DEV Community: Kozmo

Why Your Zapier Workflow Broke (And the 5-Minute Diagnosis That Finds It Every Time)

Kozmo — Sun, 19 Apr 2026 18:18:16 +0000

Why Your Zapier Workflow Broke (And the 5-Minute Diagnosis That Finds It Every Time)

Your Zap was working fine for six months. Then — nothing. No errors, no alerts, just silence.

You open Zapier, run a test, it passes. You check the logs: green. You reload the page like that will help. It does not help.

Here is what actually happened, and exactly how to find it.

The Four Ways Zaps Break Silently

1. OAuth Token Expiry

The most common culprit. Zapier authenticates to apps using OAuth tokens that expire — and when they do, Zapier does not always surface this as a clean error. Instead, your Zap continues firing, the auth step silently fails, and the rest of the workflow either runs on stale data or does nothing at all.

Diagnostic: Go to My Apps in Zapier. Look for any connection with a yellow warning triangle or last-tested timestamp older than 30 days. Reconnect it, then go back into every Zap that uses that app, reselect the account in each step (not just the trigger — each individual step), and republish.

Note: reconnecting the app in the top-level My Apps view does not automatically update references inside existing Zaps. You must reselect inside the Zap.

2. Source App Changed Its API Schema

Every time a source app updates its API — which happens quietly, in changelogs nobody reads — your field mappings can break. Zapier mapped contact.email and now the app sends contact.primary_email. Your Zap still fires. The email field just arrives empty.

Diagnostic: Pull your Zap history and look at recent failed or "halted" tasks. Expand the payload. Find the step where something is blank that should not be blank. Compare that field name against the app's current API documentation or support docs. Usually there is a changelog entry that explains the rename.

3. Trigger Polling vs. Webhook — The Silent Gap

Polling-based triggers (Zapier checks every 5-15 minutes) have a window problem: events that happen and resolve between checks never get caught. Webhooks fire immediately on every event. If your trigger type is polling and you are missing events, this is likely why.

Fix: Check whether the app supports a webhook-based trigger instead of polling. If it does, switch. You will stop missing events.

4. Multi-Step Chain — The Cascade Failure

In a 5-step Zap, if step 3 errors out, steps 4 and 5 do not run. But unless you have error notifications enabled, Zapier just marks the task as "error" in the history and moves on. You never hear about it.

Fix: Turn on Zap error notifications (Settings → Notifications). Set them to immediate, not the daily digest. Also add a Zapier step at the end of any critical workflow that logs a record to a Google Sheet or sends you a Slack message — this is your "it ran successfully" receipt.

The 5-Minute Triage Protocol

When a Zap breaks and you do not know why, run this in order:

Zap History — look at the last 50 runs. Find the first one that failed. Read the error.
Connected Account — test the connection for every app in the Zap. Reconnect any that fail.
Payload inspection — expand the output of each step in a recent failed run. Find where null or empty data starts appearing. That is your break point.
Live test with real data — do not use Zapier's cached test data. Trigger the Zap with a real event and watch the run in real time.
Check the source app — if everything above looks fine, the issue is often on the other side. The source app stopped sending events, or changed how it formats them.

When It Is Not Worth Debugging Yourself

Some workflow breaks are worth fixing in 20 minutes. Others take three hours because the root cause is buried in an API response format mismatch that requires reading documentation for two different tools simultaneously while cross-referencing a payload that is only visible in Zap history under a specific error code.

If you are a business owner and your time is worth more than $50/hour, the math on DIY debugging often does not work.

The Broken Workflow Clinic fixes broken Zapier, Make, and n8n automations for a flat fee — $150 to $2,500 depending on complexity, 48-hour turnaround, scoped upfront so you know the price before any work starts. There is a free async audit first: send a Loom walkthrough of the broken workflow and get a diagnosis back within 24 hours.

If you have a broken automation that is costing you time or money, start with the free audit here — or drop a comment below describing what broke and I will tell you exactly what is wrong.

Useful if this saved you time? Share it with the person in your company who owns the Zapier account.

I Built a Shopify-Specific AI Prompt Kit (Not Another Generic "E-Commerce" Pack)

Kozmo — Sun, 19 Apr 2026 08:03:57 +0000

Most AI prompt packs for e-commerce are useless for Shopify specifically.

They say "e-commerce" but they're really built for Amazon or some vague DTC abstraction. The prompts reference "email marketing" instead of Klaviyo flows. "Ad copy" instead of Meta Ads + Shopify Catalog. "Product descriptions" instead of Shopify PDPs with above-the-fold bullets.

I spent a few months fixing that.

What actually makes a Shopify prompt different

Shopify has a specific stack: Shopify product pages, Klaviyo for email, Meta Ads with Shopify Catalog sync, Shopify Analytics dashboards, and a specific set of apps (Judge.me, Smile.io, ReCharge, ReConvert). When your prompts reference the actual tools, the AI output is dramatically more useful.

Compare:

Generic prompt:

"Write an abandoned cart email for my store."

Shopify-specific prompt:

"You're a DTC email copywriter who specializes in Klaviyo abandoned cart recovery for Shopify brands. You write emails that feel human, not automated. I sell [product] to [customer]. Our most common abandon reason is [objection]. Write Email 1 of a 3-part Klaviyo flow: soft reminder, 1 hour post-abandon, under 120 words, no pressure, no discount."

Same AI. Completely different output — because the second prompt references Klaviyo flows, specifies the email's position in a 3-part sequence, and gives the AI a role that understands Shopify's email ecosystem.

The 5 categories in the kit

1. Product Descriptions & Shopify Listings (Prompts 1–10)
Not just "write a product description." These prompts are structured for Shopify PDPs: above-the-fold bullet points, Shopify SEO titles, FAQs formatted for Shopify themes, comparison pages.

2. Klaviyo Email Marketing Flows (Prompts 11–20)
Welcome flows, 3-email abandoned cart sequences, post-purchase onboarding, win-back flows, VIP segment emails, launch sequences. Every prompt references Klaviyo send events and flow triggers — not just "email."

3. Customer Service & Retention (Prompts 21–30)
Complaint responses, Shopify return scripts, 1-star review responses (Judge.me format), Smile.io loyalty intro emails, ReCharge/Skio subscription cancellation deflection.

4. Meta Ads for Shopify (Prompts 31–40)
Cold audience Meta ads, Shopify retargeting ads (dynamic catalog), UGC creator briefs for TikTok/Reels, BFCM automatic discount copy, 30-second video scripts.

5. Shopify Analytics & Growth Strategy (Prompts 41–50)
Weekly performance reviews using Shopify Analytics + Klaviyo + Meta together, CAC calculator prompts, A/B test design, Klaviyo list growth tactics, LTV optimization, Shopify scale readiness checklist.

The bonus: Notion income tracker

For store owners who also run brand deals, affiliate income, or UGC contracts alongside their Shopify business — there's a 5-database Notion template bundled in:

Brand Deal CRM (pipeline from outreach to payment)
Income Tracker (all revenue by platform and type)
Invoice & Payment Log (auto-flags overdue)
Monthly P&L Dashboard
Content Calendar (linked to active deals)

What makes these prompts actually work

Three things separate a prompt that produces something you edit for 5 minutes vs. something you delete:

Role specificity — every prompt opens with a precise role: "DTC copywriter specializing in Shopify brands," not "copywriter." The more specific the role, the more specialized the output.
Stack references — Klaviyo instead of "email," Shopify PDP instead of "product page," Meta Ads + Shopify Catalog instead of "Facebook ads." The AI has trained on all of these tools — use the right vocabulary.
Output constraints — length limits, tone constraints, banned words. Freedom produces generic. Constraints produce craft.

AI Shopify Operations Kit — 50 Shopify-specific prompts + Notion income tracker.
PDF + editable Markdown. $29 on Gumroad: https://kozmoinc.gumroad.com/l/yklqoz

Works with Claude, ChatGPT, or Gemini.

The 4-Part Prompt Structure That Writes E-Commerce Ad Copy Worth Running

Kozmo — Sun, 19 Apr 2026 03:30:25 +0000

Most AI-generated ad copy is immediately recognizable as AI-generated ad copy.

It's not because the AI can't write. It's because the brief was bad.

"Write a Facebook ad for my e-commerce store selling handmade candles" is not a brief. It's a category. The AI fills in the blanks with the most generic possible patterns — because you didn't give it anything specific to work with.

Here's what changes when you give it the right context.

The 4-part context frame for ad copy

Before you write the prompt, you need four things:

1. Audience
Not "women who like candles." The specific customer. "Women 28–45, homeowners, follow accounts like @apartmenttherapy and @thesill, spend on home aesthetic but research before buying."

2. Platform + format
Facebook feed ad, Instagram Story, TikTok 30-second video script, Google Search headline — each has different length constraints, scroll behavior, and expectations. Don't let the AI guess.

3. Offer hook
What is the thing that makes this product worth stopping for? Not your brand story — the specific thing about this product that would make the right customer immediately interested. "Made from single-origin coconut wax, burns 60+ hours, no synthetic fragrance" is a hook. "Quality handmade candles" is not.

4. Desired action
What exactly do you want the reader to do? Click to product page, click to learn more, watch the video, visit the link in bio. One action. State it explicitly.

The prompt

Here's how all four go into the brief:

You are a direct-response copywriter who specializes in paid social for DTC e-commerce brands. You write ads that stop the scroll and move product.

Product: [NAME] — [one sentence description]
Audience: [SPECIFIC DEMOGRAPHIC + PSYCHOGRAPHIC — interests, behaviors, what they already buy]
Platform: [Facebook feed / Instagram Story / TikTok script / Google Search]
Offer hook: [THE SPECIFIC THING THAT MAKES THIS WORTH STOPPING FOR]
Desired action: [EXACTLY WHAT YOU WANT THEM TO DO — one click, one URL]

Write 3 ad variations:

Variation A: Lead with the problem the customer has before they buy this product
Variation B: Lead with the specific outcome they get after
Variation C: Lead with social proof or specificity (a number, a detail, a result)

Each variation: headline + 3-sentence body + CTA. No exclamation points. No "limited time offer." Write like a brand that doesn't need to beg.

Run that with your actual product details and you get three distinct creative angles, each leading from a different place. Your job is to pick the strongest, not to start from scratch.

Why three variations matter

Ad creative testing isn't about finding "the best ad." It's about finding the best angle for the best audience.

Problem-led creative tends to perform for cold audiences who don't know your brand. They need to feel seen before they'll click.

Outcome-led creative performs for warm audiences — people who've seen your brand before and are closer to a decision.

Social proof or specificity-led creative cuts through because it's concrete. "Scented with [X], not synthetic fragrance" creates instant trust with an informed buyer. A number like "burns 60+ hours" is more credible than "long-lasting."

Having all three ready lets you run a legitimate test instead of a guess.

The ad copy section in the full pack

The ad copy prompts (31–40) in the AI Prompt Pack for E-Commerce use this 4-part context frame across every paid channel:

Prompt 31: Cold audience Facebook/Instagram feed ad (the one above)
Prompt 32: Retargeting copy (for people who visited but didn't buy)
Prompt 33: TikTok/Reels 30-second video script
Prompt 34: Google Search headlines and descriptions
Prompt 35: UGC creator brief (gets creators to make content worth running as an ad)
Prompts 36–40: Black Friday/sale copy, seasonal hooks, product launch sequence

Every prompt in this section uses the same architecture: role assignment, context frame (audience + platform + offer hook + desired action), structured output, tone constraints.

The difference between prompts 31–40 in v1.1 of the pack and a generic "write me ad copy" prompt is the context frame. Fill in four fields and the AI has enough to produce a real draft instead of a placeholder.

The full pack

50 prompts. 5 categories. PDF + Markdown.

If you run paid social for an e-commerce brand and currently spend more than 15 minutes per ad brief, this is a faster path. Every ad copy prompt is ready to fill in and run. Each includes the prompt, an example output, and the one customization detail that makes it work for your specific product.

$19. One purchase.

Get it on Gumroad →

If you run the ad copy prompt above and get generic output, the issue is almost always the "offer hook" field — it's too vague. Drop your product in the comments and I'll show you how to sharpen it.

The 4-Part Prompt Structure That Writes Shopify Ad Copy Worth Running

Kozmo — Sun, 19 Apr 2026 03:24:26 +0000

Most AI-generated ad copy is immediately recognizable as AI-generated ad copy.

It's not because the AI can't write. It's because the brief was bad.

"Write a Facebook ad for my Shopify store selling handmade candles" is not a brief. It's a category. The AI fills in the blanks with the most generic possible patterns — because you didn't give it anything specific to work with.

Here's what changes when you give it the right context.

The 4-part context frame for ad copy

Before you write the prompt, you need four things:

4. Desired action
What exactly do you want the reader to do? Click to product page, click to learn more, watch the video, visit the link in bio. One action. State it explicitly.

The prompt

Here's how all four go into the brief:

You are a direct-response copywriter who specializes in paid social for DTC Shopify brands. You write ads that stop the scroll and move product.

Write 3 ad variations:

Variation A: Lead with the problem the customer has before they buy this product
Variation B: Lead with the specific outcome they get after
Variation C: Lead with social proof or specificity (a number, a detail, a result)

Each variation: headline + 3-sentence body + CTA. No exclamation points. No "limited time offer." Write like a brand that doesn't need to beg.

Run that with your actual product details and you get three distinct creative angles, each leading from a different place. Your job is to pick the strongest, not to start from scratch.

Why three variations matter

Ad creative testing isn't about finding "the best ad." It's about finding the best angle for the best audience.

Problem-led creative tends to perform for cold audiences who don't know your brand. They need to feel seen before they'll click.

Outcome-led creative performs for warm audiences — people who've seen your brand before and are closer to a decision.

Having all three ready lets you run a legitimate test instead of a guess.

The ad copy section in the full pack

The ad copy prompts (31–40) in the AI Shopify Operations Kit use this 4-part context frame across every paid channel:

Prompt 31: Cold audience Facebook/Instagram feed ad (the one above)
Prompt 32: Retargeting copy (for people who visited but didn't buy)
Prompt 33: TikTok/Reels 30-second video script
Prompt 34: Google Search headlines and descriptions
Prompt 35: UGC creator brief (gets creators to make content worth running as an ad)
Prompts 36–40: Black Friday/sale copy, seasonal hooks, product launch sequence

Every prompt in this section uses the same architecture: role assignment, context frame (audience + platform + offer hook + desired action), structured output, tone constraints.

The full pack

50 prompts. 5 categories. PDF + Markdown.

If you run paid social for an Shopify brand and currently spend more than 15 minutes per ad brief, this is a faster path. Every ad copy prompt is ready to fill in and run. Each includes the prompt, an example output, and the one customization detail that makes it work for your specific product.

$29. One purchase.

Get it on Gumroad →

Why Your AI-Generated Product Descriptions Don't Convert (And How to Fix the Prompt)

Kozmo — Sun, 19 Apr 2026 03:18:53 +0000

There's a pattern in Shopify that wastes more time than any other single workflow.

Founder opens ChatGPT. Types "write a product description for my [product]." Gets 150 words of filler. Spends 20 minutes editing out the adjectives. Publishes something mediocre. Wonders why their conversion rate is flat.

The problem is the prompt, not the AI.

What a product description is actually supposed to do

Before we get to the prompt: most product descriptions fail before the AI is even involved, because the seller doesn't have a clear answer to four questions:

Who is reading this? Not "everyone who buys candles." The specific customer — demographics, context, what they're trying to achieve or feel.
What is the lead benefit? Not a feature. The outcome the customer cares about.
What's the voice? Premium and minimal? Playful and irreverent? Warm and personal?
What's the constraint? Word count, what to avoid, what not to say.

If you don't know those four things before you open the AI, you're not going to get a description worth using. You're going to get a generic draft that matches your vague brief.

The prompt structure that produces usable copy

Here's the prompt I use for product descriptions. It's in the AI Shopify Operations Kit as Prompt #1, and it's the template I run on every new product listing:

You are a DTC copywriter for [BRAND CATEGORY — e.g., "premium home goods" or "performance athletic wear"]. You write product descriptions that lead with benefit, not feature. Your tone is [ADJECTIVE, ADJECTIVE — e.g., "clean and confident" or "warm and slightly irreverent"].

Product: [PRODUCT NAME]
Key details: [2-3 sentences about what it is, what it's made of, what makes it different]
Customer: [WHO IS BUYING THIS AND WHY — be specific: "Women 30–50 who are furnishing a home for the first time and care more about quality than trend"]
Primary benefit (lead with this): [THE OUTCOME, NOT THE FEATURE — e.g., "sleep that actually recovers you" not "100% merino wool"]

Write a product description: 120–160 words. Lead with the primary benefit in the first sentence. No vague adjectives (premium, luxurious, high-quality). No passive voice. End with a concrete use-case sentence that helps the customer picture themselves using it.

Run that with your actual product details and you'll have something worth editing. Not something worth deleting.

The three mistakes that produce bad product descriptions

1. Starting with features instead of benefits

"Made from 100% organic cotton with reinforced stitching and a relaxed fit."

Nobody buys a fabric. They buy comfort, confidence, or longevity. Lead with what the customer gets, not what you made.

2. Vague adjectives

"Premium quality. Luxurious feel. High-end craftsmanship."

These phrases mean nothing because every brand says them. If your product is genuinely good, say something specific. "Stitched to survive 200+ washes" is more credible than "premium quality" and takes the same number of words.

3. Writing to everyone

If your customer is "anyone who likes candles," your description will read like it was written for no one. Pick the most likely buyer and write for them. Everyone else who's a fit will recognize themselves in it. Everyone who isn't won't buy regardless of how broad your copy is.

One more thing the prompt does that matters

The prompt above assigns the AI a role: "DTC copywriter for [brand category]." This isn't decoration — it changes the output significantly.

When you tell the AI it's a copywriter for premium home goods, it activates a different set of writing patterns than if you just say "write a product description." The AI defaults to the specific vocabulary, tone conventions, and structure of that role.

This is the single highest-leverage thing you can add to any AI prompt: a specific expert role with a specific domain. "You are a [specific expert] for [specific type of company]" before any writing task.

The abandoned cart prompts, ad copy prompts, and customer service prompts in the pack all use this same structure.

The full pack

This prompt is one of 50 in the AI Shopify Operations Kit — built to cover every writing task a Shopify, Etsy, or Amazon seller handles week-to-week.

50 prompts. 5 categories. PDF + Markdown format.

Each prompt includes the prompt itself, an example output, and a customization tip — the thing that makes the prompt actually work for your brand versus any other.

$29. Use them forever.

Get it on Gumroad →

If the product description prompt above produces something generic even after filling in your details, leave a comment. The customization tip in the full pack (Prompt #1) usually fixes it — but I'm happy to debug here too.

The AI Prompt That Writes Abandoned Cart Emails That Don't Suck

Kozmo — Sun, 19 Apr 2026 03:18:52 +0000

Most abandoned cart emails are embarrassing.

"You left something behind!" with a sad little cart emoji. Maybe a 10% discount that trains your customers to abandon carts on purpose. A subject line that reads like it was written by a guilt-tripping aunt.

The problem isn't the email. The problem is that the person who wrote it didn't understand what an abandoned cart actually means.

An abandoned cart is not a forgotten cart. It's a considered cart. The customer looked at your product, added it, and then stopped. That stop has a reason. Your job is to address the reason — not just remind them the cart exists.

The 3-email sequence that recovers revenue

High-performing abandoned cart flows aren't a single email. They're three separate conversations:

Email 1 (1 hour after abandonment): Soft reminder
No pressure. No discount. Just a clean, useful reminder that the cart is waiting. This email exists because sometimes people get interrupted. Give them the benefit of the doubt.

Email 2 (24 hours after): Address the objection
This is the money email. You need to identify the most common reason people abandon your specific product and address it directly. Is it price? Shipping cost? Uncertainty about sizing or fit? Returns policy anxiety? Pick the right objection and write directly to it.

Email 3 (72 hours after): Final nudge
Small incentive if you're going to use one — and only here, never in Email 1. Urgency framing if your inventory or offer genuinely warrants it.

Three different conversations. Three different jobs. One generic email does none of them well.

The prompt that writes all three

Here's the exact prompt structure I use. Copy it, fill in your product details, and run it in Claude, ChatGPT, or Gemini:

You are a DTC email copywriter who specializes in abandoned cart recovery. You write emails that feel like they came from a real human, not a marketing automation tool.

I run an Shopify store selling [PRODUCT TYPE] to [TARGET CUSTOMER]. My brand voice is [ADJECTIVE, ADJECTIVE, ADJECTIVE — e.g., "direct, warm, slightly irreverent"].

The most common reason customers abandon our carts is [YOUR MAIN OBJECTION — e.g., "shipping cost uncertainty" or "unsure if the sizing will work for them"].

Write a 3-email abandoned cart sequence:

Email 1 (1 hour post-abandonment): Soft, no-pressure reminder. Subject line + body under 120 words.
Email 2 (24 hours): Directly address [MAIN OBJECTION]. Subject line + body under 180 words.
Email 3 (72 hours): Final nudge. Include [small incentive or urgency framing]. Subject line + body under 150 words.

Tone rules: No exclamation points. No "You left something behind!" No guilt. Write like a confident brand that knows the product is good.

Run that once with your real product details and you'll have a draft sequence worth editing. Not a template worth deleting.

Why this works when generic prompts don't

Most AI prompt packs tell you to "write an abandoned cart email for [brand]." That's a brief, not a prompt. The AI has no idea:

What objection to address
What tone your brand uses
How long each email should be
What role it's playing in the sequence

The prompt above gives the AI a job (DTC email copywriter), a context (your product, your customer, your objection), a structure (3-email sequence with specific timing and word counts), and a constraint (tone rules). That's the difference between a generic draft you spend 30 minutes fixing and a draft you spend 5 minutes editing.

The 49 other prompts

This is one of 50 prompts in the AI Shopify Operations Kit — a collection I built to cover every writing task a Shopify, Etsy, or Amazon seller actually faces.

The pack covers:

Product descriptions and listings (10 prompts)
Email marketing sequences (10 prompts, including the full abandoned cart flow above)
Customer service responses (10 prompts)
Ad copy and creative briefs (10 prompts)
Analytics and growth strategy (10 prompts)

Every prompt includes the prompt itself, an example output, and a customization tip. PDF + Markdown format — paste directly into Notion or Obsidian.

$29. One purchase. Use them forever.

Get the full pack on Gumroad →

If you try the abandoned cart prompt above and it doesn't produce something usable, leave a comment with your product details and I'll debug the prompt with you.

The AI Prompt That Writes Abandoned Cart Emails That Don't Suck

Kozmo — Sun, 19 Apr 2026 03:13:36 +0000

Most abandoned cart emails are embarrassing.

The problem isn't the email. The problem is that the person who wrote it didn't understand what an abandoned cart actually means.

The 3-email sequence that recovers revenue

High-performing abandoned cart flows aren't a single email. They're three separate conversations:

Email 3 (72 hours after): Final nudge
Small incentive if you're going to use one — and only here, never in Email 1. Urgency framing if your inventory or offer genuinely warrants it.

Three different conversations. Three different jobs. One generic email does none of them well.

The prompt that writes all three

Here's the exact prompt structure I use. Copy it, fill in your product details, and run it in Claude, ChatGPT, or Gemini:

You are a DTC email copywriter who specializes in abandoned cart recovery. You write emails that feel like they came from a real human, not a marketing automation tool.

I run an e-commerce store selling [PRODUCT TYPE] to [TARGET CUSTOMER]. My brand voice is [ADJECTIVE, ADJECTIVE, ADJECTIVE — e.g., "direct, warm, slightly irreverent"].

The most common reason customers abandon our carts is [YOUR MAIN OBJECTION — e.g., "shipping cost uncertainty" or "unsure if the sizing will work for them"].

Write a 3-email abandoned cart sequence:

Email 1 (1 hour post-abandonment): Soft, no-pressure reminder. Subject line + body under 120 words.
Email 2 (24 hours): Directly address [MAIN OBJECTION]. Subject line + body under 180 words.
Email 3 (72 hours): Final nudge. Include [small incentive or urgency framing]. Subject line + body under 150 words.

Tone rules: No exclamation points. No "You left something behind!" No guilt. Write like a confident brand that knows the product is good.

Run that once with your real product details and you'll have a draft sequence worth editing. Not a template worth deleting.

Why this works when generic prompts don't

Most AI prompt packs tell you to "write an abandoned cart email for [brand]." That's a brief, not a prompt. The AI has no idea:

What objection to address
What tone your brand uses
How long each email should be
What role it's playing in the sequence

The 49 other prompts

This is one of 50 prompts in the AI Prompt Pack for E-Commerce — a collection I built to cover every writing task a Shopify, Etsy, or Amazon seller actually faces.

The pack covers:

Product descriptions and listings (10 prompts)
Email marketing sequences (10 prompts, including the full abandoned cart flow above)
Customer service responses (10 prompts)
Ad copy and creative briefs (10 prompts)
Analytics and growth strategy (10 prompts)

Every prompt includes the prompt itself, an example output, and a customization tip. PDF + Markdown format — paste directly into Notion or Obsidian.

$19. One purchase. Use them forever.

Get the full pack on Gumroad →

If you try the abandoned cart prompt above and it doesn't produce something usable, leave a comment with your product details and I'll debug the prompt with you.

Why Your AI-Generated Product Descriptions Don't Convert (And How to Fix the Prompt)

Kozmo — Sun, 19 Apr 2026 03:13:09 +0000

There's a pattern in e-commerce that wastes more time than any other single workflow.

The problem is the prompt, not the AI.

What a product description is actually supposed to do

Before we get to the prompt: most product descriptions fail before the AI is even involved, because the seller doesn't have a clear answer to four questions:

Who is reading this? Not "everyone who buys candles." The specific customer — demographics, context, what they're trying to achieve or feel.
What is the lead benefit? Not a feature. The outcome the customer cares about.
What's the voice? Premium and minimal? Playful and irreverent? Warm and personal?
What's the constraint? Word count, what to avoid, what not to say.

If you don't know those four things before you open the AI, you're not going to get a description worth using. You're going to get a generic draft that matches your vague brief.

The prompt structure that produces usable copy

Here's the prompt I use for product descriptions. It's in the AI Prompt Pack for E-Commerce as Prompt #1, and it's the template I run on every new product listing:

Run that with your actual product details and you'll have something worth editing. Not something worth deleting.

The three mistakes that produce bad product descriptions

1. Starting with features instead of benefits

"Made from 100% organic cotton with reinforced stitching and a relaxed fit."

Nobody buys a fabric. They buy comfort, confidence, or longevity. Lead with what the customer gets, not what you made.

2. Vague adjectives

"Premium quality. Luxurious feel. High-end craftsmanship."

3. Writing to everyone

One more thing the prompt does that matters

The prompt above assigns the AI a role: "DTC copywriter for [brand category]." This isn't decoration — it changes the output significantly.

The abandoned cart prompts, ad copy prompts, and customer service prompts in the pack all use this same structure.

The full pack

This prompt is one of 50 in the AI Prompt Pack for E-Commerce — built to cover every writing task a Shopify, Etsy, or Amazon seller handles week-to-week.

50 prompts. 5 categories. PDF + Markdown format.

Each prompt includes the prompt itself, an example output, and a customization tip — the thing that makes the prompt actually work for your brand versus any other.

$19. Use them forever.

Get it on Gumroad →

Test

Kozmo — Sun, 19 Apr 2026 03:12:44 +0000

Hi

test