DEV Community: Prompt & Skill Co.

The TikTok Shop Product Title Formula: The 200 Characters That Decide If Anyone Finds You

Prompt & Skill Co. — Wed, 21 Oct 2026 09:32:52 +0000

Most TikTok Shop sellers treat the product title as paperwork — something you type once so the listing can go live. The platform treats it differently: the title is a primary input to search, to the recommendation feed's understanding of your product, and to the listing-quality score TikTok assigns every page. Sellers who audit underperforming listings keep finding the same thing: before the product was wrong, before the price was wrong, the first 40 characters were wrong.

Why 40 characters run your listing

Two facts govern everything below:

Mobile truncates. Most TikTok shopping happens on a phone, in a feed, at scroll speed. Roughly the first 30–40 characters are what a buyer actually sees before deciding whether your product is even relevant. Everything after that is for the algorithm and for the small minority who open the page.
TikTok scores completeness and accuracy. Listings are graded on how complete and truthful their information is, and that grade feeds organic visibility. A vague title doesn't just under-sell — it quietly suppresses the listing.

The formula

Working titles on the platform follow a stable order:

[Brand] + [Product Type] + [Primary Benefit or Use] + [Key Feature or Spec] + [Size / Quantity / Variant]

Run it on a real-shaped example. Weak: "Mini Humidifier Cute Portable USB Cool Mist 500ml For Bedroom Office Travel Gift". The information is all there — in scrambled order, with the benefit implied and filler ("cute", "gift") eating the front.

Rebuilt: "Mini Humidifier for Bedroom — 500ml Cool Mist, USB, 30H Runtime, Quiet Portable". Same facts. But the product type and the use case land inside the first screen, the deciding specs carry numbers, and nothing up front is decoration.

The working rules inside the formula:

Product type + main use first. That's the search match and the human skim, in one move.
Numerals, always. "2 Pack", "500ml", "30H" — digits stop eyes and match how buyers type searches.
Specs beat adjectives. "Quiet (26dB)" outperforms "super quiet". Numbers are checkable; adjectives are noise.
One of each thing. No repeated keywords, no stuffing every variant into the title — TikTok flags repetition, and buyers read it as desperation.
Capitalize normally. ALL CAPS and special-character confetti read as spam to both the algorithm and the human.

What gets titles — and listings — in trouble

The platform's rejection and suppression triggers are worth memorizing because half of them are habits sellers import from other marketplaces:

Promotional language in the title ("Best Seller", "Hot Deal", "#1 Trending") — exaggerated claims are a compliance violation, not a strategy.
Off-platform direction anywhere on the page — QR codes, "link in bio", external store mentions. This is the fastest route to a suppressed listing.
Competitor platform names and unapproved health or performance claims in the description.
Mismatch between title and product — if the title promises a variant or quantity the page doesn't deliver, your listing score and your reviews both pay for it.

The description's only job

If the title gets you found and clicked, the description closes. The pattern on converting pages is consistent: lead with the problem and the change the product makes in a real routine, then specs, then specifics (what's in the box, shipping, returns). Show what it does before listing what it is. "Real beats perfect" — pages written like a person explaining the product to a friend outperform pages written like a warehouse catalog.

The habit that compounds

Revisit every listing every 2–4 weeks and change one variable at a time — title first, then main image, then price. Sellers who rewrite everything at once learn nothing, because they can't tell which change moved the number. The title is the cheapest test you own: no new photos, no new stock, ten minutes, and it touches search, feed comprehension, and listing score simultaneously.

If your shop has views but no orders, don't start by blaming the product or slashing the price. Open your three best listings and read only the first 40 characters of each title, the way a stranger scrolling at midnight would. That read will tell you most of what you need to fix.

I keep a pre-publish listing checklist for exactly this pass — titles, image honesty, claims, pricing truth, policy zero-tolerance checks — inside my seller toolkit. The five starter prompts (first creator DM plus four video scripts) are free:

Free mini-pack (5 prompts): https://promptandskillco.gumroad.com/l/free-tiktok-shop-mini-pack

The full AI Toolkit for TikTok Shop Sellers has the checklist plus 50 prompts (30 video scripts, 12 creator outreach, 8 live selling), and the Creator Outreach Tracker spreadsheet. Personal $39 / Commercial $79, one-time: https://promptandskillco.gumroad.com/l/ai-toolkit-for-tiktok-shop-sellers

Disclosure: I sell this toolkit. Title limits and listing rules shift by market; check your Seller Center for your region's current version.

How to Get Clients to Approve Social Media Content Faster — The Batch Approval Workflow I Use for Every Client

Prompt & Skill Co. — Wed, 21 Oct 2026 09:30:53 +0000

If you manage social media for clients, you already know the math problem nobody warns you about: the writing is not the job. The job is the loop around the writing — drafts in one doc, feedback in three DM threads, a post quietly waiting two weeks for a one-word approval, and a monthly report rebuilt from memory at midnight.

This is the approval workflow I rebuilt after one too many of those months. It runs on a spreadsheet and plain-text prompts — no specific platform required — and it scales from 2 clients to about 10 before you need real tooling.

Why approvals actually stall

Three reasons, and none of them is "clients are difficult":

Approving is annoying in chat. One post at a time, in a DM thread, with no clear ask. Your client means to reply; the thread scrolls away.
Feedback has no home. Comments arrive by email, DM, voice note and doc comments. You reconcile four channels before you can revise anything.
Nobody can prove what was approved. When a client asks "who signed off on this?", the answer shouldn't be a screenshot hunt.

Every fix below maps to one of these.

Step 1: Produce in batches, from one brief

Never write posts one at a time on posting day. For each client, take the brief (what they sell, audience, voice, pillars, off-limits topics, this month's verified promos) and generate the whole month's idea batch first: working titles, hooks, formats, platform variants. Then write the batch — scripts, carousels, captions — in one sitting per client.

Context-switching between clients daily is where quality and hours leak. Batching closes a client's context before you open the next one.

Step 2: Give every post a row and a status

One tracker for all clients. One row per post, same columns for everyone:

Column	Why it exists
Client	Keeps multi-client work separated in one file
Platform	TikTok / Instagram / LinkedIn / …
Working title	The handle you refer to in messages
Scheduled date	This column is your content calendar
Status	See the ladder below
Client feedback	Their words, next to the post — not in a DM
Published date	Closes the loop for reporting

Status ladder, fixed for every client:

Draft → In Review → Revision Requested → Approved → Scheduled → Published

The rule that makes it work: if a post isn't a row with a status, it doesn't exist. Add dropdowns for status/platform/client, and a small summary that counts posts per status, your approval rate ((Approved + Scheduled + Published) / Total), and published posts per client. With spreadsheet formulas this takes ten minutes to build and updates itself forever.

Step 3: Send approvals as one package

This is the single highest-leverage change. One message per batch, containing:

What's in the batch (titles, platforms, planned dates)
Where to view everything (one folder/link)
The exact ask: approve or request changes per item — never one blanket reply
The deadline that keeps the schedule, and what silence means under your agreement (pause posting, or proceed — decide at onboarding, not mid-crisis)
One closing question, so replying is easy

Then the nudge rule: one gentle reminder with the deadline restated and the items still unmarked listed by name. One follow-up, written kindly, then the agreement's silence rule applies. Chasing beyond that trains clients to ignore you.

And when feedback comes back as "this feels off"? Don't rewrite on a guess. Reply by translating it into a written change list ("shorter hook / warmer tone / remove the price mention — did I miss anything?") and get a yes on the list before you touch the draft. Vague feedback converted into a checklist is a deliverable, not a delay.

Step 4: Report from exports, never from memory

At month end, the report almost writes itself:

Shipped work = your tracker's Published rows
The numbers = platform exports only, every metric with its source, quoted exactly
Highlights = top posts and what made them work
What didn't work = named plainly, with the likely cause
What the data can't tell us = a short mandatory section — the one that makes clients trust everything above it
Next month = keep / stop / start, and decisions you're waiting on

Attribution honesty matters: if a post was boosted, say the reach jump came with ad spend attached. Clients can smell a padded report, and one honest limits section buys more trust than ten glossy charts.

The mistakes that keep this broken

Sending posts for approval one by one (see Step 3 — it's the root of most stalls)
Letting feedback live in DMs because "the client prefers it" — mirror it into the tracker row yourself, every time
A different process per client — same columns, same ladder, same package format; only the content changes
Reporting from impressions — if a number isn't in an export, it doesn't go in the report; mark it missing instead of estimating

Steal this workflow

You can build everything above from this article: the tracker columns, the status ladder, the approval package anatomy, the report structure. If you'd rather start from a finished version, I packaged the one I use — 50 fill-in prompts covering batch production, approval & client communication, and reporting, plus the tracker spreadsheet with the formulas built in and the report template with a filled example:

Free mini pack (5 full prompts, including the batch approval request and the monthly-report highlights prompt): https://promptandskillco.gumroad.com/l/free-smm-mini-pack
Full system (Personal $39 / Commercial $79 for client work): https://promptandskillco.gumroad.com/l/social-media-manager-production-system

Disclosure: that's my product, sold through my Gumroad store. Every example in it is fictional and labeled as such — the prompts run only on facts and numbers you paste in. The workflow in this article stands on its own either way; questions welcome in the comments.

Your Samples Aren't the Problem. Your Creator Picks Are: TikTok Shop Seeding, by the Numbers

Prompt & Skill Co. — Wed, 21 Oct 2026 07:41:31 +0000

Ask a struggling TikTok Shop seller what their affiliate plan is and you'll hear a quantity: "We're sending out 200 samples this month." Ask a profitable one and you'll hear a filter. The data on seeding is unusually clear on this point, and it points the opposite way from most sellers' instincts.

The correlation that should change your budget

One widely circulated analysis of TikTok Shop affiliate programs ran the correlations between seeding activity and GMV. Two numbers from it are worth taping to your monitor:

Monthly samples sent ↔ GMV: correlation ≈ 0.28. Weak. How much you seed barely predicts how much you sell.
Top-10 affiliate revenue ↔ GMV: correlation ≈ 0.86. Strong. Who your best creators are predicts almost everything.

Read that again: the count of free products leaving your warehouse is close to noise. The identity of your top creators is close to destiny. Sellers running efficient programs describe the same shape from the inside — of everything they seed, 90–95% produces no direct sales, and the program is carried by the 5–10% of creators who connect. Mass seeding doesn't fail because samples are bad. It fails because it spends the budget randomly across a distribution that was always going to be top-heavy.

What "picking better" actually looks like

If volume isn't the lever, selection is. The criteria experienced sellers use are boring and checkable:

Average views, not followers. Follower count is one of the worst signals in the system. A 40k-follower creator averaging 20k views per post is a better bet than a 500k account averaging 2k. Judge the last 10 posts, not the bio number.

Comments that contain buying questions. Open a creator's recent videos and read the comments. "Where did you get that?", "Does it work for X?", "Link?" — that is an audience already shopping. A creator whose comments are all fire emojis has reach; a creator whose comments are purchase intent has a register. For a seller, the second one wins.

Already demonstrating products like yours. If a creator's feed shows them competently demoing products in your category, the sample you send slots into an existing skill. If they've never posted a product video, your sample is funding their first experiment — sometimes worth it with hungry small creators, but know that's the bet you're making.

Fit with the niche one step over. The buyers of your product follow adjacent creators — the desk-setup account for the desk gadget, the dorm account for the home product. Searching your buyer's feed beats searching your product's keywords.

And once you've picked: don't kill it in operations

Two operational failures quietly burn more sample budget than bad picking:

Slow shipping. Momentum on this platform is perishable. A creator who was excited on Monday and receives product the following week posts something polite, if anything. Sellers who treat sample fulfillment as same-day work keep the energy they paid for.

Silence after the sample lands. The sample arriving is the start of the highest-leverage week in the relationship: one message confirming it arrived, the terms restated in one line (commission, disclosure on, post whenever it fits), and then — if nothing posts — one polite nudge with something useful in it. Not a guilt trip; creators who feel managed stop replying, and TikTok tracks sample obligations on its side anyway. Your job is to be the seller whose samples creators want to post, because the next launch's outreach starts from this launch's reputation.

Track all of it per creator — picked why, sample date, posted date, attributed orders — because after two cycles, that sheet is the answer to the only question that matters: which five names do I build the program around?

The reframe

Stop asking "how many samples should I send?" The better question is "which twenty creators, chosen on views and buying-intent comments, get my samples this month — and what do I do in the seven days after each box lands?" Twenty well-picked samples with fast shipping and a real follow-up will beat two hundred sprayed into the void, and they'll cost you a tenth as much to learn from.

My own seeding runs on the system in my toolkit — sample terms, follow-ups, briefs, reactivations, and the tracker that keeps the per-creator record. The five starter prompts (including the first outreach DM and four video scripts) are free:

Free mini-pack (5 prompts): https://promptandskillco.gumroad.com/l/free-tiktok-shop-mini-pack

The full AI Toolkit for TikTok Shop Sellers adds the other 45 prompts (30 video scripts, 12 creator messages, 8 live support scripts), the Creator Outreach Tracker with live formulas, and a pre-publish listing checklist. Personal $39 / Commercial $79, one-time: https://promptandskillco.gumroad.com/l/ai-toolkit-for-tiktok-shop-sellers

Disclosure: I sell this toolkit. The correlations cited are from a publicly circulated analysis of TikTok Shop affiliate programs; treat them as directional, not as a promise about your shop.

The Outreach Math: What 100 Creator DMs Actually Turn Into on TikTok Shop

Prompt & Skill Co. — Wed, 21 Oct 2026 07:40:13 +0000

Every TikTok Shop seller I talk to has the same story: "I messaged 30 creators and got nothing, so affiliate doesn't work for my product." Thirty messages is not a test of affiliate. It is a rounding error. Here is the arithmetic the channel actually runs on — the cold-start funnel numbers that experienced sellers and agencies consistently report — and what they mean for how you should budget your week.

The funnel, with real numbers

Cold outreach to creators you've never worked with converts roughly like this:

Replies: 1–4 out of 100. Reported cold-start response rates cluster at 1–3%, with new shops citing 2–4% when the messages are personalized. Either way: single digits.
Sample requests: about half of replies. A creator who replies is usually interested; roughly 50% of responders will ask for a sample.
Posts: about 70% of samples. Of creators who receive product, around 70% actually post within a reasonable window.
Posts that generate real ROI: 10–15%. Most posted videos do very little. A small minority pays for the whole program.

Multiply it through: 100 outreaches → ~3 replies → ~1.5 sample requests → ~1 posted video → a ~0.15 chance that any single batch of 100 produces a money-making video. Sellers who run this professionally summarize it as roughly one ROI-generating video per 100 cold outreaches — and they plan accordingly.

TikTok itself underwrites this math: a brand-new shop gets an allocation of around 1,000 outreach messages. At a 2–4% response rate, that's 20–40 affiliate conversations from your first push. The platform is telling you, in its own units, that this is a volume game with a long funnel — not a "message ten influencers" game.

What the math changes about your behavior

1. Your sample size is the strategy. Judging affiliate after 30 DMs is like judging a coin after two flips. The sellers who "got lucky" with one small creator whose video took off ran enough attempts for luck to have a surface area. Practitioners commonly cite 50–100 creator videos before concluding a product doesn't fit the channel at all.

2. Follow-up is not optional — it's half the replies. One message is rarely enough; creators are busy and inboxes are brutal. A single polite follow-up 3–5 days later, carrying one new piece of information (not "just bumping this"), is where a large share of responses actually come from. Then stop. Two touches, professionally spaced, is a system. Five is a reputation problem.

3. Track it like a funnel or you'll lie to yourself. If outreach lives in your DM history, you don't have a funnel — you have anecdotes. Every creator needs a row somewhere: contacted date, status, sample sent, posted, orders attributed. The statuses are the management tool: they tell you whether your problem is targeting (no replies), terms (replies, no sample requests), or follow-through (samples out, nothing posted) — three different problems with three different fixes.

4. Don't aim at the top. The top 1–3% of affiliates can see your shop's numbers, and a shop doing $2K a month is not worth their slot. The wins sellers report come from hungry Level 1–2 creators who need the opportunity as much as you do — the greeting-card brand that scaled on a single near-unknown affiliate is a genre of story, not an exception. Your first creators are also, frankly, an awareness channel: a new shop has no reviews and no social proof to convert with, so early content seeds credibility before it seeds sales.

The honest conclusion

Affiliate outreach on TikTok Shop is a numbers funnel with a 1–3% front door. That is not a reason to avoid it — creator content still carries the majority of the platform's GMV. It is a reason to run it like a pipeline: enough volume to matter, one structured follow-up, every attempt recorded, and conclusions drawn at 100 attempts, not 10.

I run my outreach on a template system — the first-touch DM, the follow-up, the sample terms, the brief, the reactivation — and the five starter prompts (the first DM plus four video scripts) are free here:

Free mini-pack (5 prompts): https://promptandskillco.gumroad.com/l/free-tiktok-shop-mini-pack

The full AI Toolkit for TikTok Shop Sellers has all 12 outreach messages plus 30 video scripts, 8 live support scripts, the outreach tracker spreadsheet with live formulas, and a pre-publish listing checklist. Personal $39 / Commercial $79, one-time: https://promptandskillco.gumroad.com/l/ai-toolkit-for-tiktok-shop-sellers

Disclosure: I sell this toolkit. The funnel figures above are the ranges sellers and agencies publicly report; your mileage will vary by niche, commission, and product.

TikTok Shop Creator Outreach: Why Your First DM Gets Ignored (and the 60-Word Affiliate DM Script I Use Instead)

Prompt & Skill Co. — Wed, 21 Oct 2026 07:09:48 +0000

On TikTok Shop in the US, creator affiliate content carries the majority of GMV — publicly reported industry figures put it above 60%, while live selling accounts for roughly 8%. (Those are the widely cited public estimates, not my numbers; check current reporting before you quote them anywhere.) Which means for most sellers, the highest-leverage message you will ever write is not a listing title. It is a 60-word DM to a creator who has never heard of you.

And most of those DMs get ignored. Not because creators are difficult. Because the average first-touch DM makes the creator do all the work: figure out the fit, guess the terms, and take all the risk. Here is what goes wrong, what creators are actually deciding, and the structure I now use instead.

Five ways a first DM dies

1. The product monologue. Three paragraphs about your product's features, materials, and story. The creator still has to translate all of it into the only question that matters to them: does this belong in front of my audience? Most won't bother.

2. No deal on the table. "Hey, want to collab?" with no sample offer and no commission percentage is a business proposal with the business part missing. Serious creators price their work. A DM with no terms reads as either amateur or a trap.

3. Asking for a free post. "We'll pay you from the sales you generate" — without even sending a sample — asks the creator to fund your marketing with their time, their audience's trust, and their content calendar. The answer is silence.

4. The obvious bulk paste. No name, no reference to anything they have ever made, sent to two hundred accounts. It signals exactly how the relationship will go if they say yes.

5. Earnings promises. "This is about to go viral," "easy money," "our last creator made $X." Even where it isn't flatly against the rules, it destroys credibility on contact. Creators have heard it from every seller who later vanished.

What a creator is actually deciding

A working creator reads your DM as a four-point check, usually in this order:

Fit. Does this product make sense in my content and my audience's life? You prove fit with one specific detail — a real video of theirs where the product would naturally live. Everything else is flattery.
Terms. Is there a free sample? What is the commission percentage? Stated plainly, up front, these two facts save both sides three rounds of messages — and their absence is the single most common reason a DM stalls.
Creative freedom. Talking points or a verbatim script? Approval process or post-and-go? Creators protect their voice because their voice is the asset you are renting. A first DM doesn't need the full brief, but it should not smell like a script they'll be forced to read.
Payout trust. Commission paid through the platform's affiliate plan, with tracked orders and tracked attribution, is credible. Commission paid "trust me, by transfer, later" is not. The system carries the trust, not your promises.

The structure: five slots, sixty words

This is the skeleton of the first-outreach template I use (it ships as prompt #31 in my toolkit). The word budget is the point — 60 words forces you to answer the four checks above and nothing else:

Hi [creator name] — your [specific video of theirs] is exactly where [product] fits.
[One factual line: what the product is + why it matches their content.]
Want a free sample? It's on our affiliate plan at [real commission]% if you ever post — zero obligation either way.

Walking through the slots:

The open is the specific video, not a compliment. "Your dorm reset video" beats "love your content!" because it is checkable. You watched something real, and the product's fit is visible in the same sentence.
The fit line is one line, and it is factual. What the product is and why it belongs in that creator's world. No adjectives doing unpaid work — if the fit needs three exclamation points, the fit isn't there.
The offer is stated plainly: sample + real commission. "Free sample" removes their risk. The actual percentage — the one you genuinely offer, on the actual affiliate plan — removes the negotiation fog.
The next step is deliberately small. Not "let's schedule a call," not a brief attachment. One yes/no question they can answer in four seconds.
"Zero obligation either way" is load-bearing. It tells the creator the sample is not a debt. Paradoxically, it is also what makes a reply more likely: nobody wants to start a conversation that feels like signing something.

Here is the skeleton filled in, with a fictional example (fictional creator, fictional product — the mini humidifier I use for all demos):

"Hi Maya — your dorm reset video is exactly where our PuraMist Mini lives. Want a free sample? It's on our affiliate plan at 20% if you ever post — zero obligation either way."

Thirty-five words. Notice what is not in it: no earnings claim, no fake discount, no "limited spots," no link. Links can come after a human replies; a cold DM carrying a link as its payload reads as spam even when it isn't.

The follow-up: once, usefully, then stop

Send one follow-up, roughly three to five days later. Make it carry something new — the answer to a question they might have, one concrete detail (shipping time, a color variant their audience asked about), not "just bumping this!" twice. Then stop. Silence after a first touch and one polite follow-up is an answer, and it costs you nothing to respect it. Creators remember the sellers who behaved professionally; when you launch a product that fits them better, you want your name to be welcome in that inbox. Burning the contact for one post is a bad trade.

The ground rules (these are not optional)

Disclosure is part of the deal. Creator posts are advertising: branded-content disclosure switched on, #ad in the caption, per FTC and platform rules. Say so in your outreach terms, not after the post goes live.
Everything runs through the platform. Affiliate links through TikTok Shop are where tracking, attribution, and buyer protection live. Never steer a creator — or their audience — off-platform to transact.
Only real numbers. Discounts, end times, stock levels, commission rates: if it isn't true and verifiable, it doesn't go in the DM or the video.
No GMV promises, in either direction. Not to creators ("you'll earn…"), and not to yourself. Outreach is a pipeline, not a lottery ticket. Track it like one: who you contacted, what you offered, who replied.

If you want the working version of this, I put the first-outreach DM template — plus four short-video script prompts (a 3-second hook generator, a pain-point demo, a mistake-alert script, and a rapid-fire multi-use format) — into a free mini-pack:

Free mini-pack (5 prompts): https://promptandskillco.gumroad.com/l/free-tiktok-shop-mini-pack

The full AI Toolkit for TikTok Shop Sellers adds the other 45 prompts (26 more video formats, 11 more creator messages — terms, briefs, negotiations, reactivations — and an 8-piece live-selling support set), a creator outreach tracker spreadsheet with live formulas, and a pre-publish listing checklist. Personal $39 / Commercial $79, one-time: https://promptandskillco.gumroad.com/l/ai-toolkit-for-tiktok-shop-sellers

Disclosure: I sell this toolkit. Every example in this post is fictional.

Your Review Mining Is Backwards: AI Should Explain, Not Count

Prompt & Skill Co. — Wed, 21 Oct 2026 05:37:51 +0000

Here's the thing nobody says out loud: when you paste 200 reviews into a chatbot and ask "what are my customers complaining about?", it doesn't count anything. It vibes.

It reads a few loud reviews, notices a theme that feels frequent, and hands you a ranking that sounds authoritative. Sometimes the ranking is even right — by accident. But you can't tell which times those are, and neither can the model.

I know because I did exactly this for months. My Seller Central had the usual graveyard: a few hundred reviews nobody has time to read, all of them potentially useful, almost none of them actually read. Pasting them into AI felt like mining. It was just scrolling with extra steps and better formatting.

The order is the whole trick

Counting is a solved problem. A spreadsheet counts perfectly, every time, for free. What a spreadsheet can't do is tell you why buyers are angry, or which complaints are actually your listing's fault.

Explaining is the opposite: it's the one thing a language model is genuinely good at — as long as you feed it a small, pre-sorted pile instead of the whole graveyard.

So the workflow that finally worked for me runs in the unglamorous order:

1. Log the reviews once, tag them by hand.
Date, stars, text, and one column I fill in myself: a pain-point tag ("lid hard to open," "paint chips," "too big for cup holder"). The tag list is the only real design decision in the whole system. Too broad and everything is soup; too narrow and everything is noise.

2. Let formulas rank.
COUNTIF over the tag column. AVERAGEIF for stars per tag. The 1–2★ share per tag. Now I have a ranked list I can defend: the tag with 30 mentions at 2.1★ is a problem. The tag with 3 mentions is one grumpy afternoon.

3. Only then, AI — on the top 2–3 tags only.
Paste just those reviews in and ask for the root cause, split three ways: (a) product issue, (b) listing/expectation issue, (c) usage issue — with quotes as evidence. Category (b) is where the money hides: a good share of "product is broken" reviews are really "the listing made me expect something else," and those you can fix tonight without touching the product.

A tiny demo run, on fictional data (an 18-review bottle set I use for testing): "lid hard to open" and "paint chips" tie at the top with 2 mentions each at 2.0★, while the praise tags cluster at 4.5–5.0★. Two minutes of counting told me more than an afternoon of scrolling ever did — and the AI's job shrank to explaining four reviews instead of hallucinating over eighteen.

What I stopped doing

Asking a chatbot to summarize more reviews than it can actually hold in view
Treating the loudest review as the most common one
Letting the model anywhere near writing, editing, or soliciting reviews — it reads existing reviews, full stop

Formulas rank. AI explains. In that order, or not at all.

I turned my setup into a small toolkit (worksheet with the formulas built in, plus the prompt set) — it's linked on my profile if you want the files instead of building your own, and there's a free 5-prompt mini pack there too.

But I'm more interested in how the rest of you do this: do you count first, or paste into ChatGPT first? And if you count — what's on your tag list? Mine's still evolving, and tag taxonomy turns out to be the whole game.

Counting, Not Scrolling: A Formula-First Workflow for Amazon Review Mining

Prompt & Skill Co. — Tue, 20 Oct 2026 12:09:28 +0000

The graveyard

Every Amazon seller I know has the same two files sitting somewhere: an export of a few hundred reviews nobody has time to read, and a Search Term Report that got downloaded, glanced at, and abandoned.

There are SaaS tools that will happily charge $99+/month to fix this. What most of them do, underneath the dashboard, is show you your own data back with charts. The data was never the problem. The counting is the problem — and counting is free.

So here's the workflow I run instead: a spreadsheet that counts, and one AI prompt that explains. In that order.

Counting first, AI second

Review mining is a counting problem before it's an AI problem.

Paste 200 reviews into a chatbot and ask "what do buyers think?", and you get a summary that sounds plausible and ranks nothing. It will spend as many words on the one-off complaint as on the problem mentioned forty times, because it has no way to know which is which.

Formulas don't have that failure mode. A COUNTIF doesn't care which review was written most passionately. So the order of operations is: formulas rank, then AI explains the top of the list. Everything below follows from that.

The worksheet: one log, one formula layer

The log is boring on purpose — one row per review:

Date
Star rating
Review text
Pain-point tag — one short label you assign: lid hard to open, sizing, packaging, instructions
Sentiment — positive / mixed / negative, which I assign while I'm tagging anyway; it takes two seconds and catches the sarcastic 5-star review a star count never will

On top of that sits a summary block of plain formulas:

COUNTIF over the tag column → mentions per tag
AVERAGEIF → average stars per tag, which separates "genuinely unhappy buyers" from "happy buyers with one gripe"
COUNTIFS → the 1–2★ share per tag
COUNTIF over the sentiment column → set totals

That's the whole machine. The one real design decision is the tag taxonomy. Tag the failing thing or the broken expectation, not the emotion. Too broad — quality, bad — and everything collapses into mush. Too narrow — lid hard to open with wet hands — and every tag has exactly one mention and nothing ranks. A good tag is the size of a fix: something you could act on in the product, the listing, or the instructions.

The prompt layer: where AI actually earns its place

Once the sheet says which two or three tags matter, I paste the review text for those tags into this prompt. It's the exact one I use — brackets are the parts you replace:

You are helping me analyze customer reviews for my own Amazon product. I will paste real review text below. Work ONLY from the text I provide — do not invent product features, specifications, or buyer complaints that are not in the reviews.

Product: [product name and type]
Reviews (all for the pain-point tag "[tag]"):
[paste review text]

Return:
1. The root cause behind these complaints, in one or two sentences — what specifically is failing or being misunderstood?
2. Split the cause: (a) product issue, (b) listing/expectation issue (the listing promised or implied something the product doesn't do), or (c) usage issue (buyers using it in a way the instructions don't cover). Quote the review phrases that support your split.
3. For (b) and (c) only: draft the exact listing or instruction change that would prevent this complaint — a rewritten bullet, a clarified size/spec line, or an added usage note. Use only facts from the reviews and my product details above; add no new claims.
4. One thing I should NOT change, based on what happy reviewers in this set said they liked.

The split in step 2 is the entire point. Only one of the three causes needs a factory, a re-order, or a tooling change. Expectation issues get fixed tonight, in the listing, for free. And the "work ONLY from the text I provide" line is not decoration — without it, the model will invent specifications your product doesn't have and complaints nobody wrote.

One full run, on fictional data

To show the loop end to end, I'll use the demo set that ships with my toolkit: 18 fictional reviews for a fictional 32 oz insulated bottle. Every number below comes from that set.

The summary block ranks the tags like this:

Tag	Mentions	Avg. stars
Keeps drinks cold (praise)	3	5.0
Lid hard to open	2	2.0
Paint / finish chips	2	2.0
Too big for cup holder	2	3.0
Ice cubes don't fit	2	4.0
One-hand open (praise)	2	5.0
Easy to clean (praise)	2	4.5

Plus three single-mention tags. Set totals: 3.4★ average, 7 positive / 3 mixed / 8 negative.

Reading those reviews one by one, I'd have remembered the 1★ "leaks in my bag" — it's vivid. The count says it's a single mention, while four separate tags tie at two. And the average-stars column separates those four instantly: two are real 2.0★ problems, one is a 4.0★ annoyance, and one — cup holder, at 3.0★ — smells like an expectation issue, because both reviewers praise the bottle while complaining: "Quality is good, but…" and "great bottle, wrong diameter."

Running the prompt on that tag's two reviews, the split comes back (b) listing/expectation, quoting "which was the whole point for my commute" and "wrong diameter for a standard car cup holder." The drafted fix is a single bullet:

Size check before you buy: this is the wide-body 32 oz. It does not fit standard car cup holders. It's built for desks and gym bags — if your commute is cup-holder-or-bust, this is the wrong bottle.

And step 4 earns its keep: the prompt warns not to shrink the bottle chasing this complaint, because the 5★ reviews are praising exactly the capacity and the insulation — "still cold at the end of my 10-hour shift," ice still inside after 12 hours in a hot car. The fix is a sentence in the listing, not a new product.

Small numbers, fictional bottle — but that's the loop at any scale. The counts tell you where to look. The prompt tells you what kind of problem it is. Most of the time, the fix is copy, not inventory.

The other two stations

The same count-first logic runs the rest of my monthly pass:

Competitor teardown. A rival's listing is public data you're allowed to analyze. I use an 8-section fill-in template — title structure, bullet angles, pricing log, their review weaknesses, image map, keyword coverage, differentiation verdict, change log — because a template gets finished and a blank page doesn't. Their 1-star clusters, counted the same way, are your differentiation angles in the buyers' own words.

PPC search-term triage. The Search Term Report is the review graveyard's twin: downloaded, never counted. Count spend per term, count conversions, and the report splits itself into waste to negate and winners to harvest into exact match. Same principle as the reviews — the data is already in your account. It just hasn't been counted.

Guardrails, stated plainly

Three lines I don't cross, and the prompts are built to hold them:

Everything runs on data you paste in. The prompts instruct the model to work only from that text — no invented specs, no invented complaints.
Nothing in this workflow writes, edits, or solicits reviews. It only reads reviews that already exist. That line is deliberate — it's the one part of "AI for Amazon" that can end a business.
No outcome is promised. This produces analysis and drafts. Whether a bullet rewrite helps is your call to make and your data to judge.

The file

I packaged my version of this loop as a one-time toolkit: the review-mining worksheet (.xlsx + .csv, formulas prebuilt — the 28 of them that produce the summary block above), the 8-section competitor teardown template, and 50 prompts in five groups (listing, review mining, competitor teardown, PPC, buyer-message replies), each with a worked fictional example. A free 5-prompt mini-pack is included, so you can test the format before paying anything.

Personal is $39, Commercial (for VAs and agencies, up to 5 seats) is $79. One-time, no subscription: https://promptandskillco.gumroad.com/l/spjfod

5 Chinese Short-Drama Lines That Break in Literal Translation (and How I'd Fix Them)

Prompt & Skill Co. — Tue, 20 Oct 2026 12:03:33 +0000

I build small Claude Skills for narrow, repetitive jobs. One of mine is not a translator. It is a first-pass localization checker for Chinese short-drama lines: paste in the Chinese, and it flags literal translations, rewrites the line in neutral Latin American Spanish and Brazilian Portuguese, explains every change in Chinese, and calls out puns or culture-specific jokes that need a workaround.

That distinction matters. Most of the lines I tested did not fail because the grammar was wrong. They failed because a perfectly literal rendering produced the wrong image, the wrong tone, or the wrong relationship between the speakers.

Here are five examples from the fictional demo set I use to test the checker.

1. "Playing dumb" becomes garlic

Chinese:

你少在这儿装蒜了，你那点小心思，我一眼就看穿了。

装蒜 (zhuāng suàn) means "playing dumb," but 蒜 is garlic. Translate the image literally and Spanish viewers get hacerse el ajo — roughly, "to make yourself the garlic."

A dubbing editor would drop the vegetable and keep the function:

Spanish: No te hagas el tonto conmigo. Sé perfectamente lo que tramas.
Brazilian Portuguese: Não se faça de bobo comigo. Eu sei muito bem o que você tá tramando.

Both use an existing local expression for playing dumb. Neither asks the audience to think about produce in the middle of an argument.

2. A sarcastic apology becomes a sincere one

Chinese:

行行行，你说的都对，我错了还不行吗？

The speaker is not really apologizing. This is the eye-roll version of "Fine, fine, you're right, I'm wrong — happy?"

A literal translation keeps the admission and loses the sarcasm:

Spanish: Ya, ya, ya… tú siempre tienes la razón. ¿Contento?
Brazilian Portuguese: Tá, tá, tá… você tá sempre certo. Satisfeito?

The repeated ya and tá do the work that the repeated 行 does in Chinese. Collapse them into one polite "okay," and the performance changes completely.

I put the complete worked checks for these five lines — original, problem flags, Spanish rewrite, Brazilian Portuguese rewrite, and the reasoning in Chinese — into a free sample:

Free sample — 5 checked short-drama lines: https://promptandskillco.gumroad.com/l/free-localization-sample

Try those first. If the format does not fit how you review scripts, the full version will not magically fit either.

3. A term of closeness turns into a family claim

Chinese:

姐妹，听我的，男人不能惯，越惯越混蛋。

姐妹 can be a warm way to address a close friend. In Spanish or Portuguese, reaching automatically for the equivalent of "sister" can make the relationship sound literal, depending on the region and context.

The safer functional choice in the demo is amiga:

Spanish: Amiga, hazme caso: a los hombres no hay que consentirlos, porque se ponen peor.
Brazilian Portuguese: Amiga, me escuta: homem não se mima, quanto mais mima, pior fica.

The check is not "never use a kinship word." It is "notice when a Chinese address term is doing social work that the target-language word may not do in the same way."

4. "Landlady-auntie" creates an accidental relative

Chinese:

房东阿姨，房租能不能宽限几天？我这个月手头有点紧。

阿姨 is a polite form of address here, not a statement that the landlady is the speaker's aunt. Spanish tía or Portuguese tia can make it sound like an actual family relationship.

The demo uses a respectful non-family form instead:

Spanish: Señora, ¿me podría dar unos días más para el alquiler? Este mes estoy un poco corta de dinero.
Brazilian Portuguese: Dona Marta, a senhora me dá mais uns dias pro aluguel? Esse mês eu tô meio apertada.

This line contains a second trap: 手头紧 literally mentions a "tight hand," but it means money is tight. Both target versions use an ordinary money expression instead of translating the body image.

5. An oath loses its oath

Chinese:

林晓，我对你是认真的，天地良心。

天地良心 is a Chinese oath formula. Rendering heaven, earth, and conscience word for word does not recreate the force of the promise. It mainly creates a phrase nobody would say during a confession scene.

The functional rewrites are:

Spanish: Lin Xiao, lo mío contigo va en serio. Te lo juro.
Brazilian Portuguese: Lin Xiao, o que eu sinto por você é sério. Juro por Deus.

The image changes. The speech act — "I am serious, and I swear it" — stays.

Why I packaged this as a checker

A one-off prompt can occasionally catch one of these problems. The harder part is checking sixty lines the same way: flag the literal image, preserve the tone, choose a natural spoken rewrite, and explain the change so the original writer can judge it.

That is why the full version is a Skill with a fixed workflow and output format rather than a loose instruction such as "translate this naturally." It is also why I keep the limit visible: this is a first-pass check, not a replacement for a native reviewer before broadcast.

If you want to inspect the format before paying for anything, start with the free five-line sample above.

If the workflow does fit your production process, the complete Short-Video Line Localization Checker — the Skill, its fixed output format, and the fully worked demo — is $19, one-time:

https://promptandskillco.gumroad.com/l/pidqzx

Which kind of line breaks most often in your work: idiom images, tone, or forms of address?

I Built a Claude Skill That Proofreads Short-Drama Lines for Spanish & Brazilian Portuguese

Prompt & Skill Co. — Tue, 20 Oct 2026 06:32:11 +0000

Chinese short dramas live and die by their lines — fast, slangy, full of idioms that carry the joke or the insult. And those are exactly the lines that break in translation.

Take this one (from a fictional demo script I wrote for testing):

林晓：你少在这儿装蒜了，你那点小心思，我一眼就看穿了。
"Stop playing dumb — I can see right through your little scheme."

装蒜 (zhuāng suàn) literally involves 蒜, garlic. A word-for-word Spanish rendering gives you hacerse el ajo — "to make yourself the garlic" — which sends viewers picturing produce instead of following the argument.

A native-speaking dubbing editor would never translate the image. They'd translate the function: "No te hagas el tonto conmigo" in Spanish, "Não se faça de bobo comigo" in Brazilian Portuguese — both are the local fixed expressions for "playing dumb." Same punch, no garlic.

That gap — between what a translator (human or AI) produces on the first pass and what an editor would flag — is what I wanted to automate. Not the translation itself. The check.

The three ways lines actually break

Working through test lines, the failures clustered into three buckets:

Idiom images. 手头紧 ("tight on money") literally says "my hand is tight." Spanish just says estar corto de dinero. The image must be dropped, not carried over.
Tone carried by form, not words. 行行行，你说的都对，我错了还不行吗？ is a sarcastic non-apology. Its literal translation reads sincere. The fix in Spanish is rhythmic: Ya, ya, ya… tú siempre tienes la razón. ¿Contento? — the triple repetition does the eye-roll.
Address terms. Chinese uses kinship words as polite address （阿姨 for a landlady). Tía in Spanish means your actual aunt. The functional equivalents are señora (ES) and dona + name (PT-BR).

A single "translate this naturally" prompt occasionally lands these. Occasionally is the problem — with 60 lines an episode, you need a fixed procedure that flags every line the same way.

Why a Skill file

I packaged the check as a Claude Skill (SKILL.md) instead of a saved prompt, for one reason: a Skill carries a workflow and an output contract, not just an instruction. The structure:

Trigger conditions — paste Chinese lines (or an existing ES/PT draft to be diagnosed), get the check; it doesn't fire on general translation chat.
A 5-step pass per line — read for intent → flag literal-translation stiffness → rewrite in neutral Latin American Spanish → rewrite in Brazilian Portuguese → handle puns/culture items explicitly.
A fixed output format — original line → problem flags → ES rewrite + note → PT-BR rewrite + note → pun/culture workaround when needed → batch summary. The notes are in Chinese, so the creator can judge the edit without reading the target language.
A fully worked example inside the file — so the model anchors on a real input→output pair, not just rules.

One rule I wrote in bold, in my head at least: flag uncertain expressions for native review instead of guessing. The skill is a first pass. Anything broadcast still gets human native eyes at the end — the file says so, and so does the product page.

What a run looks like

For the sarcasm line above, the output flags that a literal rendering "只剩字面认错，反讽语气全丢" (keeps the literal apology, loses all the sarcasm), then gives:

ES: Ya, ya, ya… tú siempre tienes la razón. ¿Contento?
PT-BR: Tá, tá, tá… você tá sempre certo. Satisfeito?
Note: the "行行行" effect is a repetition effect — the target language must reproduce the rhythm with its own repeated filler (ya / tá), not collapse it into a single "okay."

Six demo lines (all fictional — slang, address terms, an oath, sarcasm) are included in the pack so you can see the full format before deciding if the workflow fits yours.

The file

The skill, the worked demo, and the fixed output format are here for $19, one-time:

https://promptandskillco.gumroad.com/l/pidqzx

If you localize short-form video into Spanish or Portuguese, I'd genuinely like to know which line type breaks most often for you — idioms, tone, or address terms. That's the category the next version of the checker will grow into.