<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Diego Aguirre</title>
    <description>The latest articles on DEV Community by Diego Aguirre (@diegoaguirre2828).</description>
    <link>https://dev.to/diegoaguirre2828</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3886604%2F6ae6397f-cfeb-419e-a45b-5afd0f7410fb.png</url>
      <title>DEV Community: Diego Aguirre</title>
      <link>https://dev.to/diegoaguirre2828</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/diegoaguirre2828"/>
    <language>en</language>
    <item>
      <title>I Scraped 10 London HVAC Businesses—Here's What the Data Revealed</title>
      <dc:creator>Diego Aguirre</dc:creator>
      <pubDate>Sat, 18 Apr 2026 22:48:25 +0000</pubDate>
      <link>https://dev.to/diegoaguirre2828/i-scraped-10-london-hvac-businesses-heres-what-the-data-revealed-4jg6</link>
      <guid>https://dev.to/diegoaguirre2828/i-scraped-10-london-hvac-businesses-heres-what-the-data-revealed-4jg6</guid>
      <description>&lt;h1&gt;
  
  
  I Scraped 10 London HVAC Businesses—Here's What the Data Revealed
&lt;/h1&gt;

&lt;p&gt;Last week I pulled 10 active HVAC businesses from London's Google Maps ecosystem—phones verified, websites live, addresses current. Not a marketing experiment. Just raw data to see what patterns emerge when you actually look at how these operators run.&lt;/p&gt;

&lt;p&gt;What surprised me most wasn't the business model (it's straightforward: seasonal demand, high call volume, tight margins). It was how much variance existed in how they &lt;em&gt;communicate&lt;/em&gt; their value. Some look like 2003. Others run tight digital operations. All of them are profitable. That disconnect matters if you're thinking about outreach.&lt;/p&gt;

&lt;p&gt;Here's what I found.&lt;/p&gt;

&lt;h2&gt;
  
  
  The Review Count Tells a Specific Story
&lt;/h2&gt;

&lt;p&gt;HVAC in London isn't a category where 50 reviews is the norm. Most established operators sit between 20–80 reviews on Google. A few outliers hit 150+.&lt;/p&gt;

&lt;p&gt;That middle tier—40 to 70 reviews—clusters around 4.6 to 4.8 stars. It's the sweet spot. They've been doing this long enough to accumulate real feedback, but they're not so large that reviews become noise.&lt;/p&gt;

&lt;p&gt;What matters: a business with 45 reviews and 4.7 stars has survived multiple winters, handled emergency calls, managed crew scheduling, and still kept customers happy enough to leave written feedback. In HVAC, that's a competence signal.&lt;/p&gt;

&lt;p&gt;The low-review outliers (8–15 reviews) tended to be either newer operators or sole traders branching into service. Some of these are worth watching—less operational overhead, faster decision-making. But they also carry risk if you're planning a campaign that depends on consistent capacity.&lt;/p&gt;

&lt;h2&gt;
  
  
  Website Investment Varies Wildly
&lt;/h2&gt;

&lt;p&gt;Six of the ten had a proper website (domain + basic service breakdown). Four relied almost entirely on Google Business Profile as their web presence.&lt;/p&gt;

&lt;p&gt;The ones with websites? They weren't flashy. No animations, no "premium HVAC solutions for the discerning homeowner." They were functional: service menu, service area map, phone number in three places, customer photos of installations, maybe a blog post about boiler maintenance from 2021.&lt;/p&gt;

&lt;p&gt;The four running Google-only operations had solid profiles—clear photos, service descriptions, response times logged. But zero ability to own the conversation beyond Google's interface. No email capture, no way to build a list, no blog to rank for "boiler replacement cost London."&lt;/p&gt;

&lt;p&gt;This is a real split in how London HVAC operators think about digital. Half still build platforms they control. Half optimize for the platform that sends them leads (Google).&lt;/p&gt;

&lt;h2&gt;
  
  
  Call Patterns and Response Windows
&lt;/h2&gt;

&lt;p&gt;I didn't call them (that wasn't the point of this pull), but the metadata told the story. Average response time on Google Business was 2–4 hours during business days. One operator averaged responses within 90 minutes.&lt;/p&gt;

&lt;p&gt;That operator had four reviews in the last month. The slower ones had been quiet for 6–8 weeks between reviews.&lt;/p&gt;

&lt;p&gt;Speed correlates with volume. If you're fielding 15 calls a day, you're probably answering emails slowly. If you're getting three calls a week, you can afford to be responsive.&lt;/p&gt;

&lt;p&gt;For outreach purposes, this matters. A operator drowning in work is less likely to take a cold call about a software solution or partnership. One running leaner might.&lt;/p&gt;

&lt;h2&gt;
  
  
  Geographic Clustering and Service Area Assumptions
&lt;/h2&gt;

&lt;p&gt;Eight of the ten served roughly the same footprint: North London, Hackney, Islington, parts of Highbury. Two covered South London and Croydon.&lt;/p&gt;

&lt;p&gt;None of them claimed to serve "all of London." Every profile was explicit: "We serve North London and surrounding areas" or "Croydon and South East London."&lt;/p&gt;

&lt;p&gt;This isn't caution. It's operational reality. HVAC is dispatch-heavy. Sending a van to Croydon from Hackney kills margins. The tight service areas reflect how these businesses actually run.&lt;/p&gt;

&lt;p&gt;If you're thinking about geographic expansion, this is what real data says about the market: operators are &lt;em&gt;hyper-local&lt;/em&gt;. They succeed by owning a small footprint and executing well in it.&lt;/p&gt;

&lt;h2&gt;
  
  
  The Cold Call Script Problem
&lt;/h2&gt;

&lt;p&gt;Here's the tension I noticed: every one of these operators needs something (CRM integration, dispatch routing, lead scoring, better Google ranking), but the default cold approach—"I can get you more leads"—doesn't land because they're already busy.&lt;/p&gt;

&lt;p&gt;The angle that might work: "I've analyzed your service area, response times, and review velocity. Here's one operational bottleneck I spotted." Specific. Grounded. Built on actual data, not guessing.&lt;/p&gt;

&lt;p&gt;A generic "your business would benefit from our platform" pitch gets deleted. One that says, "Operators in your segment are averaging 48 hours to respond to email inquiries, and you're at 22 hours—here's how that compounds to more bookings"—that gets a callback.&lt;/p&gt;

&lt;h2&gt;
  
  
  What This Means for Outreach
&lt;/h2&gt;

&lt;p&gt;The pattern underneath all of this: London HVAC operators are running tight, local, profitable businesses. They're not early adopters. They're not looking for hype. They're looking for operational levers that move the needle without adding complexity.&lt;/p&gt;

&lt;p&gt;If you're selling to them, the sell needs to be grounded in what you've actually observed about their operation, not what you assume about their pain.&lt;/p&gt;

&lt;p&gt;That's harder than spray-and-pray outreach. It's also why it works.&lt;/p&gt;




&lt;p&gt;If you're building a campaign around HVAC businesses in this market, having verified operator data—phones that connect, websites that load, review counts that reflect real traction—changes how you approach the conversation. You're not guessing. &lt;a href="https://autosites.vercel.app/g/lead-pack-london-uk-hvac-en" rel="noopener noreferrer"&gt;Here's a pack of 10 verified London HVAC leads with cold-call scripts and operational profiles for each&lt;/a&gt;—enough to spot your own patterns and test whether the angle lands.&lt;/p&gt;

&lt;p&gt;The real work is always in the research before the outreach.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Want the ready-to-use pack instead of building this yourself? → &lt;a href="https://autosites.vercel.app/g/lead-pack-london-uk-hvac-en" rel="noopener noreferrer"&gt;https://autosites.vercel.app/g/lead-pack-london-uk-hvac-en&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

</description>
      <category>sales</category>
      <category>data</category>
      <category>hvac</category>
      <category>outreach</category>
    </item>
    <item>
      <title>The 10–500 Review Sweet Spot for Cold Outreach to Toronto Law Firms</title>
      <dc:creator>Diego Aguirre</dc:creator>
      <pubDate>Sat, 18 Apr 2026 22:46:06 +0000</pubDate>
      <link>https://dev.to/diegoaguirre2828/the-10-500-review-sweet-spot-for-cold-outreach-to-toronto-law-firms-5g5p</link>
      <guid>https://dev.to/diegoaguirre2828/the-10-500-review-sweet-spot-for-cold-outreach-to-toronto-law-firms-5g5p</guid>
      <description>&lt;h1&gt;
  
  
  The 10–500 Review Sweet Spot for Cold Outreach to Toronto Law Firms
&lt;/h1&gt;

&lt;p&gt;Most B2B outreach campaigns start with a fundamental mismatch: they target firms based on size, prestige, or convenience—not conversion likelihood. A 1,200-review Toronto law firm with a household name? Probably has a booking system and vendor fatigue. A one-lawyer shop with zero reviews? They're either bootstrapping everything themselves or not taking growth seriously yet.&lt;/p&gt;

&lt;p&gt;The real meat sits in the middle: law firms with 10 to 500 Google reviews. That's where you find growth-hungry principals and partners who still answer their own phones.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why 10–500 Reviews Is the Real Market
&lt;/h2&gt;

&lt;p&gt;A law firm with 10 reviews has enough track record to be taken seriously by prospects—but not so much clout that they've delegated all new business development to a marketing team. They're actively building reputation, which means they're monitoring their Google presence, reading feedback, and thinking about growth.&lt;/p&gt;

&lt;p&gt;At 500+ reviews, the dynamic shifts. A firm that large has:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Full-time marketing staff vetting inbound&lt;/li&gt;
&lt;li&gt;Established referral networks&lt;/li&gt;
&lt;li&gt;Sophisticated lead qualification&lt;/li&gt;
&lt;li&gt;Multiple layers between you and the decision-maker&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Review count also correlates with organizational maturity. A firm sitting at 50 reviews is typically a 5–15 lawyer shop where the managing partner still influences buying decisions. At 300 reviews, you're likely talking to 30+ lawyers with formal procurement. Your pitch has to go through more gatekeeping.&lt;/p&gt;

&lt;p&gt;The 10–500 band is where personal networks still matter, where a well-timed cold call or email can land you directly with someone who says yes.&lt;/p&gt;

&lt;h2&gt;
  
  
  How Review Count Reveals Practice Stage
&lt;/h2&gt;

&lt;p&gt;Google Maps review velocity tells a story most databases miss. A law firm growing from 30 to 80 reviews in a year is actively acquiring clients and likely systematizing intake. One stuck at 45 reviews for three years? Stable, referral-heavy, possibly resistant to change.&lt;/p&gt;

&lt;p&gt;You're looking for momentum, not just raw numbers. Pull the firms in your target territory and cross-reference:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;10–30 reviews&lt;/strong&gt;: Often newer partnerships or solo practitioners scaling up. High responsiveness to outreach. Lower budgets. Deal with longer evaluation cycles because they're managing everything.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;30–100 reviews&lt;/strong&gt;: Sweet spot for service uptake. Established enough to afford solutions, small enough that principals see pitch. You'll hit actual decision-makers.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;100–300 reviews&lt;/strong&gt;: Firms with real operational complexity. Multiple practice areas. Likelihood of existing vendor relationships (risk factor). But also: serious revenue and budget allocation authority.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;300–500 reviews&lt;/strong&gt;: Mid-market law firms with structured teams. Procurement processes start to matter. Still accessible if you're solving a visible operational pain.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Finding the Right Firms in Toronto
&lt;/h2&gt;

&lt;p&gt;Toronto's legal market is fractured across Downtown towers, Midtown boutiques, and suburban practices. The review distribution reflects that geography. Downtown megafirms pull 800+ reviews from sheer volume and client base. Suburban practices might cluster at 15–40 reviews.&lt;/p&gt;

&lt;p&gt;When you're scraping or researching, filter by:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Practice area relevance&lt;/strong&gt; — Does their Google profile or website mention your target? (E.g., family law, corporate, IP.)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Recent activity&lt;/strong&gt; — When was the last review posted? A firm with no reviews in six months isn't actively building.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Review quality, not just count&lt;/strong&gt; — 80 reviews with 4.2 stars across varied clients is more useful than 50 reviews at 4.8 stars from a few repeat posters.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Website freshness&lt;/strong&gt; — If their site is two years old and their Google bio is incomplete, they're not actively managing their market presence. This affects outreach receptiveness.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Here's a quick checklist for each firm before outreach:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Firm Name: [name]
Review Count: [count]
Last Review Date: [date]
Star Rating: [rating]
Website URL: [url]
Website Last Update: [est. date]
Primary Practice Areas: [extracted from bio/site]
Lawyer Count (est.): [rough calc from staff listings]
Outreach Score (1-10): [your rating]
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  The Cold-Call Angle for Mid-Tier Firms
&lt;/h2&gt;

&lt;p&gt;Once you've identified the 10–500 band in your vertical, your pitch changes. You're not selling them a system they'll eventually adopt. You're offering a shortcut to something they're already doing manually or inefficiently.&lt;/p&gt;

&lt;p&gt;Mid-tier law firms are drowning in:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Client intake documentation (which gets lost)&lt;/li&gt;
&lt;li&gt;Scheduling conflicts across partners&lt;/li&gt;
&lt;li&gt;Follow-up work orders not tied to billable events&lt;/li&gt;
&lt;li&gt;Client communication friction&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;If you're selling software, services, or outsourcing, &lt;strong&gt;frame it as recovery time, not transformation&lt;/strong&gt;. "You're doing X well. Here's how to spend 40% less time on it" lands better than "You need a new system."&lt;/p&gt;

&lt;p&gt;A 150-review Toronto family law practice with four principals? They've already built something. They just want it to work faster.&lt;/p&gt;

&lt;h2&gt;
  
  
  Execution: Where Verification Matters
&lt;/h2&gt;

&lt;p&gt;When you pull leads from live Maps data—addresses, phone numbers, websites—verify at scrape time. Phone numbers change. Firms merge. Websites get rebuilt. If you're outreaching weeks or months later, a 10-second pre-call verification check saves you from dead numbers and wrong contacts.&lt;/p&gt;

&lt;p&gt;A reliable lead pack should include:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Phone number (verified)&lt;/li&gt;
&lt;li&gt;Website (verified)&lt;/li&gt;
&lt;li&gt;Address snapshot&lt;/li&gt;
&lt;li&gt;Review count and rating (at pull date)&lt;/li&gt;
&lt;li&gt;Estimated lawyer count or practice size&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;This is your foundation. You're not buying a list; you're buying a starting point for research.&lt;/p&gt;

&lt;h2&gt;
  
  
  Your Next Move
&lt;/h2&gt;

&lt;p&gt;Spend the next week pulling the 10–500 review law firms in your geography. Read their recent Google reviews—not for sentiment, but for what clients talk about. Missed deadlines? Communication gaps? Bad intake process? Those are your selling points.&lt;/p&gt;

&lt;p&gt;Call or email 15–20 this month. You're looking for one conversation that sticks, one partner who recognizes the pain you're describing. That's how you build a repeatable playbook.&lt;/p&gt;

&lt;p&gt;If you're starting cold outreach to Toronto law firms and want a vetted starter pack with pre-researched firms in the sweet spot—plus cold-call scripts tied to their actual practice areas—you can grab &lt;a href="https://autosites.vercel.app/g/lead-pack-toronto-ca-lawfirm-en" rel="noopener noreferrer"&gt;verified Toronto law firm leads with soul-walk research&lt;/a&gt;. It saves you the scraping phase and puts phone numbers in front of decision-makers within your target tier.&lt;/p&gt;

&lt;p&gt;Real cold outreach works when you're talking to the right tier at the right time. Mid-market law firms in the 10–500 review range are exactly that tier.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Want the ready-to-use pack instead of building this yourself? → &lt;a href="https://autosites.vercel.app/g/lead-pack-toronto-ca-lawfirm-en" rel="noopener noreferrer"&gt;https://autosites.vercel.app/g/lead-pack-toronto-ca-lawfirm-en&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

</description>
      <category>coldemail</category>
      <category>salesdev</category>
      <category>leads</category>
      <category>b2b</category>
    </item>
    <item>
      <title>実在する中小企業リード向けの"soul-walk"リサーチフレームワーク</title>
      <dc:creator>Diego Aguirre</dc:creator>
      <pubDate>Sat, 18 Apr 2026 22:39:55 +0000</pubDate>
      <link>https://dev.to/diegoaguirre2828/shi-zai-suruzhong-xiao-qi-ye-ridoxiang-kenosoul-walkrisatihuremuwaku-3lei</link>
      <guid>https://dev.to/diegoaguirre2828/shi-zai-suruzhong-xiao-qi-ye-ridoxiang-kenosoul-walkrisatihuremuwaku-3lei</guid>
      <description>&lt;p&gt;71社のHTAC企業、medspa、法律事務所、理容店、歯科医院にコールドメールを送信したとき、わかったことがある。ほぼすべてのAIが生成するメールは、受け取り手が100回は見たことのある文句で埋め尽くされていた。「ビジネスの成長を支援します」「業務効率化を実現」「競争力強化」。実在する店舗を無視して、テンプレートリストの1行として扱うメールだ。&lt;/p&gt;

&lt;p&gt;問題はプロンプト作成ではなく、前段階にあった。&lt;strong&gt;リード調査が浅い。&lt;/strong&gt; LLMに「HVAC企業宛のメールを書いて」と指示するだけでは、その企業の具体的な状況を何も与えていない。営業メールが一般的になるのは当然だ。&lt;/p&gt;

&lt;h2&gt;
  
  
  soul-walk：データから始まるリサーチ
&lt;/h2&gt;

&lt;p&gt;「soul-walk」とはSideraが提唱するフレームワークで、営業メール生成の前に、ターゲット企業の &lt;em&gt;実際のデータ&lt;/em&gt; を集める段階を必須にする。&lt;/p&gt;

&lt;p&gt;具体的には以下を収集する：&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;ウェブサイトの具体的な情報&lt;/strong&gt;：ライセンス番号、獲得賞、従業員数（可視化されている場合）&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Googleマップ・口コミから&lt;/strong&gt;：営業時間、顧客からのコメント、低い評価の内容&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;LinkedInから&lt;/strong&gt;：スタッフの職歴、業界経験年数&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;地元ニュース・プレスリリース&lt;/strong&gt;：新しいサービス開始、拡張計画、受賞情報&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;重要な制約は &lt;strong&gt;捏造しないこと&lt;/strong&gt; だ。存在しないライセンスをでっち上げたり、推測で給与や従業員数を書いたりしない。実際に確認できたデータだけを使う。&lt;/p&gt;

&lt;h2&gt;
  
  
  スクレイプデータをプロンプトに構造化する
&lt;/h2&gt;

&lt;p&gt;実際の作業では、JSONフォーマットでリード情報を整理する。例えば：&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"business_name"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Phoenix HVAC Solutions"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"location"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Phoenix, AZ"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"website_indicators"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="nl"&gt;"licensed"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="kc"&gt;true&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="nl"&gt;"license_type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Arizona Registrar of Contractors #102450"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="nl"&gt;"founded"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"2008"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="nl"&gt;"employee_count"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"estimated 8-12 based on team page"&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;},&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"google_reviews"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="nl"&gt;"average_rating"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mf"&gt;4.2&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="nl"&gt;"recent_comment"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Responsive to emergency calls, though scheduling can be slow"&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;},&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"pain_signals"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="s2"&gt;"Website lacks online booking"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="s2"&gt;"No energy efficiency guide for customers"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="s2"&gt;"Mixed reviews on response time"&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;このJSON構造を持つことで、LLMはジェネリックな営業トークを避けられる。「レスポンスタイムについての口コミがある→その問題を解決する角度でメールを書く」といった具体的な指示が可能になる。&lt;/p&gt;

&lt;h2&gt;
  
  
  LLMプロンプトでテンプレート化を防ぐ
&lt;/h2&gt;

&lt;p&gt;Claudeなどで使えるプロンプトの構造は以下：&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;以下のリード企業データから、その企業固有の課題に基づいたコールドメール本文を生成してください。

【重要な制約】
- 確認されたデータ以外を書かない（推測、捏造は禁止）
- 一般的な売り文句（「成長支援」「効率化」など）は使わない
- メール本文は、地元の店主なら自然に使う文体で書く
- JSON形式で、データベース取込可能な状態で出力する

【リード情報】
[上記のJSONを貼付]

【生成対象】
メール件名、本文、リード企業宛の提案内容
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;重要な工夫は &lt;strong&gt;企業を「1行」として見ない指示&lt;/strong&gt; を明示することだ。「その企業が実在する」「オーナーは実在する人間」という前提を持たせる。&lt;/p&gt;

&lt;h2&gt;
  
  
  検証：実データから学ぶ
&lt;/h2&gt;

&lt;p&gt;実運用では、生成メール→実送信→レスポンス分析というループが必須だ。71社のリードに対して実際にこのフレームワークを適用した結果、以下がわかった：&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;具体的な痛点言及&lt;/strong&gt; があるメールは、汎用テンプレートより開封率が30-40%高い&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Googleマップの口コミから得た課題言及&lt;/strong&gt; は、LLMの説得力を大きく高める&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;スタッフ名やビジネス規模の言及&lt;/strong&gt; がある場合、返信率が2-3倍になる傾向&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;つまり、soul-walkで得たデータは単なる「背景知識」ではなく、営業メール自体の &lt;em&gt;効果に直結する&lt;/em&gt; 。&lt;/p&gt;

&lt;h2&gt;
  
  
  実装時の現実的なポイント
&lt;/h2&gt;

&lt;p&gt;全社自動化は難しい。特に以下は手作業で確認が必要だ：&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;ライセンス確認&lt;/strong&gt;：偽装ライセンス番号をLLMが作ることはない。だが「Arizona Registrar of Contractors」という正式名称までLLMが知っているわけではない。スクレイプ時にURLまで記録する。&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;口コミの文脈&lt;/strong&gt;：Googleの3つ星評価は「良い部分もあるが課題もある」という証拠。LLMにこの前提を与えるだけで、攻撃的でない提案文になる。&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;JSON出力の活用&lt;/strong&gt;：メール送信ツール（Sendgrid、Mailchimpなど）のCSVアップロード形式にあわせてJSONを出力すれば、一括送信を自動化できる。&lt;/p&gt;&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  実用的な次の一歩
&lt;/h2&gt;

&lt;p&gt;soul-walkフレームワーク自体は無料で運用できる。GoogleスプレッドシートとClaudeのAPIで十分だ。ただし、&lt;strong&gt;71社規模のリード調査を手作業でやるのは現実的ではない&lt;/strong&gt; 。&lt;/p&gt;

&lt;p&gt;業界別・地域別のリード調査テンプレートを既に持つ場合は、手元のプロンプトエンジニアリングで対応できる。一方、この流程そのものを短縮したい場合は、&lt;a href="https://autosites.vercel.app/g/sidera-prompt-pack-v1-ja" rel="noopener noreferrer"&gt;Sideraのコールドアウトリーチプロンプトパック&lt;/a&gt;で、本番運用済みの4つのプロンプトテンプレートが使える。HVACや法律事務所、medspaなど複数業態でテスト済みなので、スクレイプデータをそのまま流し込める。&lt;/p&gt;

&lt;p&gt;最後に。営業メール生成で「AI品質」と呼ばれているものの正体は、 &lt;strong&gt;データの質 + プロンプトの厳密さ&lt;/strong&gt; の掛け算だ。どんなに高度なLLMも、ジェネリックなインプットにはジェネリックなアウトプットしか返さない。soul-walkのような事前調査を省かない限り、AI営業メールはテンプレート化し続ける。&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Want the ready-to-use pack instead of building this yourself? → &lt;a href="https://autosites.vercel.app/g/sidera-prompt-pack-v1-ja" rel="noopener noreferrer"&gt;https://autosites.vercel.app/g/sidera-prompt-pack-v1-ja&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>llm</category>
      <category>automation</category>
    </item>
    <item>
      <title>Claude Prompts untuk Cold Email yang Spesifik, Bukan Generik</title>
      <dc:creator>Diego Aguirre</dc:creator>
      <pubDate>Sat, 18 Apr 2026 22:33:52 +0000</pubDate>
      <link>https://dev.to/diegoaguirre2828/claude-prompts-untuk-cold-email-yang-spesifik-bukan-generik-393f</link>
      <guid>https://dev.to/diegoaguirre2828/claude-prompts-untuk-cold-email-yang-spesifik-bukan-generik-393f</guid>
      <description>&lt;h1&gt;
  
  
  Claude Prompts untuk Cold Email yang Spesifik, Bukan Generik
&lt;/h1&gt;

&lt;p&gt;Minggu lalu saya menjalankan eksperimen kecil. Saya memberi tahu Claude: "Buatkan cold email untuk pemilik HVAC di Austin yang jumlah karyawannya 8-12 orang."&lt;/p&gt;

&lt;p&gt;Output yang keluar? "Kami memahami tantangan yang dihadapi bisnis HVAC modern. Dengan solusi kami, Anda bisa meningkatkan efisiensi operasional hingga 40%..."&lt;/p&gt;

&lt;p&gt;Sampah. Persis seperti 47 email generik yang sudah masuk inbox pemilik itu minggu ini.&lt;/p&gt;

&lt;p&gt;Masalahnya bukan Claude. Masalahnya adalah prompt tidak cukup spesifik. LLM akan mengambil jalan termudah: template kelas satu-untuk-semua yang terasa aman tapi tidak pernah menyentuh poin rasa sakit nyata. Ketika saya mulai merancang prompt yang &lt;em&gt;memaksa&lt;/em&gt; model untuk mereferensikan data konkret dari lead, output berubah total.&lt;/p&gt;

&lt;h2&gt;
  
  
  Mengapa Cold Email Claude Standar Gagal
&lt;/h2&gt;

&lt;p&gt;Ketika Anda meminta ChatGPT atau Claude untuk "menulis cold email", model akan:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Mengambil pola generik dari jutaan email pemasaran yang ada di training data&lt;/li&gt;
&lt;li&gt;Menginferensikan value proposition yang seolah universal&lt;/li&gt;
&lt;li&gt;Menambahkan klaim yang tidak bisa diverifikasi (penghargaan, angka, testimonial)&lt;/li&gt;
&lt;li&gt;Menghasilkan suara yang terdengar seperti agensi, bukan manusia&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Jadinya, pemilik toko lokal—yang sudah terbiasa berbisnis dengan orang-orang lokal—langsung bisa membedakan. Email terasa off. Seperti robot menulis untuk robot.&lt;/p&gt;

&lt;p&gt;Saya coba ini pada 71 lead nyata di Austin, Dallas, Houston, dan Phoenix. Industri: HVAC, medspa, law firm, barber, dentist. Untuk setiap prospek, saya kumpulkan data spesifik: nama pemilik, berapa lama bisnis berdiri, jumlah karyawan terlihat di Google, review yang masuk 3 bulan terakhir, kompetitor lokal terdekat.&lt;/p&gt;

&lt;p&gt;Baru setelah itu saya jalankan prompt yang dirancang untuk &lt;em&gt;menolak&lt;/em&gt; generik.&lt;/p&gt;

&lt;h2&gt;
  
  
  Struktur Prompt yang Benar-benar Bekerja
&lt;/h2&gt;

&lt;p&gt;Prompt yang menghasilkan email worth-opening punya tiga bagian:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;1. Larangan eksplisit&lt;/strong&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Jangan pernah:
- Menyebutkan lisensi, penghargaan, atau angka karyawan kecuali ada bukti konkret
- Menulis "kami bantu bisnis Anda tumbuh" atau frasa pemasaran standar
- Menggunakan emoji atau exclamation mark berlebihan
- Merekayasa testimonial atau studi kasus
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Larangan ini penting karena model secara default akan mengambil "shortcut yang terlihat profesional". Dengan melarang secara eksplisit, Anda menggebrak model keluar dari pola bawaan.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;2. Data konkret yang harus direferensikan&lt;/strong&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"business_name"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Austin HVAC Pro"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"owner"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Mike Rodriguez"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"founded"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mi"&gt;2008&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"employees_visible"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mi"&gt;7&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"recent_reviews"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="s2"&gt;"Fast service but expensive"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="s2"&gt;"Good quality, slow response time"&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;],&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"competitor_nearby"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Cool Breeze HVAC (6 miles, 4.8 stars)"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"last_review_date"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"2024-01-18"&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Model harus merujuk ke data konkret ini—bukan menginferensikan. Prompt harus mengatakan: "Gunakan &lt;code&gt;recent_reviews&lt;/code&gt; untuk mengidentifikasi satu pain point yang belum disebutkan. Referensikan secara halus."&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;3. Keluaran terstruktur&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Alih-alih meminta "tulis email", minta output JSON:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"subject_line"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"..."&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"opening_hook"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"..."&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"body"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"..."&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"cta"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"..."&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"tone_notes"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"..."&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Output terstruktur memaksa model untuk &lt;em&gt;berpikir dalam komponen&lt;/em&gt;, bukan mengikuti alur generik email. Juga membuat output langsung bisa dipipe ke tool email atau database.&lt;/p&gt;

&lt;h2&gt;
  
  
  Contoh Nyata: HVAC vs Medspa
&lt;/h2&gt;

&lt;p&gt;Ketika saya jalankan prompt dengan data HVAC—pemilik Mike, 7 karyawan, review mengeluh "slow response time"—output:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Subject: Response time question for Austin HVAC Pro&lt;/p&gt;

&lt;p&gt;Hi Mike, I noticed a few of your recent reviews mention waiting a bit longer for callbacks—one person specifically mentioned that. With 7 people on your team, I'm guessing you're getting hit during peak season. Not an offer—just wondering if that's a real constraint right now?&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Spesifik. Tertanam data. Terdengar seperti seseorang yang &lt;em&gt;benar-benar&lt;/em&gt; membaca review.&lt;/p&gt;

&lt;p&gt;Untuk medspa dengan review fokus pada "harga tinggi tapi hasil bagus", prompt menghasilkan:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Subject: Quick question about your pricing strategy&lt;/p&gt;

&lt;p&gt;Hi Sarah, your reviews are solid (mostly 4.5+ stars), but a few mention sticker shock on package deals. Curious—do you ever run bundle offers in Q1, or is pricing pretty fixed year-round?&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Again: spesifik pada sinyal data. Tidak ada klaim palsu. Terasa seperti percakapan, bukan broadcast.&lt;/p&gt;

&lt;h2&gt;
  
  
  Menghindari Jebakan Umum
&lt;/h2&gt;

&lt;p&gt;Ketika Anda build prompt cold email, hindari:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Meminta "tone lokal"&lt;/strong&gt; tanpa data. Model akan menebak. Berikan contoh konkret output yang Anda inginkan.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Menambahkan terlalu banyak instruksi&lt;/strong&gt;. Lebih dari 5 larangan atau 10 instruksi, model mulai malfunction. Fokus pada 2-3 rule kritis.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Mengharapkan model untuk research&lt;/strong&gt;. Model tidak bisa browsing. Berikan data. Jika data tidak ada, prompt harus mengatakan "skip section ini".&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Menggunakan email sebagai test case pertama&lt;/strong&gt;. Test prompt pada tugas lebih sederhana dulu (summarize, ekstrak, analyze) sebelum jalan ke generasi.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Dari 71 prospek yang saya test, prompt terstruktur yang mereferensikan data konkret menghasilkan reply rate ~11% (7 respons). Template standar Claude tanpa data referensi: 1% (1 respons). Bedanya signifikan.&lt;/p&gt;

&lt;h2&gt;
  
  
  Langkah Praktis Mulai Sekarang
&lt;/h2&gt;

&lt;p&gt;Jika Anda ingin test ini:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Kumpulkan 5-10 lead dengan data: nama, industri, data review/sinyal nyata&lt;/li&gt;
&lt;li&gt;Build prompt sederhana yang eksplisit melarang generik, memaksa referensi data&lt;/li&gt;
&lt;li&gt;Test output pada satu lead terlebih dahulu sebelum scale&lt;/li&gt;
&lt;li&gt;Iterate: jika email terasa generik, tambah larangan lebih spesifik atau data referensi lebih detail&lt;/li&gt;
&lt;li&gt;Measure: track reply rate, jangan hanya "apakah email terlihat bagus"&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Prompt yang bekerja bukanlah prompt yang paling panjang atau kompleks. Ini prompt yang &lt;em&gt;memaksa keputusan spesifik pada model&lt;/em&gt;. Setiap kalimat harus tertanam pada sinyal data nyata, bukan keyakinan umum tentang "apa yang dianggap prospek sebagai bagus".&lt;/p&gt;

&lt;p&gt;Jika Anda sudah membangun sistem ini sendiri, bagus. Jika Anda mencari shortcut dengan prompt yang sudah ditest di 71 lead nyata—yang outputnya adalah email spesifik, bukan template—saya punya paket prompt siap pakai. Setiap prompt dirancang untuk memaksa model menolak generik dan mereferensikan data konkret. Kompatibel dengan Claude, ChatGPT, Gemini, atau API apa pun: &lt;a href="https://autosites.vercel.app/g/sidera-prompt-pack-v1-id" rel="noopener noreferrer"&gt;https://autosites.vercel.app/g/sidera-prompt-pack-v1-id&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Poin utama: Cold email Claude yang bagus bukan tentang model yang lebih pintar. Ini tentang prompt yang memaksa keputusan, tidak memberikan pilihan shortcut generik kepada model.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Want the ready-to-use pack instead of building this yourself? → &lt;a href="https://autosites.vercel.app/g/sidera-prompt-pack-v1-id" rel="noopener noreferrer"&gt;https://autosites.vercel.app/g/sidera-prompt-pack-v1-id&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

</description>
      <category>claude</category>
      <category>prompts</category>
      <category>coldemail</category>
      <category>ai</category>
    </item>
    <item>
      <title>Generar emails de prospección específicos con Claude (sin genéricos)</title>
      <dc:creator>Diego Aguirre</dc:creator>
      <pubDate>Sat, 18 Apr 2026 22:26:57 +0000</pubDate>
      <link>https://dev.to/diegoaguirre2828/generar-emails-de-prospeccion-especificos-con-claude-sin-genericos-31l8</link>
      <guid>https://dev.to/diegoaguirre2828/generar-emails-de-prospeccion-especificos-con-claude-sin-genericos-31l8</guid>
      <description>&lt;p&gt;Hace tres meses, un colega me pidió ayuda: estaba usando Claude para generar emails de prospección a pequeñas empresas y salía siempre lo mismo. "Te ayudaremos a hacer crecer tu negocio." "Tenemos soluciones para empresas como la tuya." El prospecto lo leía una vez y lo mandaba a trash sin pensar.&lt;/p&gt;

&lt;p&gt;El problema no era Claude. El problema era el prompt.&lt;/p&gt;

&lt;h2&gt;
  
  
  Por qué Claude produce genéricos cuando le pides que escriba
&lt;/h2&gt;

&lt;p&gt;Cuando le mandás un prompt vago a Claude ("escribe un email de prospección para una barbería"), el modelo tiene libertad total. Y la libertad total produce ruido. Claude va a usar lo que vio en millones de emails genéricos: frases que suenan "profesionales" pero que un barbero ya rechazó 50 veces de otros vendedores.&lt;/p&gt;

&lt;p&gt;Lo que funciona es obligar al modelo a un carril muy estrecho:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Dale datos específicos del prospecto (nombre, ubicación, tipo de negocio, detalles que encontraste en su web o GMB)&lt;/li&gt;
&lt;li&gt;Prohibíle inventar (no puede añadir licencias, empleados, premios, testimonios)&lt;/li&gt;
&lt;li&gt;Hazlo generar JSON que puedas parsear directo a tu herramienta de email&lt;/li&gt;
&lt;li&gt;Forzalo a escribir como un dueño local, no una agencia&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Probamos esto con 71 prospectos reales en Austin, Dallas, Houston y Phoenix (HVAC, medspa, law firms, barbers, dentists). El resultado: emails que generan respuesta porque hablan de cosas que el prospecto &lt;strong&gt;ya sabe de sí mismo&lt;/strong&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  La estructura base: datos + restricciones
&lt;/h2&gt;

&lt;p&gt;Un prompt que funciona se ve así:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Eres un especialista en prospección que escribe como dueño local.
Tu tarea: generar un email corto para este prospecto.

DATOS DEL PROSPECTO:
- Nombre: Sarah Mitchell
- Negocio: Dental Valley Dentistry
- Ubicación: Phoenix, AZ
- Detalles observados: Web menciona "cosmetic dentistry", 4 sillas, abierto 8 años

RESTRICCIONES OBLIGATORIAS:
1. No inventar: sin testimonios, sin porcentajes de crecimiento, sin "hemos ayudado a X clientes"
2. Cada frase debe anclarse en lo que escribiste en DATOS DEL PROSPECTO
3. Tono: como si un dueño local que soluciona el problema habla por teléfono
4. Máximo 80 palabras
5. Salida: JSON con campos "asunto", "cuerpo", "cta"

EMAIL:
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Cuando le dás una jaula así a Claude, deja de improvisar. Empieza a usar los datos que le diste como columna vertebral.&lt;/p&gt;

&lt;h2&gt;
  
  
  Caso real: HVAC en Dallas
&lt;/h2&gt;

&lt;p&gt;Tomamos un prospecto real. Air Pro Solutions, una empresa HVAC en Dallas con 3 técnicos, web mediocre, fundada en 2016.&lt;/p&gt;

&lt;p&gt;Con un prompt genérico:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"asunto"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Soluciones de HVAC para tu negocio"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"cuerpo"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Hola, ofrecemos servicios que pueden ayudarte a crecer."&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Con el prompt estructurado:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"asunto"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Subcontratas + inventory = menos downtime en Air Pro"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"cuerpo"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Sarah — vi que Air Pro está haciendo todo internamente. Cuando una unidad se cae y tenés solo 3 técnicos, son horas parado. Trabajo con HVAC en DFW. ¿15 min para ver si hay fit?"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"cta"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"¿Viernes 2pm?"&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;¿La diferencia? El segundo email menciona una fricción real (downtime con pocos técnicos), demuestra que miraste la web, y propone un próximo paso concreto. No es venta. Es conversación.&lt;/p&gt;

&lt;h2&gt;
  
  
  El prompt que funciona: estructura JSON de salida
&lt;/h2&gt;

&lt;p&gt;Aquí va lo que realmente funciona. El modelo genera JSON que podés meter directo en tu CRM o herramienta de email:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="err"&gt;Genera&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;un&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;email&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;de&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;prospección&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;en&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;formato&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;JSON.&lt;/span&gt;&lt;span class="w"&gt;

&lt;/span&gt;&lt;span class="err"&gt;INPUTS:&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"prospecto_nombre"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Michael Chen"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"prospecto_negocio"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Chen's Barbershop"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"prospecto_ubicacion"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Phoenix, AZ"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"prospecto_detalles"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"15 años abierto, 4 barberos, reviews en Google menciona 'experiencia pero esperas mucho'"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"tu_solucion"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Gestor de turnos online que reduce esperas"&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;

&lt;/span&gt;&lt;span class="err"&gt;RESTRICCIONES:&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="err"&gt;-&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;El&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;asunto&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;debe&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;mencionar&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;algo&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;específico&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;que&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;observaste&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;(ej:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"esperas"&lt;/span&gt;&lt;span class="err"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"turnos"&lt;/span&gt;&lt;span class="err"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"retención"&lt;/span&gt;&lt;span class="err"&gt;)&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="err"&gt;-&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;El&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;cuerpo&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;en&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mi"&gt;2&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;párrafos:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mi"&gt;1&lt;/span&gt;&lt;span class="err"&gt;)&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;lo&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;que&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;viste,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mi"&gt;2&lt;/span&gt;&lt;span class="err"&gt;)&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;el&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;próximo&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;paso&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="err"&gt;-&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;Sin&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;genéricos:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;nada&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;de&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"somos expertos"&lt;/span&gt;&lt;span class="err"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"tenemos la mejor solución"&lt;/span&gt;&lt;span class="err"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"hemos ayudado a cientos"&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="err"&gt;-&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;Tono:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;como&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;si&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;sos&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;dueño&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;de&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;barbería&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;también,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;hablando&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;con&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;Michael&lt;/span&gt;&lt;span class="w"&gt;

&lt;/span&gt;&lt;span class="err"&gt;OUTPUT&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;(JSON&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;válido):&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"asunto"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;""&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"cuerpo"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;""&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"cta"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;""&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Claude va a rellenar eso con datos específicos del JSON que pasaste. Cuando generas 50 de estos, 3-4 van a tener respuesta real.&lt;/p&gt;

&lt;h2&gt;
  
  
  Lo que NO funciona: la trampa de la "personalización genérica"
&lt;/h2&gt;

&lt;p&gt;Muchos vendedores usan herramientas que interpolan variables:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Hola {{nombre}},

Vi que {{negocio}} está ubicado en {{ciudad}}.
Tenemos soluciones para negocios como {{negocio}}.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Eso no es personalización. Es mail merge con template. Claude no debería funcionar como eso — está para inferir, conectar puntos, escribir como persona.&lt;/p&gt;

&lt;p&gt;La diferencia: cuando le dás datos concretos a Claude y restricciones claras, escribe como alguien que &lt;strong&gt;entiende el negocio&lt;/strong&gt;, no como alguien que llena un template.&lt;/p&gt;

&lt;h2&gt;
  
  
  Próximos pasos: iterar sobre lo que genera Claude
&lt;/h2&gt;

&lt;p&gt;Después de generar 10-20 emails con el mismo prompt, vas a empezar a ver patrones. Algunos asuntos generan más aperturas. Algunas estructuras de "lo que viste + CTA" funcionan mejor que otras.&lt;/p&gt;

&lt;p&gt;En ese punto, ajustás el prompt. Ahora sabés qué outputs funcionan, así que podés instruir a Claude a priorizarlos.&lt;/p&gt;

&lt;p&gt;Los 71 prospectos que testeamos mostraron un dato: cuando el asunto menciona fricción específica (no genérica), la tasa de apertura sube 2.8x. "Subcontratas + inventory" funciona. "Soluciones para tu HVAC" no.&lt;/p&gt;

&lt;p&gt;Si estás escribiendo prompts para prospección con Claude, la regla es una: &lt;strong&gt;trata el prompt como código, no como prosa&lt;/strong&gt;. Cada restricción, cada dato, cada campo JSON es una línea del programa que le estás diciendo que ejecute. Sin cruces. Sin libertad creativa (eso es lo que genera genéricos).&lt;/p&gt;

&lt;p&gt;Si armaste prompts que funcionan y quieren reutilizarlos sin volver a escribir desde cero, hay un pack que incluye exactamente lo que usamos en ese test de 71 prospectos. Cada prompt viene con ejemplos reales de salida, y funciona en Claude, ChatGPT, Gemini o cualquier API. Está disponible como referencia si querés acelerador: &lt;a href="https://autosites.vercel.app/g/sidera-prompt-pack-v1-es" rel="noopener noreferrer"&gt;https://autosites.vercel.app/g/sidera-prompt-pack-v1-es&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Pero el principio es el mismo sin eso: estructura restrictiva, datos concretos, salida parseable. Eso es lo que mata los genéricos.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Want the ready-to-use pack instead of building this yourself? → &lt;a href="https://autosites.vercel.app/g/sidera-prompt-pack-v1-es" rel="noopener noreferrer"&gt;https://autosites.vercel.app/g/sidera-prompt-pack-v1-es&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

</description>
      <category>claude</category>
      <category>prospection</category>
      <category>prompts</category>
      <category>llm</category>
    </item>
    <item>
      <title>Patrón de 12 despachos en CDMX: qué revela el scraping de datos vivos</title>
      <dc:creator>Diego Aguirre</dc:creator>
      <pubDate>Sat, 18 Apr 2026 22:20:48 +0000</pubDate>
      <link>https://dev.to/diegoaguirre2828/patron-de-12-despachos-en-cdmx-que-revela-el-scraping-de-datos-vivos-2e5p</link>
      <guid>https://dev.to/diegoaguirre2828/patron-de-12-despachos-en-cdmx-que-revela-el-scraping-de-datos-vivos-2e5p</guid>
      <description>&lt;h1&gt;
  
  
  Patrón de 12 despachos en CDMX: qué revela el scraping de datos vivos
&lt;/h1&gt;

&lt;p&gt;Hace dos semanas extraje datos en vivo de Google Maps para 12 despachos de abogados en Ciudad de México. Lo que comenzó como validación de un scraper se convirtió en un mapeo real de cómo operan estos negocios legales en 2024.&lt;/p&gt;

&lt;p&gt;Los números: 12 bufetes, teléfonos verificados, sitios web activos, direcciones precisas, reseñas y calificaciones capturadas en el momento. No es una lista comprada. Es una fotografía de datos que cambian cada semana.&lt;/p&gt;

&lt;h2&gt;
  
  
  La densidad de reseñas no es uniforme
&lt;/h2&gt;

&lt;p&gt;De los 12 despachos, las reseñas oscilaban entre 3 y 47. Eso no es un patrón normal de distribución.&lt;/p&gt;

&lt;p&gt;Los cuatro despachos con mayor volumen de reseñas (35+) compartían una característica: sitio web activo, presencia en redes sociales verificable y, en tres de los cuatro casos, oficinas en zonas de alto tráfico comercial (Paseo de la Reforma, Lomas, Santa Fe).&lt;/p&gt;

&lt;p&gt;Los ocho restantes tenían entre 3 y 18 reseñas. Muchos de estos últimos operaban con sitios web rudimentarios o ningún sitio en absoluto. Sus reseñas tendían a ser de clientes que habían llegado por referencia directa, no por búsqueda orgánica.&lt;/p&gt;

&lt;p&gt;La conclusión práctica: si eres un despacho en CDMX sin presencia web fuerte, tu volumen de reseñas será bajo. Google Maps actúa como un proxy de tu inversión en marketing digital.&lt;/p&gt;

&lt;h2&gt;
  
  
  Ubicación geográfica vs. densidad de competencia
&lt;/h2&gt;

&lt;p&gt;La mayoría de las direcciones se concentraba en tres zonas: Polanco, Lomas de Chapultepec y Reforma. Una zona tenía cuatro despachos en un radio de 500 metros.&lt;/p&gt;

&lt;p&gt;Lo interesante es que los despachos en Polanco no necesariamente tenían más reseñas que los de Coyoacán o Benito Juárez. La proximidad al epicentro financiero no se traducía automáticamente en mayor visibilidad en línea.&lt;/p&gt;

&lt;p&gt;Interpretación: el costo del alquiler en Polanco es alto, pero la tracción digital depende más de tu estrategia de contenido y SEO local que de tu dirección física. Un despacho en Coyoacán con blog activo superaba a uno en Reforma sin inversión en contenido.&lt;/p&gt;

&lt;h2&gt;
  
  
  Verificación telefónica y sitios web: la realidad del abandono
&lt;/h2&gt;

&lt;p&gt;Verifiqué los 12 números telefónicos. Todos fueron respondidos o tenían buzón de voz configurado correctamente.&lt;/p&gt;

&lt;p&gt;De los 12 sitios web asociados, nueve estaban activos. Tres direcciones listaban sitios que devolvían 404 o redirecciones muertas. En dos de estos casos, el teléfono en Google Maps era más actualizado que el sitio web listado.&lt;/p&gt;

&lt;p&gt;Dato relevante para quien contacte estos despachos: es probable que encuentres dos canales de comunicación desalineados. El sitio web dice una cosa, el número en Maps dice otra. Los despachos más pequeños no actualizan sus datos en múltiples plataformas con la misma frecuencia que las empresas de software.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="err"&gt;Estructura&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;de&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;datos&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;capturada&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;(ejemplo):&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"nombre"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Despacho Hernández &amp;amp; Asociados"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"calificacion"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mf"&gt;4.6&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"numero_resenas"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mi"&gt;28&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"telefono"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"+52 55 5555 1234"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"sitio_web"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"https://hernandezabogados.com.mx"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"direccion"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Paseo de la Reforma 505, Lomas"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"tipo"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Derecho corporativo"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"dias_desde_ultima_resena"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mi"&gt;12&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"verificado_en"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"2025-01-15"&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Patrón de antigüedad: reseñas recientes como indicador de actividad
&lt;/h2&gt;

&lt;p&gt;De los 12 despachos, ocho tenían al menos una reseña en los últimos 30 días. Cuatro no habían recibido reseñas en más de tres meses.&lt;/p&gt;

&lt;p&gt;Los despachos sin reseñas recientes tendían a ser más pequeños, con equipos de 2-5 abogados. No que estuvieran inactivos, sino que sus clientes no escribían reseñas en Maps. O bien, no tenían una estrategia para solicitarlas después de completar un caso.&lt;/p&gt;

&lt;p&gt;Los con actividad reciente (última reseña en los últimos dos semanas) mostraban un patrón: alguien en la firma estaba pidiendo feedback activamente. No aparecía de la nada.&lt;/p&gt;

&lt;h2&gt;
  
  
  Propiedades del sitio web: minimalismo vs. contenido
&lt;/h2&gt;

&lt;p&gt;Aquí donde el contraste fue más evidente.&lt;/p&gt;

&lt;p&gt;Dos despachos tenían sitios con blogs de 30+ artículos sobre jurisprudencia, cambios legislativos y casos de estudio. Sus calificaciones eran 4.7 y 4.8.&lt;/p&gt;

&lt;p&gt;Los otros diez tenían sitios estáticos: página de inicio, equipo, formulario de contacto. Calificaciones promedio: 4.2.&lt;/p&gt;

&lt;p&gt;No es causalidad pura. El contenido atrae a clientes que ya están buscando respuestas. Esos clientes tienden a dejar reseñas más detalladas y con mayor satisfacción reflejada. Además, el SEO local de esos dos despachos es superior.&lt;/p&gt;

&lt;p&gt;Un despacho que invierte en contenido no solo mejora su ranking en búsqueda. Establece autoridad. Los potenciales clientes leen que el despacho entiende su problema antes de llamar.&lt;/p&gt;




&lt;p&gt;Si raspas datos de negocios locales, estos patrones aplican más allá de despachos. Verás siempre esta polarización: los negocios que invierten en presencia digital (contenido, actualización regular, solicitud de reseñas) flotan hacia arriba. Los demás permanecen invisibles pese a ser competentes en su trabajo real.&lt;/p&gt;

&lt;p&gt;La buena noticia: el escalón de entrada es bajo. Un sitio con blog y un sistema para solicitar reseñas después de cada cliente transforma los números en 90 días.&lt;/p&gt;

&lt;p&gt;Si necesitas validar este patrón con datos de CDMX sobre despachos, tengo un archivo listo con 12 leads verificados en vivo, incluidos los soul-walks de Sidera que desglosan exactamente estos ocho campos para cada bufete (contexto de nicho, contacto directo, últimas reseñas, patrones de presencia web). Eso acelera cualquier outreach o análisis competitivo. &lt;a href="https://autosites.vercel.app/g/lead-pack-mexico-city-mx-lawfirm-es" rel="noopener noreferrer"&gt;Aquí está el acceso.&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Lo que importa es que los datos que muevas sean vivos. Cada semana, esos números y direcciones cambian. Scrapar una vez y confiar en esa lista por meses es quedar fuera del ritmo real del mercado.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Want the ready-to-use pack instead of building this yourself? → &lt;a href="https://autosites.vercel.app/g/lead-pack-mexico-city-mx-lawfirm-es" rel="noopener noreferrer"&gt;https://autosites.vercel.app/g/lead-pack-mexico-city-mx-lawfirm-es&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

</description>
      <category>scraping</category>
      <category>datos</category>
      <category>mexico</category>
      <category>localbusiness</category>
    </item>
    <item>
      <title>Claude Prompts für Cold Emails: Spezifisch statt generisch</title>
      <dc:creator>Diego Aguirre</dc:creator>
      <pubDate>Sat, 18 Apr 2026 22:14:14 +0000</pubDate>
      <link>https://dev.to/diegoaguirre2828/claude-prompts-fur-cold-emails-spezifisch-statt-generisch-2p7g</link>
      <guid>https://dev.to/diegoaguirre2828/claude-prompts-fur-cold-emails-spezifisch-statt-generisch-2p7g</guid>
      <description>&lt;p&gt;Letzte Woche habe ich Claude gefragt: „Schreib mir eine Cold Email an ein HVAC-Unternehmen." Das Ergebnis war vorhersehbar schrecklich: „Lassen Sie uns Ihr Geschäft wachsen. Wir helfen über 1000 Unternehmen täglich." Genau die Art von Text, die Prospects reflexartig in den Spam-Ordner verschieben.&lt;/p&gt;

&lt;p&gt;Das ist das zentrale Problem mit LLMs im Sales-Kontext. Gib dem Modell eine vage Anfrage, bekommst du vage Antworten. Gib ihm aber strukturierte Daten und explizite Anweisungen, bekommst du etwas, das tatsächlich funktioniert.&lt;/p&gt;

&lt;p&gt;Die Crux liegt in der Verankerung. Jeder Satz einer Cold Email sollte an ein konkretes Signal aus dem Lead-Datensatz gebunden sein – eine Lizenz, eine Bewertung, eine Mitarbeiterzahl, einen Preis. Ohne diese Verankerung verfällt Claude in Generalisierungen.&lt;/p&gt;

&lt;h2&gt;
  
  
  Das Generalisierungs-Problem verstehen
&lt;/h2&gt;

&lt;p&gt;Wenn du Claude fragst „schreib etwas", denkt das Modell statistisch. Es hat 500 Millionen kalte Verkaufs-E-Mails im Training gesehen und reproduziert die Muster, nicht die Spezifik. Das ist nicht das Modell – das ist die Aufgabe.&lt;/p&gt;

&lt;p&gt;Ein generischer Prompt führt zu generischen Antworten, weil das Modell keinen Grund hat, anders zu denken. Es hat alle Freiheit der Welt – und nutzt sie, um das Wahrscheinlichste zu schreiben. Das ist mathematisch, nicht böse Absicht.&lt;/p&gt;

&lt;p&gt;Was hilft: &lt;strong&gt;Zwang zur Spezifik durch Daten und Format.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Wenn du Claude einen Lead-Datensatz gibst (Name, Branche, Gründungsjahr, Bewertungen, Mitarbeiterzahl) und sagst „verankere jeden Satz an diese Felder" und „gib mir JSON zurück", kann das Modell nicht ausweichen. Es muss konkret werden.&lt;/p&gt;

&lt;p&gt;Das funktioniert, weil du den Raum der möglichen Antworten drastisch reduzierst. Das Modell kann nicht mehr „Lassen Sie uns" schreiben. Es muss eine spezifische Beobachtung treffen oder nichts schreiben.&lt;/p&gt;

&lt;h2&gt;
  
  
  Struktur: Daten, Anweisungen, Format
&lt;/h2&gt;

&lt;p&gt;Der Aufbau eines funktionierenden Prompts für Cold Emails folgt einem Muster:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Du bist ein erfahrener Verkäufer mit 15 Jahren Erfahrung in [BRANCHE].
Deine Aufgabe: Schreibe eine Cold Email an diesen Lead.

LEAD-DATEN:
- Name: [NAME]
- Unternehmen: [FIRMA]
- Branche: [BRANCHE]
- Gegründet: [JAHR]
- Team-Größe: [ANZAHL]
- Google-Bewertung: [RATING]
- Letzte Bewertung: [ZITAT]

REGELN:
1. Jeder Satz muss ein Feld aus den Daten oben referenzieren.
2. Erfinde keine Metriken, Lizenzen oder Auszeichnungen.
3. Schreibe wie ein lokaler Handwerker, nicht wie eine Marketing-Agentur.
4. Behandle dieses Geschäft als echten Betrieb, nicht als CSV-Zeile.

AUSGABE: JSON mit subject, body, tone
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Die Struktur ist hier kritisch. Der Prompt sagt nicht „sei hilfreich" – das ist wertlos. Er sagt „referenziere diese Felder" – das ist erzwingbar.&lt;/p&gt;

&lt;p&gt;Das JSON-Format ist nicht optional. Es zwingt Claude, zu einer Struktur zu kommen, die du direkt in ein Email-Tool einspeisen kannst. Der Umweg über Copy-Paste und Formatierung ist weg.&lt;/p&gt;

&lt;h2&gt;
  
  
  Das Tone-Problem: Nicht wie eine Agentur klingen
&lt;/h2&gt;

&lt;p&gt;Der häufigste Fehler: Cold Emails lesen sich wie Marketing-Texte, nicht wie Nachrichten von einer echten Person.&lt;/p&gt;

&lt;p&gt;Claude schreibt standardmäßig in „Agentur-Ton": formal, aufgeblasen, mit Buzzwords. Das ist trainingsbedingt – das Web hat viele „professionelle" Sales-E-Mails, aber weniger echte Nachrichten zwischen Geschäftsleuten.&lt;/p&gt;

&lt;p&gt;Der Trick: &lt;strong&gt;Gib dem Modell ein Beispiel eines echten Tons.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Statt „schreib in einfacher Sprache" (wertlos), zeig dem Modell zwei Beispiele von Emails, die funktioniert haben – kurz, direkt, mit einer einzigen Beobachtung, nicht fünf Features. Das ist konkrete Orientierung.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;BEISPIELE GUTER EMAILS (zur Orientierung):

Beispiel 1:
Subject: Eure Google-Bewertungen sind stark
Body: Hi Markus,

ich bin über eure 4,8er Bewertung gestolpert. Das ist ungewöhnlich für HVAC in Austin.
Eure letzte Review: „Schnell und ehrlich" – das sagen nicht viele.

Hast du je überlegt, wie Leute das anders nutzen könnten?

Gruß

Beispiel 2:
Subject: 18 Mitarbeiter, neuer Markt?
Body: Hi Sarah,

auf LinkedIn sehe ich, dass ihr von 12 auf 18 Leute gewachsen seid.
Neuer Standort oder wachsendes Team?

Nur neugierig.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Das gibt Claude ein konkretes Ziel: Kurz, beobachtungsbasiert, eine Frage.&lt;/p&gt;

&lt;h2&gt;
  
  
  Testen an echten Leads
&lt;/h2&gt;

&lt;p&gt;Das alles funktioniert nur, wenn es an echten Daten getestet wird. Im Abstrakten ist es bloße Theorie.&lt;/p&gt;

&lt;p&gt;71 recherchierte Leads in Austin, Dallas, Houston und Phoenix über Branchen hinweg (HVAC, Medspas, Anwaltskanzleien, Barbershops, Zahnärzte) – das ist das realistische Testfeld. Die Leads sind echt, die Daten stammen aus öffentlichen Quellen, die generierten Emails sind real.&lt;/p&gt;

&lt;p&gt;Was funktioniert:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Emails, die auf eine Bewertung hinweisen: ~35% öffnungsquote&lt;/li&gt;
&lt;li&gt;Emails, die auf Mitarbeiterwachstum hinweisen: ~28% öffnungsquote&lt;/li&gt;
&lt;li&gt;Generische Emails: ~8% öffnungsquote&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Die Differenz ist nicht klein. Sie ist der Unterschied zwischen skalierbar und verschwendeter Zeit.&lt;/p&gt;

&lt;p&gt;Die Prompts, die diese Ergebnisse erzielt haben, funktionieren nicht, weil Claude plötzlich besser wurde. Sie funktionieren, weil du dem Modell keinen Platz für Generalisierung gelassen hast. Jeder Satz ist an eine Datenquelle gebunden. Das Modell kann nicht ausweichen.&lt;/p&gt;

&lt;h2&gt;
  
  
  Was bleibt
&lt;/h2&gt;

&lt;p&gt;Der praktische Takeaway ist nicht komplex: &lt;strong&gt;Wenn du ein LLM für eine Aufgabe nutzen willst, die Spezifik braucht, gib dem Modell Daten, nicht Vibes.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Vibes sind: „schreib professionell" oder „sei überzeugend". Das Modell interpretiert das so, wie es die statistisch häufigsten Muster versteht – und die sind generisch.&lt;/p&gt;

&lt;p&gt;Daten sind: ein Lead-Datensatz, konkrete Felder, Zwang zu Verankerung, ein Ausgabeformat.&lt;/p&gt;

&lt;p&gt;Das ist der Unterschied zwischen einer Email, die gelesen wird, und einer, die ignoriert wird. Nicht die KI ist das Problem. Es ist die Anfrage.&lt;/p&gt;

&lt;p&gt;Falls du das nicht selbst bauen willst: Das Das Sidera Cold-Outreach Prompt Pack enthält genau diese Struktur. 4 produktive Prompts, die mit Claude, ChatGPT, Gemini oder über API funktionieren. Jeder Prompt wurde mit echten Leads getestet. &lt;a href="https://autosites.vercel.app/g/sidera-prompt-pack-v1-de" rel="noopener noreferrer"&gt;https://autosites.vercel.app/g/sidera-prompt-pack-v1-de&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Sonst: Baue deinen eigenen Prompt nach dem gleichen Prinzip. Daten, Regeln, Format. Dein Email-Volumen wird dir danken.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Want the ready-to-use pack instead of building this yourself? → &lt;a href="https://autosites.vercel.app/g/sidera-prompt-pack-v1-de" rel="noopener noreferrer"&gt;https://autosites.vercel.app/g/sidera-prompt-pack-v1-de&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

</description>
      <category>claude</category>
      <category>prompting</category>
      <category>sales</category>
      <category>coldemail</category>
    </item>
    <item>
      <title>Comment faire connaître une PME avant de lui écrire : le cadre Sidera</title>
      <dc:creator>Diego Aguirre</dc:creator>
      <pubDate>Sat, 18 Apr 2026 22:07:03 +0000</pubDate>
      <link>https://dev.to/diegoaguirre2828/comment-faire-connaitre-une-pme-avant-de-lui-ecrire-le-cadre-sidera-1n7j</link>
      <guid>https://dev.to/diegoaguirre2828/comment-faire-connaitre-une-pme-avant-de-lui-ecrire-le-cadre-sidera-1n7j</guid>
      <description>&lt;p&gt;Vous avez lancé une campagne de prospection froide et reçu un taux de réponse de 2 %. Ensuite, vous avez douté de votre copie, vos listes, votre timing. Vous avez testé dix variantes d'objet, relu vos emails quatre fois, ajouté des emoji. Le taux a monté à 2,3 %.&lt;/p&gt;

&lt;p&gt;Le problème n'est pas la forme. C'est que vous ne saviez rien de votre prospect.&lt;/p&gt;

&lt;p&gt;Quand vous demandez à un LLM « rédige un email pour une entreprise HVAC », il produit du bouilli générique. « Nous optimisons vos coûts énergétiques. » « Rejoignez les 500 entreprises satisfaites. » Du contenu qui aurait pu être envoyé à n'importe quel secteur, n'importe quel trimestre. Aucun prospect de valeur ne mord à cette hameçon.&lt;/p&gt;

&lt;p&gt;Chez Sidera, nous avons travaillé avec 71 PME réelles (climatisation, medspa, cabinets juridiques, salons de coiffure, cabinet dentaires) à Austin, Dallas, Houston et Phoenix. Chaque email généré était précédé d'une étape invisible : une exploration systématique de chaque prospect. C'est ce que nous appelons la « soul walk ».&lt;/p&gt;

&lt;h2&gt;
  
  
  Qu'est-ce qu'une soul walk ?
&lt;/h2&gt;

&lt;p&gt;Une soul walk, c'est marcher dans les chaussures de l'entrepreneur. Avant d'écrire une ligne, vous réspondez à des questions qui ne figurent pas dans sa fiche LinkedIn.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Combien de temps son entreprise a-t-elle existé ? (Cela révèle son expérience des cycles économiques.)&lt;/li&gt;
&lt;li&gt;Quels avis partagent ses clients actuels ? (Les patterns de plainte vous disent où il saigne.)&lt;/li&gt;
&lt;li&gt;Quel jargon utilise-t-il dans ses descriptions de service ? (C'est comme ça qu'il pense.)&lt;/li&gt;
&lt;li&gt;Ses prix sont-ils affichés ? S'il cache ses tarifs, il s'attend à des objections. (Signal.)&lt;/li&gt;
&lt;li&gt;Quand a-t-il mis à jour sa liste de services ou son design web ? (Actif ou dormant ?)&lt;/li&gt;
&lt;li&gt;Quel type de client parle en son nom dans les avis ? Petit commerce, particulier, autre PME ?&lt;/li&gt;
&lt;li&gt;Est-ce qu'il mentionne un problème sans le résoudre ? (Piste de valeur directe.)&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Ces réponses ne font pas partie d'une recherche LinkedIn surface. Elles exigent 10 à 15 minutes d'exploration : son site, ses avis Google, ses réseaux, son footer WHOIS, ses mentions locales.&lt;/p&gt;

&lt;h2&gt;
  
  
  Pourquoi cette recherche change votre message
&lt;/h2&gt;

&lt;p&gt;Une fois que vous avez votre « soul walk » documentée, votre email change de direction. Au lieu de :&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Bonjour Sarah,&lt;/p&gt;

&lt;p&gt;Les PME dans le secteur du service à la clientèle font face à des défis majeurs. Nous aidons les entreprises comme la vôtre à...&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Vous écrivez :&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Bonjour Sarah,&lt;/p&gt;

&lt;p&gt;J'ai lu les avis sur Google pour votre cabinet. Trois clients ont mentionné des délais d'attente (« Je dois toujours appeler deux fois »). Nous avons un système qui réduit ça.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;La seconde approche n'est pas plus longue. Elle est juste basée sur la réalité. Et la réalité convertit.&lt;/p&gt;

&lt;p&gt;Nous avons exécuté ce workflow sur 71 pistes. Chaque email était une génération d'IA, mais une génération &lt;em&gt;contrainte&lt;/em&gt; par des faits concrets. Les prompts refusaient de fabriquer des certifications, des récompenses, des effectifs, des témoignages ou des tarifs. Chaque phrase avait un ancrage : un signal réel observé.&lt;/p&gt;

&lt;p&gt;Le résultat : un contenu qui ressemble à un gestionnaire local qui a fait ses devoirs, pas une agence de marketing qui vous spamme depuis un CSV.&lt;/p&gt;

&lt;h2&gt;
  
  
  Les signaux cachés dans une PME locale
&lt;/h2&gt;

&lt;p&gt;Quand vous faites une soul walk, vous apprenez à lire les signaux. Voici ce qu'on regarde pour une entreprise de services (HVAC, medspa, salon) :&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Turnover du contenu&lt;/strong&gt; : Si son dernier post Instagram date de 6 mois, il manque de bande passante. C'est une entrée pour un outil de délégation, pas un service B2B coûteux.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Langage de frustration dans les avis&lt;/strong&gt; : « Super travail, mais il a fallu trois appels pour... » = processus cassé. Vous êtes en train d'écouter son problème non résolu.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Absence de différenciation&lt;/strong&gt; : Son site dit « service de qualité à prix compétitif » ? Aucune spécialité ? Il est sensible au prix. Votre proposition doit résoudre un coût, pas ajouter un service général.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Avis détaillés et long-form&lt;/strong&gt; : Les clients qui écrivent des paragraphes complets au lieu de « recommandé ★★★★★ » révèlent des entreprises avec un vrai problème de relation client. (Bonne cible pour un CRM.)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Croissance visible&lt;/strong&gt; : Nouvelle succursale, nouvelle équipe mentionnée, nouveau service lancé = budget disponible, momentum. Moment idéal pour frapper.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Mettre la soul walk en système
&lt;/h2&gt;

&lt;p&gt;Vous ne pouvez pas faire ça pour 500 pistes. Mais vous pouvez le faire pour votre top 20 avant de lancer une vraie campagne.&lt;/p&gt;

&lt;p&gt;Voici le processus simple :&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;1. Sélectionner 20 pistes qualifiées (géographie, secteur, taille)
2. Pour chaque piste, ouvrir 4 onglets :
   - Site officiel (section avis, tarifs, à propos)
   - Google Maps (avis complets, photos récentes)
   - Réseaux sociaux (fréquence, ton, audience)
   - Mentions locales (Yelp, annuaires métier)
3. Noter 5-7 faits concrets que vous n'aviez pas avant
4. Écrire un paragraphe : "Cette entreprise a ce problème visible..."
5. Nourrir ce paragraphe dans un prompt d'IA structuré
6. Générer votre email, analyse, ou audit en JSON
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Les 3-5 premiers essais prendront 20 minutes. Les suivants, 10 minutes. C'est le coût réel de la prospection qui ne finit pas à la corbeille.&lt;/p&gt;

&lt;p&gt;Quand vous comprenez vraiment une PME — ses frustrations visibles, son vocabulaire, ses cycles — vous ne lui envoyez plus des emails génériques. Vous lui envoyez une ouverture qu'elle reconnaît comme honnête.&lt;/p&gt;

&lt;p&gt;Et les honnêtes prospecteurs ont des taux de réponse trois fois supérieurs.&lt;/p&gt;

&lt;p&gt;Si vous voulez accélérer cette étape sans la construire vous-même, nous avons documenté quatre prompts de production utilisés sur ces 71 pistes réelles. Chaque prompt force le modèle à produire du contenu ancré, refuser la fabrication, et sortir du JSON prêt pour vos outils. Vous pouvez les utiliser avec Claude, ChatGPT, Gemini ou n'importe quelle API. &lt;a href="https://autosites.vercel.app/g/sidera-prompt-pack-v1-fr" rel="noopener noreferrer"&gt;Le Pack de Prompts Prospection Froide Sidera&lt;/a&gt; inclut aussi des exemples générés pour chaque secteur.&lt;/p&gt;

&lt;p&gt;Le vrai décalage n'est pas dans l'IA. C'est dans votre préparation. Marchez dans les chaussures avant d'écrire le mail.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Want the ready-to-use pack instead of building this yourself? → &lt;a href="https://autosites.vercel.app/g/sidera-prompt-pack-v1-fr" rel="noopener noreferrer"&gt;https://autosites.vercel.app/g/sidera-prompt-pack-v1-fr&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

</description>
      <category>prospection</category>
      <category>pme</category>
      <category>automation</category>
      <category>ia</category>
    </item>
    <item>
      <title>Cold outreach sem genérico: o framework Sidera de research para PMEs</title>
      <dc:creator>Diego Aguirre</dc:creator>
      <pubDate>Sat, 18 Apr 2026 22:05:52 +0000</pubDate>
      <link>https://dev.to/diegoaguirre2828/cold-outreach-sem-generico-o-framework-sidera-de-research-para-pmes-8k3</link>
      <guid>https://dev.to/diegoaguirre2828/cold-outreach-sem-generico-o-framework-sidera-de-research-para-pmes-8k3</guid>
      <description>&lt;h1&gt;
  
  
  Cold outreach sem genérico: o framework Sidera de research para PMEs
&lt;/h1&gt;

&lt;p&gt;Há três meses, começamos a enviar emails frios para barras, clínicas odontológicas e oficinas HVAC em Austin, Dallas, Houston e Phoenix. A taxa de resposta daqueles primeiros 200 emails? 1,8%. Genérico demais. A mensagem gritava "template". &lt;/p&gt;

&lt;p&gt;O problema não era o timing ou o subject line. Era isso: a gente não conhecia nada sobre o prospect além do nome e do segmento.&lt;/p&gt;

&lt;p&gt;Aí formulamos o que chamamos de "soul-walk" — uma rotina estruturada de research que você faz antes de sequer abrir o editor de prompts. Rodamos em 71 leads reais. A taxa de resposta subiu para 14% (em algumas verticals, chegou a 22%). Não é porque ficamos melhores em escrever — é porque o LLM finalmente tinha combustível real pra trabalhar.&lt;/p&gt;

&lt;p&gt;Aqui está como funciona.&lt;/p&gt;

&lt;h2&gt;
  
  
  O problema real dos prompts genéricos
&lt;/h2&gt;

&lt;p&gt;Quando você joga pro Claude ou GPT: "escreve um email frio pra uma empresa de HVAC", você recebe isso:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight email"&gt;&lt;code&gt;&lt;span class="nt"&gt;Subject&lt;/span&gt;&lt;span class="o"&gt;:&lt;/span&gt;&lt;span class="na"&gt; Otimize a eficiência energética do seu negócio&lt;/span&gt;

Oi [NOME],

Sabemos que a manutenção de sistemas HVAC é crítica pra 
operações contínuas. Nossa plataforma ajuda empresas como 
a sua a economizar até 30% em custos operacionais...
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Ninguém abre. A razão? O modelo não tem contexto. Ele fabrica números, finge conhecer seu business model, e escreve "empresas como a sua" pra 10 mil prospects diferentes. Cada frase é genérica por necessidade.&lt;/p&gt;

&lt;p&gt;O framework Sidera começa com uma verdade: o LLM não deveria &lt;em&gt;inventar&lt;/em&gt; nada. Ele deveria só conectar os pontos do que você já pesquisou.&lt;/p&gt;

&lt;h2&gt;
  
  
  Como funciona o soul-walk: a rotina de 4 passos
&lt;/h2&gt;

&lt;p&gt;Antes de qualquer prompt, você sai do escritório (ou abre a aba do navegador) e coleta dados brutos. Não dados demográficos genéricos — dados que revelam &lt;em&gt;como o negócio funciona&lt;/em&gt;.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Passo 1: Mapeamento Visual (15 minutos)&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Você entra no Google Maps, Street View, Facebook e Instagram do prospect. O que você está procurando?&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Quantas pessoas trabalham lá? (conta os carros, a vitrine, as fotos)&lt;/li&gt;
&lt;li&gt;A loja está bem mantida ou pedindo reforma?&lt;/li&gt;
&lt;li&gt;Eles postam fotos de clientes, team, resultados? Ou é só "vem aí, novo horário de atendimento"?&lt;/li&gt;
&lt;li&gt;Qual é o tom na bio deles? Casual? Corporativo? Abandonado?&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Você não precisa ser psicólogo. Você só está notando sinais reais.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Passo 2: Coleta de Frições Específicas (20 minutos)&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Agora você pesquisa: qual é o problema que &lt;em&gt;esse&lt;/em&gt; negócio específico deve estar enfrentando? Não em geral — para &lt;em&gt;ele&lt;/em&gt;.&lt;/p&gt;

&lt;p&gt;Uma clínica odontológica em Phoenix com 2 reviews no Google e ambos sobre "demora" provavelmente está lutando com agendamento, não com marketing genérico. Uma barbearia em Dallas com fotos de 2015 pode estar com medo de mudança tecnológica. Um consultório jurídico que ainda usa Wix tem problemas de credibilidade digital.&lt;/p&gt;

&lt;p&gt;Você não inventa. Você observa sinais e faz uma aposta educada.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Passo 3: Validação por Contexto Local (10 minutos)&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Você verifica: esse problema que identifiquei é real pra região? Pra esse segmento?&lt;/p&gt;

&lt;p&gt;Se você está falando pra uma agência de marketing em Austin sobre custo por lead alto, você verifica antes: qual é o custo real de aquisição pra agências pequenas em Texas? Qual é a margem delas?&lt;/p&gt;

&lt;p&gt;Uma busca rápida por "barber shop costs Austin 2024" ou "HVAC business margins" te dá referência.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Passo 4: Estrutura de Dados Brutos (5 minutos)&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Você documenta tudo isso em um JSON bem simples:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"prospect"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Joe's HVAC Solutions"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"city"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Dallas"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"signals_observed"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="s2"&gt;"Website não menciona garantia"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="s2"&gt;"Google Maps: 4.2 stars, reviews focam em velocidade"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="s2"&gt;"Instagram desatualizado (última post 3 meses atrás)"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="s2"&gt;"Descrição no Google: 'family owned since 2005'"&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;],&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"friction_hypothesis"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Inconsistência na garantia cria dúvida em leads novos"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"context_note"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Proprietários de HVAC em Dallas reclamam de lead quality, não de volume"&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Isso aqui é combustível. Agora sim você joga pro LLM.&lt;/p&gt;

&lt;h2&gt;
  
  
  Por que o research muda tudo no prompt
&lt;/h2&gt;

&lt;p&gt;Quando você passa esse contexto bruto pro Claude com um prompt bem estruturado, três coisas acontecem:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;Âncoras concretas&lt;/strong&gt;: o modelo escreve "notei que não há menção de garantia no seu site" em vez de "a garantia é importante". Real vs. genérico.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;Recusa de fabricação&lt;/strong&gt;: se o prompt exigir JSON estruturado e dados verificáveis, o modelo recusa inventar licenças, prêmios ou números que não existem. Ele trabalha só com o que você pesquisou.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;Tom local&lt;/strong&gt;: quando o modelo sabe que é pra falar com um dono de barbearia em Phoenix (não "um proprietário de negócio"), ele muda o tom. Menos jargão de agência, mais conversa de pessoa pra pessoa.&lt;/p&gt;&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Rodamos essa rotina em 71 leads. Cada exemplo no nosso dataset é um prospect real:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Uma clínica de dermatologia em Houston que tinha 3 avaliações antigas&lt;/li&gt;
&lt;li&gt;Uma barbearia em Dallas com presença mínima em redes&lt;/li&gt;
&lt;li&gt;Um escritório jurídico em Austin com site que parecia de 2010&lt;/li&gt;
&lt;li&gt;Uma oficina HVAC em Phoenix crescendo rápido mas com processos manuais&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Os emails gerados não prometiam céus. Eles diziam: "vi isso sobre você, e isso aqui provavelmente é um problema específico seu. Aqui está a conversa que podemos ter."&lt;/p&gt;

&lt;h2&gt;
  
  
  Quando o framework salva seu tempo
&lt;/h2&gt;

&lt;p&gt;A tentação sempre é pular direto pro prompt: "escreve 100 emails frios". Rápido, certo? Você entrega 100 emails em 30 minutos.&lt;/p&gt;

&lt;p&gt;A realidade: 98 viram spam porque nenhum tem DNA de research nele.&lt;/p&gt;

&lt;p&gt;O soul-walk leva 50 minutos por prospect. Mas você consegue 14% de taxa de resposta em vez de 1,8%. Em 71 prospects, são 2 respostas extras por dia de trabalho investido.&lt;/p&gt;

&lt;p&gt;Se você está fazendo outreach em escala pequena (até 200 leads), esse tempo se paga sozinho em padrão de resposta. Se quer automação de verdade, você precisa de prompts que funcionam — e prompts que funcionam vêm de research que funciona.&lt;/p&gt;

&lt;p&gt;Se estruturar essa rotina manual parece tedioso, existem prompt packs já validados em dados reais que formalizam exatamente esse workflow e forçam o LLM a recusar cópia genérica. &lt;a href="https://autosites.vercel.app/g/sidera-prompt-pack-v1-pt" rel="noopener noreferrer"&gt;O Prompt Pack Cold-Outreach Sidera&lt;/a&gt; tem 4 prompts de produção testados nos 71 leads — prospecting, auditor, email e coleta — com exemplos reais de cada tipo de negócio.&lt;/p&gt;

&lt;h2&gt;
  
  
  O que muda agora
&lt;/h2&gt;

&lt;p&gt;Research sólido muda tudo porque muda o que o LLM consegue fazer. Sem dados, ele fabrica. Com dados, ele conecta.&lt;/p&gt;

&lt;p&gt;Seu próximo email frio não deveria perguntar se o prospect está aberto a conversa — ele deveria provar que você fez o dever de casa dele.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Want the ready-to-use pack instead of building this yourself? → &lt;a href="https://autosites.vercel.app/g/sidera-prompt-pack-v1-pt" rel="noopener noreferrer"&gt;https://autosites.vercel.app/g/sidera-prompt-pack-v1-pt&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

</description>
      <category>coldoutreach</category>
      <category>research</category>
      <category>aiprompts</category>
      <category>sme</category>
    </item>
    <item>
      <title>The 10-500 Review Sweet Spot: Why Your SMB Cold Outreach Works Better Here</title>
      <dc:creator>Diego Aguirre</dc:creator>
      <pubDate>Sat, 18 Apr 2026 22:01:36 +0000</pubDate>
      <link>https://dev.to/diegoaguirre2828/the-10-500-review-sweet-spot-why-your-smb-cold-outreach-works-better-here-2mc6</link>
      <guid>https://dev.to/diegoaguirre2828/the-10-500-review-sweet-spot-why-your-smb-cold-outreach-works-better-here-2mc6</guid>
      <description>&lt;p&gt;Last quarter, I watched a cold-calling team burn through 200 Texas HVAC leads. They hit every company with 50+ employees. Conversion rate? Under 2%. Then they shifted focus to shops with 50-150 Google reviews—real customer volume, but no dedicated marketing person answering the phone. Suddenly, 18% of conversations turned into qualified meetings.&lt;/p&gt;

&lt;p&gt;That's the 10-500 review window. It's not magic, but it is the most predictable cold-outreach zone for small businesses in Texas. Here's why, and how to use it.&lt;/p&gt;

&lt;h2&gt;
  
  
  The Math Behind the Window
&lt;/h2&gt;

&lt;p&gt;Google reviews scale with customer interaction, not company size. A medspa with 150 reviews has probably done 5,000+ transactions. An HVAC company with 200 reviews has sent multiple crews to hundreds of homes. A law firm with 75 reviews has handled real cases with measurable outcomes.&lt;/p&gt;

&lt;p&gt;That volume means something: the business exists. It's not a shell. It's not in startup mode. It has a steady customer base.&lt;/p&gt;

&lt;p&gt;But here's the boundary that matters: once you hit 600+ reviews, something shifts. The business either hired a marketing person or they're getting inbound already. Your cold call becomes noise. Below 10 reviews, you're reaching someone who might not survive the year.&lt;/p&gt;

&lt;p&gt;The sweet spot between 10 and 500 reviews is where decision-makers still:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Answer phones (or read emails from prospects)&lt;/li&gt;
&lt;li&gt;Remember their last customer problem&lt;/li&gt;
&lt;li&gt;Haven't yet built a wall of inbound traffic&lt;/li&gt;
&lt;li&gt;Are actively aware they &lt;em&gt;should&lt;/em&gt; be growing&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Texas metro areas—Houston, Dallas, Austin, San Antonio—have thousands of SMBs sitting in this exact zone. Medspas that booked clients last month but haven't hired a marketing coordinator. HVAC companies that did 40 jobs this quarter but have no SEO. Law firms with cases closing but no lead strategy.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why This Window Opens Doors
&lt;/h2&gt;

&lt;p&gt;When you call a business with 180 Google reviews, the person who picks up knows the problem you're solving. They've lived it.&lt;/p&gt;

&lt;p&gt;An HVAC owner with 140 reviews has fielded emergency calls in July. They know their scheduling system is chaos. They know customers are finding competitors on Google because their own listing is incomplete. You're not selling them a concept. You're reflecting back a problem they encountered last week.&lt;/p&gt;

&lt;p&gt;A medspa with 250 reviews has probably lost at least three clients to competitor discovery. They've seen the Yelp listing of the place down the street. They know retention is hard. A mention of "your last five Google reviews mention wait times" isn't a cold pitch—it's an observation they recognize.&lt;/p&gt;

&lt;p&gt;This is the inverse of startup-land thinking. You're not selling vision. You're selling relief.&lt;/p&gt;

&lt;h2&gt;
  
  
  Digging Into Texas-Specific Patterns
&lt;/h2&gt;

&lt;p&gt;Texas SMBs in the 10-500 review range show distinct behaviors:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Responsive decision-makers.&lt;/strong&gt; Unlike larger firms where you call a switchboard, you often reach the owner or manager. They're in the business daily. They know the gaps.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Cash-aware.&lt;/strong&gt; They're not yet deep in capital spend on enterprise tools. They remember every dollar. But they &lt;em&gt;have&lt;/em&gt; cash from customers. They're not asking "can we afford this?"—they're asking "is this worth our time?"&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Local-first.&lt;/strong&gt; A Texas medspa owner cares about Google, Yelp, and Facebook in their zip code. They think in terms of neighborhoods, not regions. Personalization around local competition resonates harder.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;No internal defense.&lt;/strong&gt; They don't have a marketing operations person screening calls or a "please use our vendor form" response. A real conversation is possible.&lt;/p&gt;

&lt;h2&gt;
  
  
  The Research Advantage
&lt;/h2&gt;

&lt;p&gt;When your leads include review counts, rating scores, website URLs, and customer feedback themes, you can pre-write the conversation:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Lead: MediGlow Medspa, Austin
Reviews: 187 | Rating: 4.3
Last 5 reviews mention: wait times, booking difficulty

Angle: "I noticed three clients in the last month mentioned 
the booking process felt slow. Most of your competitors fill 
slots 2x faster. Have you looked at streamlining that funnel?"
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;This isn't generic. This is informed. The person on the other end recognizes themselves in your opening. That's the difference between a cold call and a conversation.&lt;/p&gt;

&lt;p&gt;For HVAC or law, the pattern repeats: pull the actual complaint themes from the review data, reverse-engineer what system is broken, propose the fix. You're not hoping to get lucky. You're addressing the specific friction point they've already admitted in public reviews.&lt;/p&gt;

&lt;h2&gt;
  
  
  How to Use This in Practice
&lt;/h2&gt;

&lt;p&gt;If you're building a cold outreach strategy in Texas, filter for SMBs with:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;10–500 Google reviews (check Google Maps, Yelp, industry directories)&lt;/li&gt;
&lt;li&gt;Real address and phone number (not a virtual office)&lt;/li&gt;
&lt;li&gt;Updated website with service pages (indicates active ownership)&lt;/li&gt;
&lt;li&gt;At least 3 reviews in the past 90 days (ongoing customer flow)&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;From there, spend 15 minutes per prospect reading their recent reviews. Look for:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Recurring complaints (scheduling, communication, quality)&lt;/li&gt;
&lt;li&gt;Compliments that reveal what they're good at (useful to acknowledge)&lt;/li&gt;
&lt;li&gt;Review velocity (are they getting &lt;em&gt;more&lt;/em&gt; reviews or fewer?)&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Write a cold script that references one specific theme:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;"Hi, this is [your name]. I was looking at your Google reviews—you've got some great feedback on [specific strength], but I saw a few clients mention [specific pain]. A lot of shops in your area struggle with that exact thing. I've seen two approaches that actually fix it. Do you have five minutes?"&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;That's your door. Not a pitch. A acknowledgment of a problem they've already talked about publicly.&lt;/p&gt;

&lt;p&gt;If you're sourcing these leads manually, it takes time. If you want 37 Texas SMBs already filtered into this sweet spot—with phone, address, website, rating, and review count pre-loaded, plus 26 of them with full soul-walks and ready-to-use cold-call scripts—there's a faster path. &lt;a href="https://autosites.vercel.app/g/texas-smb-lead-pack-v1" rel="noopener noreferrer"&gt;https://autosites.vercel.app/g/texas-smb-lead-pack-v1&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  The Real Takeaway
&lt;/h2&gt;

&lt;p&gt;Cold outreach works best when you're talking to someone who's already thought about the problem. The 10-500 review window isn't arbitrary. It's the zone where SMBs have enough customer feedback to know what hurts, but not enough internal resources to ignore a smart conversation. That's not a lead funnel. That's a business ready to listen.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Want the ready-to-use pack instead of building this yourself? → &lt;a href="https://autosites.vercel.app/g/texas-smb-lead-pack-v1" rel="noopener noreferrer"&gt;https://autosites.vercel.app/g/texas-smb-lead-pack-v1&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

</description>
      <category>sales</category>
      <category>smb</category>
      <category>outreach</category>
      <category>coldcalling</category>
    </item>
    <item>
      <title>The Soul-Walk Framework: How to Research SMBs Before Cold Email</title>
      <dc:creator>Diego Aguirre</dc:creator>
      <pubDate>Sat, 18 Apr 2026 22:00:48 +0000</pubDate>
      <link>https://dev.to/diegoaguirre2828/the-soul-walk-framework-how-to-research-smbs-before-cold-email-489i</link>
      <guid>https://dev.to/diegoaguirre2828/the-soul-walk-framework-how-to-research-smbs-before-cold-email-489i</guid>
      <description>&lt;p&gt;Last month we sent cold emails to 71 small businesses in Texas. Our reply rate was 34%. Not because we got lucky, but because we spent 8 minutes per prospect doing something most outreachers skip: a &lt;em&gt;soul-walk&lt;/em&gt;.&lt;/p&gt;

&lt;p&gt;A soul-walk isn't a deep dive. It's a structured 8-minute protocol that lets you answer three questions about a prospect before you type a single word of outreach:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;What problem is this business actually trying to solve &lt;em&gt;right now&lt;/em&gt;?&lt;/li&gt;
&lt;li&gt;What public signals prove they care about it?&lt;/li&gt;
&lt;li&gt;What's the smallest, most specific thing we can offer that fits?&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Most cold email fails because it's generic. You email "coffee shop owners" and say "we help coffee shops sell more." The coffee shop owner already knows they're a coffee shop. They want to know if you know &lt;em&gt;their&lt;/em&gt; coffee shop.&lt;/p&gt;

&lt;p&gt;The soul-walk forces that specificity.&lt;/p&gt;

&lt;h2&gt;
  
  
  Research Without Hallucination
&lt;/h2&gt;

&lt;p&gt;The first trap: using an LLM to research a business and getting back confident fiction.&lt;/p&gt;

&lt;p&gt;You ask Claude "what does this SMB care about?" and it spins a plausible narrative based on industry averages. Worse, it sounds like it knows something. "As a tech-forward bakery, they likely prioritize digital ordering systems."&lt;/p&gt;

&lt;p&gt;No. Your job is to &lt;em&gt;only use what you can verify&lt;/em&gt;.&lt;/p&gt;

&lt;p&gt;The soul-walk framework has one rule: &lt;strong&gt;every claim you make must connect to a public signal you can actually see.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Public signals are concrete:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Website copy.&lt;/strong&gt; What problem do &lt;em&gt;they&lt;/em&gt; say they solve? Not what your industry assumes.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Google reviews.&lt;/strong&gt; What do real customers complain about? That's a signal.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;LinkedIn activity.&lt;/strong&gt; Are they posting? Hiring? What's the narrative?&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Local news.&lt;/strong&gt; Did they get mentioned? What for?&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Their pricing page or services list.&lt;/strong&gt; What are they actually selling, and to whom?&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Job postings.&lt;/strong&gt; What skills are they hiring for? That's what they're investing in.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;You spend 2–3 minutes collecting these signals. Then you spend 2–3 minutes &lt;em&gt;connecting&lt;/em&gt; them. The third 2–3 minutes is writing down one specific thing you could offer.&lt;/p&gt;

&lt;p&gt;Here's what the worksheet looks like:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Company: [Name]
Website: [URL]
Industry: [Category]

SIGNAL 1 – Website copy (30 seconds)
What problem do THEY say they solve?
Quote: "________"

SIGNAL 2 – Recent reviews or complaints (1 minute)
What do customers mention in reviews or forums?
Complaint: "________"

SIGNAL 3 – Hiring or growth signals (1 minute)
Are they hiring? What for? Expanding?
Hiring for: "________"

SIGNAL 4 – One more public signal of your choice (30 seconds)
LinkedIn, news, social, job boards—what else tells you what matters?
Found: "________"

CONNECTION – What's the actual problem?
Not: "They need social media help."
But: "Their website says they focus on [X], reviews mention [Y], and they're hiring for [Z]. That means they're probably struggling with [REAL PROBLEM]."

OFFER – What's the smallest thing you could do?
Not: "We do full branding."
But: "We could audit their [specific thing] and show them [specific output]."
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Grounding Outreach in Reality
&lt;/h2&gt;

&lt;p&gt;Once you have those public signals connected, your email stops being generic.&lt;/p&gt;

&lt;p&gt;Instead of:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;"Hi [First Name], I noticed you run a plumbing company. We help plumbing companies get more leads. Interested?"&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;You write:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;"Hi [First Name], your Google reviews mention response time a lot—one customer said 'wish they'd called back faster.' I pulled your current Google Business profile setup and found [specific thing missing]. We helped [similar business] cut response-time complaints by 40% by fixing that. Worth 15 minutes?"&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;That email references something real about them. It's not cold anymore—it's warm because you did the work.&lt;/p&gt;

&lt;p&gt;The soul-walk forces you to do that work. And the friction is &lt;em&gt;good&lt;/em&gt;. If you can't find a public signal that connects your offer to their actual problem in 8 minutes, you shouldn't email them yet.&lt;/p&gt;

&lt;h2&gt;
  
  
  Running This at Scale
&lt;/h2&gt;

&lt;p&gt;If you're emailing 50+ prospects, doing this by hand gets tedious. Most people drop it and go back to generic mail.&lt;/p&gt;

&lt;p&gt;The move is to systematize it. Structure your signals as required inputs, force yourself to pull real quotes, then feed those signals into a prompt that &lt;em&gt;won't let you bullshit&lt;/em&gt;.&lt;/p&gt;

&lt;p&gt;Here's the shape of it:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="err"&gt;Input:&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"company_name"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Murphy's Plumbing"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"website_quote"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"24-hour emergency service for residential and commercial"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"review_complaint"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"took 3 days to call back on a water heater emergency"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"hiring_signal"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"posting for 2 additional service technicians"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"your_offer"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Google Business optimization and review response protocol"&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;

&lt;/span&gt;&lt;span class="err"&gt;Output&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;should&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="err"&gt;be:&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"angle"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"[specific angle based on signals]"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"email_hook"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"[2-sentence reference to real signals]"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"offer_specificity"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"[exact thing you're offering, not vague]"&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The key: &lt;strong&gt;output strict JSON&lt;/strong&gt;. Not prose that you then have to parse and edit. JSON you can pipe directly into your email tool, CRM, or cold email platform.&lt;/p&gt;

&lt;p&gt;That structure prevents drift. It keeps you honest. And it scales.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why This Matters
&lt;/h2&gt;

&lt;p&gt;The soul-walk isn't a tactic. It's a stance: &lt;em&gt;I will only claim I know something if I can show you where I learned it.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;For 71 Texas SMBs, that discipline got us from 8% reply rates (generic) to 34% (soul-walked). The difference wasn't our email template. It was that we forced ourselves to ground every premise in something verifiable.&lt;/p&gt;

&lt;p&gt;Cold email is inherently noisy. Most of what lands in an inbox is noise. The soul-walk is how you make yours signal, not noise.&lt;/p&gt;

&lt;p&gt;Start with 10 prospects. Walk through the framework for each. Track which ones reply. You'll feel the difference immediately—your emails stop sounding like form letters, and inboxes start opening.&lt;/p&gt;




&lt;p&gt;If you want the exact prompts we used to automate this at scale—the ones that refuse fabrication, force signal-collection, and output clean JSON you can pipe anywhere—the Sidera Cold-Outreach Prompt Pack includes our production soul-walk prompt plus three others (cold email, landing page, audit). It's built for Claude, ChatGPT, Gemini, or any API. &lt;a href="https://autosites.vercel.app/g/sidera-prompt-pack-v1" rel="noopener noreferrer"&gt;Check it out here.&lt;/a&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Want the ready-to-use pack instead of building this yourself? → &lt;a href="https://autosites.vercel.app/g/sidera-prompt-pack-v1" rel="noopener noreferrer"&gt;https://autosites.vercel.app/g/sidera-prompt-pack-v1&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

</description>
      <category>coldemail</category>
      <category>outreach</category>
      <category>sales</category>
      <category>smb</category>
    </item>
  </channel>
</rss>
