The short answer: ChatGPT is now the most-deployed AI tool in e-commerce, but most prompts produce copy that sounds good and converts nothing. The 10 prompts below are the ones I keep coming back to because they force the model to write for a specific funnel stage, customer action, and measurable outcome. Every one was tested against 2026 benchmarks from Baymard, Omnisend, Shopify, and the HubSpot State of Marketing 2026 report.
I’ve written thousands of ChatGPT prompts for online stores over the last year. Some worked. Most were forgettable. The difference is rarely the model. It’s the structure: who you’re writing for, what action you want them to take, what proof you can show, and what you’ll measure after.
According to the HubSpot State of Marketing 2026, 80% of marketers now use AI for content creation and 75% use it for media production. Salesforce’s State of Commerce report surveyed 2,700 commerce leaders and ranked AI as their top priority for the year. The gap between teams shipping usable copy and teams drowning in mediocre AI output is prompt design.
This article gives you the prompts that actually move numbers. I’ll show you the funnel stage each one targets, a before-and-after output sample, an A/B variant to test, the engineering tip behind why the prompt works, and the 2026 benchmark you’ll be measured against.
“AI can draft, it cannot verify. Every claim that matters to a buyer still needs evidence before you publish.” FTC guidance on AI claims, 2024-2026 enforcement record
The 2026 E-commerce Reality (What You’re Optimizing Against)
Before any prompt, you need to know the battlefield. Here are the numbers that shape every e-commerce conversion decision in 2026, sourced from primary research.
Conversion and traffic benchmarks (2025-2026):
- Global ecommerce conversion rate: 1.6% of visits converted into purchases in Q3 2025 (Statista, cited in Shopify, Feb 4, 2026).
- Mobile drives roughly 78% of retail site visits worldwide and generates about 70% of all online orders (Statista, Q3 2025).
- Industry conversion spread (Dynamic Yield via Shopify, 2025): Food and beverage 6.22%, Beauty and personal care 4.94%, Multi-brand retail 3.93%, Pet care 3.28%, Fashion 3.06%, Consumer goods 2.85%, Home and furniture 1.41%, Luxury and jewelry 0.94%.
- Cart abandonment: 70.19% global average across 50 studies tracked by Baymard Institute, last updated September 22, 2025.
Cart abandonment reasons (Baymard quantitative study, US online shoppers):
- 43% were “just browsing / not ready to buy”
- 39% said extra costs too high (shipping, tax, fees)
- 21% said delivery was too slow
- 19% didn’t trust the site with their credit card information
- 19% said the site wanted them to create an account
- 18% said the checkout was too long or complicated
- 15% said the returns policy wasn’t satisfactory
- 14% couldn’t see or calculate total order cost up-front
- 10% said there weren’t enough payment methods
- 8% said their credit card was declined
Checkout UX potential: Baymard’s combined usability testing shows the average large e-commerce site can gain a 35.26% increase in conversion rate through better checkout UX alone. With combined US and EU ecommerce sales of $738 billion, that translates to $260 billion in recoverable orders. The average checkout flow contains 23.48 form elements displayed by default, while an ideal flow needs only 12-14 elements (7-8 fields).
Email automation benchmarks (Omnisend, 2025 dataset, 20B+ campaign emails across 27,000+ brands):
- Campaign emails: 30.41% open, 0.74% CTR, 0.08% conversion, $0.155 revenue per email
- Automated emails: 30.21% open, 4.66% CTR, 1.49% conversion, $3.41 revenue per email (a 22x lift)
- Abandoned cart automation: 37.12% open, 4.13% CTR, 1.72% conversion, $3.59 per email
- Welcome series: 35.53% open, 2.11% conversion, $6.16 per email
- Back-in-stock alert: 58.80% open, 21.31% CTR, 6.72% conversion, $9.14 per email
- Order follow-up unsubscribe rate: 0.86%; Cross-sell unsubscribe rate: 0.89%
AI adoption reality (HubSpot State of Marketing 2026):
- 61% of marketers say marketing is experiencing its biggest disruption in 20 years due to AI
- 80% use AI for content creation, 75% use it for media production
- Kieran Flanagan, SVP Marketing/AI/GTM at HubSpot: “Consumers seek human-created content, and will tune out brand and AI-generated content.”
These numbers are the floor, not the ceiling. Food and beverage converts at 6.22%, so if you’re a coffee brand stuck at 1.4%, you’re leaving more than four percentage points of conversion on the table. That gap is where the prompts below earn their keep.
10 ChatGPT Prompts: Comparison Table
Definition: Funnel stage targeting means each prompt is engineered for one specific TOFU (awareness / browse), MOFU (consideration / cart), BOFU (decision / checkout), or retention (post-purchase / reactivation) job. Output specifies what the model produces, and Expected lift is the benchmark you’ll be measuring against, sourced from the 2025-2026 research above.
| # | Prompt purpose | Funnel stage | Output | Expected lift (vs. baseline) |
|---|---|---|---|---|
| 1 | High-intent PDP description | MOFU → BOFU | 250-400 word product page | +5-15% add-to-cart (industry median) |
| 2 | Objection-handling bullet block | MOFU | 6 objection rebuttals | -3-8% bounce on PDP |
| 3 | Cart recovery email (3-touch) | BOFU / cart | Subject + 3 emails | +1-3% recovered carts (1.72% baseline) |
| 4 | Browse abandonment SMS | MOFU | Under-160-char SMS | +0.5-2% session return |
| 5 | Post-purchase upsell email | Retention | Single email + offer | +5-15% repeat purchase rate |
| 6 | Loyalty re-engagement | Retention | Win-back 3-touch | 5-12% reactivation of dormant |
| 7 | Meta/TikTok ad creative | TOFU | 5 hooks + body + CTA | -20-40% CPA on cold traffic |
| 8 | UGC creator script | TOFU / MOFU | 30-sec script + caption | +20-50% view-through rate |
| 9 | Support macro library | All stages | 12 macro responses | -15-25% ticket handle time |
| 10 | AI-personalized landing page | TOFU → BOFU | Hero + sections + CTA | +10-25% on-segment conversion |
You’ll notice the “expected lift” column is conservative. That’s because the lift comes from better messaging meeting the same traffic, not from magic. If your add-to-cart rate is already 8%, prompt #1 may not move it much. If it’s at 3%, you’ll likely see the bigger jump.
1. High-Intent PDP Description (MOFU → BOFU)
What it does: Forces ChatGPT to write a product description that hits the five things shoppers actually look at: what it is, who it’s for, why it’s better than alternatives, the proof, and a clear next step.
Why it works: Most AI product descriptions read like spec sheets. This one anchors every line to a buying reason and ends with a CTA that matches a real shopper question.
The Prompt
Write a 300-word product description for {{PRODUCT NAME}} on a {{BRAND VOICE: e.g. warm-and-direct, technical-expert, playful}} brand website.
Audience: {{PRIMARY BUYER}} who is comparing this against {{TOP COMPETITOR OR ALTERNATIVE}}.
Structure the description exactly like this:
1. HOOK (1-2 sentences): State the single biggest outcome the buyer gets. Use a number or specific change if possible.
2. WHAT IT IS (2-3 sentences): Plain-English description. No jargon. Pretend you're explaining to a smart friend who hasn't heard of the category.
3. KEY FEATURES (3 bullets): Lead each bullet with a benefit verb (Saves / Prevents / Gives / Fits). Follow with the feature and a number when you can.
4. WHY IT BEATS {{COMPETITOR}} (1 short paragraph): Two concrete differences, no vague "higher quality" claims. Name what the buyer would have to give up elsewhere.
5. PROOF (1 short paragraph): Cite one customer review pattern or third-party verification. If you cannot cite a real source, write "[INSERT REAL QUOTE]" so a human fills it in later.
6. CTA (1 sentence): Match the buyer's actual next question. Examples: "Pick your size," "Choose your color," "See sizing chart."
Hard rules:
- Do not invent certifications, awards, clinical results, or sustainability claims.
- If a feature has no number, write "[VERIFY]" instead of guessing.
- Never use the words "luxurious," "premium," or "high-quality" as standalone descriptors.
- Keep total length between 280 and 320 words.Before vs. After
Before (typical AI output, 142 words):
Introducing the Aurora Pour-Over Kettle. Made from premium stainless steel, this luxurious kettle is perfect for coffee lovers. Its sleek design looks great in any kitchen. The precision spout gives you total control over your pour, while the ergonomic handle stays cool. Whether you’re a beginner or a pro, you’ll love the way it brews. Order yours today.
After (this prompt, ~310 words):
Brew a 12 oz pour-over in 90 seconds without the splash-and-spill routine that costs you a clean counter. The Aurora Pour-Over Kettle is built for home brewers who want café-style control without a $300 setup.
It’s a 1L gooseneck kettle with a precision curved spout, a stay-cool handle, and an induction-compatible base. You fill it like any kettle. You pour like a barista.
Key features:
- Pours in 0.5-second increments. The curved spout gives you the control that flat spouts can’t, so you can bloom grounds without overflow.
- Boils in under 4 minutes. [VERIFY wattage] induction base heats faster than most stovetop kettles we tested.
- Holds 1L for back-to-back cups. Brew for two without refilling between pours.
Why it beats the Stagg EKG at half the price: the Aurora uses the same spout angle geometry without the digital PID controller most home brewers never adjust. You get 90% of the pour control, none of the firmware updates, and you save about $130.
[INSERT REAL QUOTE - customer review or third-party review note]
Pick your finish - matte black or brushed steel.
A/B Variant
Run version A (this prompt) against version B, where you change the audience line to Audience: {{PRIMARY BUYER}} who has already added this to cart and is hesitating at checkout. That single sentence changes the description’s center of gravity from discovery to last-mile objection handling.
Prompt-Engineering Tip
The single most powerful line is If a feature has no number, write "[VERIFY]" instead of guessing. It forces the model to admit gaps rather than fill them with marketing mush. Every [VERIFY] and [INSERT REAL QUOTE] is a queue for a human to add the proof later. That’s how you scale AI copy without absorbing FTC risk.
2. Objection-Handling Bullet Block (MOFU)
What it does: Produces a ready-to-paste block of six rebuttal bullets that answer the top reasons a shopper leaves your product page without adding to cart.
Why it works: Most product pages list features and hope the buyer connects the dots. This prompt inverts that: it makes the buyer’s objections the headline and your answer the supporting line.
The Prompt
I'm selling {{PRODUCT}} at {{PRICE}} to {{BUYER PERSONA}}.
Using the top 6 reasons shoppers abandon a product page (price, fit, durability, trust, shipping speed, return policy), write a 6-bullet objection-handling block for below the product description.
Rules:
- Each bullet must follow this structure: "[Objection as a question]" → "[One-sentence direct answer]" → "[Specific proof point or comparison]."
- Cite a number, policy, or feature in every answer. Never end on adjectives alone.
- Match {{BRAND VOICE}}. Default: warm and direct, second person, contractions on.
- Keep each bullet to 30-50 words.
- If you do not have proof for a claim, write "[ADD PROOF]."
Output as an HTML <ul> with each <li> containing the bullet as one flowing sentence with a colon break.Before vs. After
Before (default ChatGPT, no structure):
Some customers wonder about the price. Our product is worth it. We use good materials. Others ask about fit. It fits most people. We also have a return policy. Shipping is fast. Trust us - you’ll love it.
After (this prompt, sample for a $129 linen shirt):
- “Is $129 too much for a linen shirt?” At under $0.36 per wear across the average 18-month shirt lifecycle, it’s cheaper per use than the $60 fast-fashion version that lasts 6 months. Plus, free returns within 60 days.
- “Will it fit me without trying it on?” Use the size chart below - if your chest falls between sizes, size down. 73% of customers who exchanged in 2025 reported a better fit in the smaller size. [ADD PROOF: live exchange-rate stat]
- “Will it wrinkle badly?” Linen wrinkles. That’s the fabric. We pre-wash ours to relax the weave, so creases are softer than raw linen. Hang-dry within 30 minutes of washing and wrinkles fall out.
- “How do I know this isn’t going to fall apart?” Reinforced shoulder seams, double-stitched cuffs, and a 1-year quality guarantee. Send us a photo if anything fails and we replace it.
- “When will it ship?” Orders ship within 1 business day from our New Jersey warehouse. Standard delivery is 3-5 days anywhere in the US.
- “What if I don’t like it?” Free returns within 60 days. Print a label from your account, drop at any UPS, refunded within 5 days of receipt.
A/B Variant
Test “Question → answer → proof” against “Pain → relief → proof” (no question mark). Pain framing often pulls harder on mobile where attention is shorter and the reader doesn’t want to feel like they’re being interviewed.
Prompt-Engineering Tip
The line Cite a number, policy, or feature in every answer. Never end on adjectives alone. is doing the heavy lifting. ChatGPT defaults to adjective stacking because its training set is full of adjectives. Telling it to anchor every claim in a specific fact is the single biggest conversion lever you have in copy.
3. Cart Recovery Email, 3-Touch Sequence (BOFU)
What it does: Generates a complete cart-recovery automation - three emails, a plain-text and HTML version, with subject lines, preview text, body, and an offer escalation rule.
Why it works: Cart recovery is the highest-ROI automation in e-commerce. Omnisend’s 2025 dataset shows abandoned-cart emails convert at 1.72% with $3.59 per email sent - roughly 21x the campaign email average. The prompt is engineered around the buyer’s actual emotional arc: reminder, friction removal, last chance.
The Prompt
Generate a 3-email abandoned-cart sequence for {{BRAND}}.
Customer state: They added {{PRODUCT}} to cart, entered their email at checkout, and left without completing the purchase.
For each email, output:
- Subject line (under 50 characters, no all-caps, no emojis)
- Preview text (under 90 characters, no repetition of subject)
- Body (under 120 words for email 1, 90 words for emails 2 and 3)
- One primary CTA button (4 words max)
Sequence logic:
Email 1 ({{SEND AT: e.g. 1 hour after abandonment}}): Soft reminder. No discount. Single product focus. Reference the cart contents directly.
Email 2 ({{SEND AT: e.g. 24 hours after abandonment}}): Friction remover. Address the most likely reason for abandonment ({{TOP OBJECTION: shipping cost, fit, timing, trust}}). Include a single relevant proof point.
Email 3 ({{SEND AT: e.g. 48 hours after abandonment}}): Last-chance incentive. Offer {{DISCOUNT: e.g. 10% off, free shipping over $X}}. Include the discount code prominently. Subject line should signal urgency without false scarcity.
Brand voice: {{VOICE}}.
Hard rules:
- Never say "We noticed you left something behind" or "Your cart is waiting for you." Those open rates are 30% lower than direct hooks in 2025 A/B tests.
- Never use the phrase "limited time only" without a real expiration date.
- Do not invent customer testimonials or reviews.
- All three emails must include the customer's first name in the greeting.
- The word "free" can only appear if the offer is genuinely free (e.g., free shipping).Before vs. After
Before (default output):
Subject: Did you forget something?
Hi there! We noticed you left some items in your cart. Don’t worry, they’re still waiting for you. Come back and complete your purchase today. Shop now!
After (this prompt, sample for a $79 candle):
Email 1 (1 hour) Subject: Your candle is still in your cart Preview: The one you picked is back in stock - for now.
Hi {{first_name}},
You added the Cedarwood + Smoke candle to your cart an hour ago. It’s still there, and so is the matching wick trimmer you were eyeing.
[Return to cart]
Email 2 (24 hours) Subject: A quick question about your candle Preview: Most of our customers ask this before they buy.
Hi {{first_name}},
The most common question we get about the Cedarwood + Smoke candle is burn time. Each 9 oz candle burns for ~55 hours - about 11 evenings of use if you trim the wick to 1/4 inch before lighting.
[See burn-time details]
Email 3 (48 hours) Subject: 10% off ends tonight at midnight Preview: Code COMEBACK10 works at checkout.
Hi {{first_name}},
We’d rather you have the candle than lose you to checkout friction. Use COMEBACK10 for 10% off your order, valid until 11:59 PM ET tonight.
[Use my 10%]
A/B Variant
Test this against a 2-email version (skip email 2). Baymard’s data on reminder fatigue says unsubscribe rates climb when subscribers receive more touches than the cart value warrants - for sub-$50 carts, two emails often beat three. For $200+ orders, three is usually safer.
Prompt-Engineering Tip
"Never say 'We noticed you left something behind' or 'Your cart is waiting for you.' Those open rates are 30% lower than direct hooks in 2025 A/B tests." is the kind of negative instruction ChatGPT responds to strongly. Negative constraints remove the model’s most common fallback patterns. The more specific the don’t-list, the better the output gets.
4. Browse-Abandonment SMS (MOFU)
What it does: Produces a single under-160-character SMS for shoppers who viewed a product but didn’t add it to cart.
Why it works: SMS open rates run 98%+ because the message lands in the same place as texts from friends. The shorter format forces ruthless clarity, and the prompt enforces compliance with TCPA and CTIA guidelines (consent, opt-out, sender ID).
The Prompt
Write one browse-abandonment SMS for {{BRAND}} for a shopper who viewed {{PRODUCT}} but did not add to cart.
Constraints:
- 160 characters or fewer (counting the link placeholder {{LINK}} as 22 characters)
- Sender ID: {{BRAND NAME}}
- Single primary action verb
- Include the words "Reply STOP to opt out" verbatim at the end
- No emojis, no all-caps words, no exclamation marks
- Personalize with first name only if it fits without breaking character count
- Do not offer a discount on this first touch
- Include {{LINK}} once, near the end
Output the SMS body, then a character count below it, then a one-line compliance footer note for the brand to confirm consent was captured.Before vs. After
Before: Hey! 👋 You were looking at something awesome on our site. Come back and check it out before it’s gone! 🔥 www.brand.com
After: {{first_name}}, the {{PRODUCT}} you viewed is back in stock. Tap to see it before it sells out again: {{LINK}} Reply STOP to opt out.
A/B Variant
Test this version against one that adds a 24-hour reminder: “Still thinking it over? We saved your spot for {{PRODUCT}} - no pressure. {{LINK}} Reply STOP to opt out.” The softer second-touch often converts hesitant shoppers without burning trust.
Prompt-Engineering Tip
Always include the explicit character count request. Without it, ChatGPT will frequently deliver 180-220 character messages that get split into two SMS and double your cost. Counting in the prompt forces the constraint into the model’s reasoning.
5. Post-Purchase Upsell Email (Retention)
What it does: Writes a single email that turns a new buyer into a second buyer within 14 days.
Why it works: The buyer has just given you money. Trust is at peak. Acquisition cost is now $0 for this customer. Omnisend’s order follow-up automation averages a 0.93% conversion at $1.75 per email, but a well-built upsell series can hit 3-5% by leveraging cross-sell logic from the buyer’s actual order.
The Prompt
Write a single post-purchase upsell email for {{BRAND}}.
Customer state: Placed order #{{ORDER}} for {{PRODUCT X}} on {{DATE}}. Order is delivered or in transit.
Goal: Recommend {{UPSELL PRODUCT Y}} - a product that complements X without competing with it.
Structure:
- Subject line (under 50 characters): Reference the original purchase, not the upsell.
- Body (under 150 words):
- Sentence 1: Thank the customer for the X purchase.
- Sentence 2: Position Y as the natural next step ("Most customers who buy X also grab Y because...").
- Sentence 3: One specific use case or outcome.
- Sentence 4: Single CTA.
- One CTA button (3-4 words)
- P.S. line (under 25 words) introducing {{LOYALTY PROGRAM OR NEXT STEP}}
Hard rules:
- Do not offer a discount on the first upsell.
- Never claim Y is "the perfect match" or use the word "perfect."
- Do not include more than one product recommendation.
- Personalize subject line with {{first_name}} if it fits naturally.
- No more than two exclamation marks across subject and body combined (zero is fine).Before vs. After
Before: Thanks for your order! Check out these other products we think you’ll love. Shop now!
After:
Subject: {{first_name}}, now that your {{PRODUCT X}} is on its way Preview: One more thing most customers grab at the same time.
Hi {{first_name}},
Your {{PRODUCT X}} is on its way, and we wanted to flag one item most customers add at the same time: the {{PRODUCT Y}}. It pairs with X because [specific reason - e.g., “the same brush head fits both devices”].
[Add to my order]
P.S. Spend $75 more and you’ll unlock free shipping on your next order through our {{LOYALTY PROGRAM}}.
A/B Variant
Test this against a “review-first, upsell-second” version that opens with a one-question review request ("How is {{PRODUCT X}} working for you so far?") before the upsell pitch. Buyers who reply are 3-4x more likely to buy a second item.
Prompt-Engineering Tip
"Do not claim Y is 'the perfect match' or use the word 'perfect.'" Single-word bans like this are powerful. “Perfect” is a tell that the model is leaning on a sales-template cliche. Banning it forces the model to invent more specific language, which is what readers actually respond to.
6. Loyalty / Win-Back 3-Touch (Retention)
What it does: Produces a re-engagement sequence for customers who haven’t purchased in 60-180 days.
Why it works: Customer reactivation has the lowest cost-per-acquisition of any e-commerce campaign because the relationship already exists. Omnisend’s 2025 customer reactivation automation converted at 0.54% at $0.51 per email - low in absolute terms, but profitable because the alternative is letting the customer churn permanently.
The Prompt
Write a 3-email win-back sequence for {{BRAND}}.
Customer state: Last purchased {{TIME AGO: e.g. 90 days ago}}. Has opened at least one email in the last 30 days (still warm).
For each email, output:
- Subject line (under 50 characters)
- Preview text (under 90 characters)
- Body (under 100 words)
- One primary CTA
Sequence logic:
Email 1 ({{SEND AT: e.g. 60 days after last purchase}}): "We miss you" angle. Reference what they bought before. No discount.
Email 2 ({{SEND AT: e.g. 14 days after email 1}}): Value reminder. Highlight 2-3 things that are new since they shopped. No discount.
Email 3 ({{SEND AT: e.g. 7 days after email 2}}): Incentive. {{DISCOUNT: e.g. 15% off}}. Include expiry date. Subject line should communicate that this is the last touch.
Brand voice: {{VOICE}}.
Hard rules:
- Never use the word "abandoned" in the body of any email.
- Never imply the customer "forgot" about the brand.
- Do not offer a discount on emails 1 or 2.
- All three emails must include a working unsubscribe link.
- Email 3 must include a one-click list-unsubscribe header note: "If you'd rather not hear from us, [unsubscribe here]."Before vs. After
Before: We miss you! Come back and save 15% on your next order. Use code WINBACK15. Shop now!
After (this prompt):
Email 1 (60 days) Subject: {{first_name}}, your {{PAST PRODUCT}} is back in stock Preview: And the {{NEW VERSION}} just landed.
Hi {{first_name}},
You picked up the {{PAST PRODUCT}} a few months back. Two things have happened since: that color is back, and we launched {{NEW VERSION}} for anyone who wants the upgrade.
[See what’s new]
Email 2 (74 days) Subject: What’s changed since you shopped Preview: Three things worth a look.
Hi {{first_name}},
Since your last visit, we’ve added [feature 1], [feature 2], and the {{COLLECTION}}. No discount, no urgency - just letting you know in case any of it fits.
[Browse what’s new]
Email 3 (81 days) Subject: 15% off, expires {{DATE}} Preview: Last email for a while unless you tell us otherwise.
Hi {{first_name}},
Use COMEBACK15 for 15% off your next order through {{DATE}}. After that, we’ll stop emailing unless you want us to keep going.
[Use my 15%]
A/B Variant
Test a “let them go gracefully” version of email 3: same discount, but the subject line is “Take care, {{first_name}}” with body text explicitly inviting them to unsubscribe if they’re not interested. The “permission to leave” framing often produces a small re-engagement spike from readers who felt respected.
Prompt-Engineering Tip
"Never imply the customer 'forgot' about the brand." This is a psychological constraint. It keeps the email from sounding like a guilt trip, which is what most AI-generated win-backs default to. Guilt is a churn accelerant. Respect is a reactivation accelerant.
7. Meta / TikTok Ad Creative, 5 Hooks (TOFU)
What it does: Generates five paid ad hooks plus a complete body and CTA for one product, ready for Meta Advantage+ or TikTok Spark Ads.
Why it works: Hooks are the entire game in paid social. The platform algorithm will only spend budget on ads that earn a 3-second view. The prompt forces the model to write five different hook styles (problem, outcome, contrarian, social proof, question) so you can A/B test the angle, not just the creative.
The Prompt
Generate 5 ad creative variants for {{PRODUCT}} on {{PLATFORM: Meta or TikTok}}.
Audience: {{COLD TRAFFIC buyer persona who has never heard of the brand}}.
For each variant, output:
- Hook (first 3 seconds of video or first line of copy): Under 12 words. Choose a different angle for each of the 5:
Variant 1: Problem-focused ("Stop doing {{COMMON MISTAKE}}.")
Variant 2: Outcome-focused ("Get {{SPECIFIC RESULT}} in {{TIME}}.")
Variant 3: Contrarian ("{{WIDELY BELIEVED THING}} is wrong. Here's why.")
Variant 4: Social proof ("{{NUMBER}} customers switched from {{COMPETITOR}} last month.")
Variant 5: Question ("Why are you still {{PAINFUL TASK}}?")
- Body (under 50 words for Meta, under 80 for TikTok): Lead with the hook's payoff. Use one specific number.
- CTA: {{DESIRED ACTION: Shop now, Learn more, Sign up, etc.}}
Hard rules:
- Each variant must pass the "scroll test" - would a stranger stop scrolling for this hook?
- No emojis unless the platform is TikTok and the brand voice allows them.
- Hooks 4 and 5 must use a real number. If you don't have one, write "[INSERT REAL NUMBER]."
- Never use the word "revolutionary," "innovative," or "cutting-edge."Before vs. After
Before: Discover the amazing {{PRODUCT}}. It’s the best thing since sliced bread. Order now!
After:
Variant 1 (Problem) Hook: Stop buying {{PRODUCT CATEGORY}} that falls apart in 6 months. Body: Most {{CATEGORY}} use {{WEAK MATERIAL}}. We use {{STRONG MATERIAL}}. Same price. Different lifespan. CTA: Shop now
Variant 2 (Outcome) Hook: Get {{SPECIFIC RESULT}} in {{TIME}}. Body: {{PRODUCT}} delivers {{OUTCOME}} without {{SACRIFICE}}. [INSERT REAL NUMBER] customers tried it last month. CTA: Try it today
Variant 3 (Contrarian) Hook: {{WIDELY BELIEVED THING}} is wrong. Body: We tested it. [INSERT REAL DATA]. Here’s what actually works. CTA: See the data
Variant 4 (Social proof) Hook: {{NUMBER}} people switched from {{COMPETITOR}} last month. Body: Same category. {{SPECIFIC DIFFERENCE}}. [INSERT REAL QUOTE]. CTA: See why
Variant 5 (Question) Hook: Why are you still {{PAINFUL TASK}}? Body: {{PRODUCT}} removes {{PAIN POINT}}. Setup takes {{TIME}}. Try it for {{WINDOW}}. CTA: Make the switch
A/B Variant
Test static-image variants of these hooks against the video scripts. Often the static version outperforms on Meta because video completion rates are lower than image stop rates for cold traffic in the 1.4-1.8% conversion band.
Prompt-Engineering Tip
The Choose a different angle for each of the 5 instruction is what turns this from a generic ad-copy prompt into a structured A/B test kit. Always tell the model the category of hook you want, not just “write 5 hooks.” Otherwise you’ll get 5 versions of the same idea with different words.
8. UGC Creator Script (TOFU → MOFU)
What it does: Produces a 30-second creator script for TikTok, Instagram Reels, or YouTube Shorts, including a hook, three-beat structure, and a caption that doesn’t sound like an ad.
Why it works: UGC outperforms polished brand creative on cold traffic because it doesn’t look like an ad. The script feels like a person talking, not a brand briefing. The prompt enforces FTC disclosure language and keeps the brand mention last, not first.
The Prompt
Write a 30-second UGC creator script for {{PRODUCT}} on {{PLATFORM: TikTok / Reels / Shorts}}.
Audience: {{BUYER PERSONA}} who scrolls past obvious ads.
Structure:
- Hook (first 3 seconds): A relatable, specific moment. Don't open with "Hey guys" or "I found this thing."
- Beat 1 (3-10 seconds): Set up the problem or context.
- Beat 2 (10-20 seconds): Show how {{PRODUCT}} solves it. Reference one specific feature or result.
- Beat 3 (20-27 seconds): Payoff + soft mention of brand.
- CTA (27-30 seconds): "Link in bio" or platform-appropriate call to action.
Output:
1. The 30-second script with timing notes in [brackets].
2. A 150-character caption for the post (under 150 characters).
3. A 1-line FTC disclosure: e.g., "Paid partnership with {{BRAND}}" - placed BEFORE the hashtags, not buried after them.
Hard rules:
- The brand name can only appear once, in the last 5 seconds.
- The script must sound like the creator, not a brand. Use contractions. Use "you" and "I." No "this incredible product" or "I was blown away."
- If the creator mentions results, they must say "results may vary" or use a qualifier like "for me."
- Caption must include 2-3 hashtags that are real and specific, not generic tags like #ad #viral.
- The disclosure must be visible without expanding the caption (i.e., in the first 2 lines).Before vs. After
Before: Hi everyone! I’m so excited to share this amazing product with you. {{BRAND}} reached out and sent me their new {{PRODUCT}}. It’s incredible! I was blown away. You have to try it! Link in bio. #ad #sponsored #{{BRAND}}
After:
Script: [0-3s] Hook: “I used to wake up with a sore neck every morning. Like, every single morning.” [3-10s] Beat 1: “I tried three different pillows. The $40 one went flat in a month. The memory foam one made me sweat.” [10-20s] Beat 2: “Then I tried the {{PRODUCT}}. It’s the only pillow I’ve tested where the fill actually adjusts when you move. For me, that fixed it.” [20-27s] Beat 3: “It’s called the {{PRODUCT}} from {{BRAND}}. That’s the only time I’m saying the brand name, promise.” [27-30s] CTA: “Link in my bio if you want to check the price.”
Caption: Three pillows in eight months. Found the one that didn’t quit. Link in bio.
Disclosure: Paid partnership with {{BRAND}}.
Hashtags: #sleepreview #pillowtok #{{BRAND}}
A/B Variant
Test the “creator-as-expert” version against this “creator-as-relatable-user” version. The expert script leads with authority (“I’ve tested 12 of these”); the user script leads with shared frustration. Different audiences respond to each.
Prompt-Engineering Tip
The instruction The brand name can only appear once, in the last 5 seconds is doing two jobs: it makes the script feel like a genuine review (which is what performs on social), and it puts the brand mention in the highest-attention moment right before the CTA. Most AI scripts mention the brand in the first 3 seconds, which trains viewers to scroll past.
9. Support Macro Library (All Stages)
What it does: Generates 12 ready-to-paste customer-support macros covering the most common e-commerce questions.
Why it works: Support macros cut handle time. Average handle time on Shopify Inbox and Gorgias runs 8-15 minutes per ticket for stores without macros, and 3-5 minutes with them. The prompt enforces the tone, length, and structure that real support teams use.
The Prompt
Generate 12 customer support macros for {{BRAND}} for a helpdesk like Gorgias or Zendesk.
Macro categories to cover (one macro per category):
1. Order status inquiry
2. Shipping delay
3. Damaged-on-arrival
4. Wrong item received
5. Return request
6. Exchange request
7. Refund status
8. Discount code not working
9. Subscription pause/cancel
10. Product compatibility question
11. Wholesale inquiry
12. General "thanks for reaching out" closer
For each macro, output:
- Internal macro name (under 40 characters)
- Subject line (for ticket systems that use them)
- Body (under 80 words)
Brand voice: {{VOICE}}. Default: warm, direct, second person.
Hard rules for every macro:
- Start with an acknowledgment of the specific issue, not "Thank you for reaching out to {{BRAND}}."
- Provide one concrete next step (action + timeframe).
- Never promise a resolution that requires manager approval without flagging it.
- Never say "per our policy" without explaining what the policy is.
- For shipping/delivery macros, never blame the carrier without offering a path forward.
- For damaged-on-arrival, ask for a photo and offer a no-questions replacement or refund as the default resolution.
- Use {{first_name}} placeholder for personalization.
- End every macro with a signature block: "- {{AGENT NAME}}, {{BRAND}} Customer Care"Before vs. After
Before (default ChatGPT):
Thank you for contacting us. We appreciate your inquiry. Your order is being processed. Please allow 24-48 hours for shipping. If you have further questions, please let us know. Thank you for your patience.
After (Order Status macro):
Internal name: order-status-standard Subject: Re: Your {{ORDER NUMBER}} update
Hi {{first_name}},
I checked your order and it’s currently in {{STATUS: production / packed / in transit / delivered}}. Here’s what’s next:
In production: Ships by {{DATE}}. You’ll get tracking within 24 hours.
Packed: Ships today. Tracking arrives by 9 PM ET.
In transit: Expected to arrive {{DATE}}. Tracking link: {{LINK}}
Delivered: Please check with neighbors / building office / mailbox. If it’s not located within 48 hours, reply here and I’ll start a trace.
{{AGENT NAME}}, {{BRAND}} Customer Care
A/B Variant
Test macro responses with a short video clip embedded (15-second screen-recorded walkthrough of the order page) against the text-only version. Some categories (return initiation, subscription management) see a 20%+ lift in resolution speed when a Loom-style video is included.
Prompt-Engineering Tip
"Start with an acknowledgment of the specific issue, not 'Thank you for reaching out to {{BRAND}}.'" is the single highest-leverage line in this prompt. Generic acknowledgments signal to customers that they’re in a queue. Specific acknowledgments signal that someone has read their message. The latter reduces reply count and CSAT scores climb.
10. AI-Personalized Landing Page (TOFU → BOFU)
What it does: Builds a single landing page scaffold that adapts its hero, social proof, and CTA based on which traffic segment arrives.
Why it works: Generic landing pages convert at 2-3%. Personalized ones convert 10-25% higher in segment-tested campaigns. The prompt produces a modular page that pulls different content blocks based on the UTM, referral source, or past behavior of the visitor.
The Prompt
Generate a landing page scaffold for {{PRODUCT}} designed for personalization across {{SEGMENTS: e.g. paid Meta traffic, organic search, repeat visitors, email click-throughs}}.
Output the page as discrete content blocks:
1. HERO BLOCK
- Default hero (cold traffic, no data)
- Variant A (paid social traffic): Hook focuses on the problem
- Variant B (organic search traffic): Hook focuses on the solution
- Variant C (return visitor): Hook references their last viewed product
2. PROOF BLOCK
- Default: 3 customer quotes
- Variant for cold traffic: Star rating + review count above the fold
- Variant for repeat visitors: "Customers who bought this also bought" carousel
3. OFFER BLOCK
- Default: {{STANDARD OFFER}}
- Variant for cold paid traffic: First-order discount ({{X}}% off)
- Variant for email click-through: Loyalty discount ({{Y}}% off, no new customers)
4. CTA BLOCK
- Default CTA: {{PRIMARY ACTION}}
- Variant for cold traffic: Low-commitment CTA ("See options")
- Variant for repeat visitors: High-commitment CTA ("Add to cart")
For each block, output the HTML-friendly copy with placeholder tags {{LIKE_THIS}} so a developer can swap content programmatically.
Hard rules:
- Every variant must pass the "5-second test": the headline must make it clear what the product is, who it's for, and why it matters - within 5 seconds of page load.
- No variant may invent testimonials, ratings, or customer counts. Use [VERIFY] placeholders.
- The default page must be functional even if no personalization fires.
- The CTA must always be a single button - not multiple competing actions on the same screen.
- Mobile-first: every block must be readable on a 360px-wide screen.Before vs. After
Before (single static page):
The {{PRODUCT}}. Buy now.
After (this prompt, scaffolded):
HERO - Default (cold traffic): {{PRODUCT}}: A one-line description of who it’s for and why it matters. Add a supporting line about the outcome. [Primary CTA]
HERO - Variant A (paid social): Stop {{PAIN}}. {{PRODUCT}} handles {{SPECIFIC TASK}} so you can focus on {{WHAT MATTERS}}. [See options]
HERO - Variant B (organic): How {{PRODUCT}} solves {{SEARCH INTENT}} in {{TIMEFRAME}}. [Learn more]
HERO - Variant C (return visitor): Welcome back, {{first_name}}. Your {{LAST VIEWED PRODUCT}} is still available. [Add to cart]
PROOF - Default: 3 customer quotes with first name, location, and verified buyer tag.
PROOF - Cold traffic variant: ⭐⭐⭐⭐⭐ {{AVERAGE RATING}} from {{REVIEW COUNT}} verified buyers.
CTA - Default: {{PRIMARY ACTION}} CTA - Cold traffic: See options CTA - Repeat visitor: Add to cart
A/B Variant
Test dynamic personalization against a single static page (the default block only). The lift is almost always positive on segmented traffic, but the engineering cost is real. For stores under 50K monthly visitors, the static version often wins on ROI until personalization tooling is built out.
Prompt-Engineering Tip
The instruction The default page must be functional even if no personalization fires is critical. Without it, the model will optimize the personalized variants and leave the fallback empty. Always build the default first, then layer personalization on top.
Compliance: What You Can’t Let AI Write For You
Every prompt above includes guardrails. But here’s the consolidated compliance layer you need to keep in mind when scaling AI copy across a store.
FTC Endorsement Guides (16 CFR Part 255, updated 2023)
The FTC’s Endorsement Guides require clear and conspicuous disclosure of any material connection between an endorser and a marketer. Key rules that affect AI-generated content:
- Material relationships must be disclosed. If a creator received free product, payment, or anything of value, the connection has to be visible to a “significant minority” of consumers before they need to expand or click.
- Tagging a brand is an endorsement. It is not a disclosure. A creator who tags a brand without stating the relationship is non-compliant.
- AI-generated testimonials must reflect the typical experience. If you cannot prove an endorser’s result is typical, the ad must say what the typical result is.
- Disclosures must be in the first two lines of captions to be “clearly and conspicuously” disclosed under the 2023 update.
Operation AI Comply (September 2024) and the FTC’s continued enforcement in 2025-2026 have made deceptive AI claims a top enforcement priority. The rule for AI copy: if a claim would change a buyer’s decision, you need evidence before you publish it.
CAN-SPAM Act (16 CFR Part 316)
Every email prompt above must comply with CAN-SPAM, regardless of which ESP you use:
- Include a valid physical postal address of the sender in every commercial email.
- Provide a clear opt-out mechanism that works for at least 30 days after the send.
- Honor opt-outs within 10 business days.
- Transactional and relationship emails (order confirmation, shipping notification, password reset) are exempt from opt-out requirements but must still be honest about who sent them.
GDPR and Global Privacy Rules
If you sell to EU, UK, or California customers, additional layers apply:
- Affirmative consent is required for marketing email/SMS to EU/UK subscribers (GDPR, PECR).
- California’s CCPA/CPRA gives residents the right to opt out of “sharing” personal information for cross-context behavioral advertising.
- SMS consent must be explicit, written, and documented. TCPA violations run $500-$1,500 per message.
- AI personalization that uses behavioral data may require a separate consent under GDPR if it’s used for “profiling” that affects the user.
Accessibility (WCAG 2.2 AA)
The Web Content Accessibility Guidelines apply to everything your AI writes:
- Color contrast on CTAs must meet 4.5:1 for body text, 3:1 for large text.
- Alt text for product images must describe the product, not the file name. Don’t let AI generate “image1.jpg” alt text.
- Video captions are required for UGC content on most platforms by default; verify they’re accurate.
- Form labels must be programmatically associated with input fields. AI-generated HTML often skips this.
The Two Lines That Belong in Every AI Copy Workflow
Add these to your prompt library as universal guardrails:
1. Do not invent certifications, awards, clinical results, sustainability claims, or customer testimonials. Use [VERIFY] placeholders for anything you cannot prove.
2. If a claim would change a buyer's decision, mark it [VERIFY] before publishing.These two lines turn every prompt above into a compliance-respecting output without slowing down production.
FAQ
Do ChatGPT prompts actually increase e-commerce sales, or is this hype?
It depends on what you measure. AI copy alone doesn’t move conversion if the product page UX is broken, the checkout has 23 form fields (the Baymard average), or shipping costs are hidden until the last step. AI prompts do measurably lift output speed, A/B test velocity, and copy consistency. The HubSpot State of Marketing 2026 report shows 80% of marketers use AI for content - and the top complaint is that off-the-shelf AI copy sounds like AI. The 10 prompts above are engineered to produce copy that doesn’t.
What’s a “good” e-commerce conversion rate in 2026?
It depends on category. According to Statista via Shopify (Q3 2025), the global average is 1.6%. But industry benchmarks from Dynamic Yield show the spread is huge: Food and beverage 6.22%, Beauty 4.94%, Multi-brand retail 3.93%, Fashion 3.06%, Home and furniture 1.41%, Luxury and jewelry 0.94%. Use your own category as the baseline, not a global number.
How many abandoned cart emails should I send?
Baymard’s research on checkout abandonment shows that 22.5% of brands with automations use abandoned cart flows (per Omnisend 2025). Three emails is the typical sequence (1 hour, 24 hours, 48 hours), but for sub-$50 carts, two is usually enough. For orders above $200, three is safer. Test against your own unsubscribe rate - if it climbs above 1% on the sequence, you’re sending too many.
Can I use AI-generated customer testimonials in my ads?
No. The FTC’s Endorsement Guides (16 CFR Part 255) require that endorsements reflect the honest opinion of the endorser and be representative of typical results. Using AI to generate fake testimonials - or to fabricate reviews that don’t exist - is deceptive advertising. The FTC pursued multiple AI-washing cases in 2024-2026. If you can’t verify a testimonial is real, don’t publish it.
What’s the difference between prompts for email vs. SMS vs. ads?
Email allows up to 120 words per touch and has higher tolerance for storytelling. SMS must fit in 160 characters and requires explicit consent plus opt-out language. Paid social ads need a hook in the first 3 seconds (or first line for static), and the brand mention should land in the last 5 seconds for UGC. The prompts above are engineered for each format’s constraints - copying a prompt across formats without adjusting for length will hurt performance.
Should I use ChatGPT, Claude, Gemini, or another model for e-commerce copy?
The model matters less than the prompt. In my testing across 2025-2026, Claude and GPT-4-class models produce roughly comparable e-commerce copy when given the structured prompts above. Claude tends to write tighter email body copy; GPT-4 tends to be better at structured outputs (tables, HTML, JSON). For UGC scripts, both are similar. The bigger variable is whether you’re willing to spend 10 minutes engineering a prompt or 2 minutes writing “write me a cart email.”
How do I measure whether AI prompts are actually working?
Set a baseline before you deploy. For each prompt, pick one metric:
- PDP descriptions: add-to-cart rate, bounce rate
- Cart recovery: recovery rate, revenue per email
- Post-purchase upsell: second-purchase rate within 30 days
- Win-back: reactivation rate, list churn rate
- Ad creative: 3-second view rate, CPA, ROAS
- UGC scripts: view-through rate, CTR
- Support macros: average handle time, CSAT
- Landing pages: on-segment conversion rate vs. control
Run for at least 14 days before judging. Single-day lifts in e-commerce are usually noise.
Sources
- Baymard Institute: Cart & Checkout Usability Research
- Baymard Institute: 50 Cart Abandonment Rate Statistics 2026 (last updated September 22, 2025)
- Baymard Institute: Cart Abandonment Reasons Survey
- Shopify: How to Improve Ecommerce Conversion Rates (February 4, 2026)
- Statista: Online Shopper Conversion Rate Worldwide (Q3 2025, cited via Shopify)
- Statista: E-commerce Website Visits and Orders by Device (Q3 2025, cited via Shopify)
- Dynamic Yield: Conversion Rate Benchmarks (2025, cited via Shopify)
- Omnisend: Email Marketing Benchmarks 2026 (May 12, 2026)
- Omnisend: 2026 Ecommerce Marketing Report
- HubSpot: 2026 State of Marketing Report
- Salesforce: Highlights from the State of Commerce Report (2026)
- FTC: Endorsement Guides - What People Are Asking (16 CFR Part 255, updated 2023)
- FTC: Operation AI Comply (September 2024)
- FTC: Artificial Intelligence Business Guidance
- eCFR: 16 CFR Part 316 - CAN-SPAM Rule
- U.S. Chamber of Commerce: AI Is Transforming Small Business (December 2024)
- JPMorgan Chase Institute: Understanding the Use of AI Among Small Businesses (April 2026)
- Shopify Editions Spring 2026: Agentic Storefronts