AI is good at producing twenty angles and hooks for an ecommerce product in a minute, and bad at knowing which of them is true, allowed or worth money. Use it to generate options inside real limits, then you pick, check and test. Below are prompts for angles, hooks, Meta and Google copy that fits character limits, a creative brief for your designer or an image tool, and a small testing plan.
Status of the prompts: untested templates. I did not run them against a live model or an ad account for this post. I also have no conversion numbers to share, and I won't invent any: results depend on your product, price, audience and budget.
Start with the offer sheet, not the ad
Ad copy goes wrong when the model improvises the offer. Write the facts down first and give them to every prompt.
OFFER SHEET
Product: [NAME]
Price: [Rs. X] (MRP [Rs. Y] if relevant)
Real offer right now: [E.g. free shipping above Rs. 499, or NONE]
Offer end date: [DATE or NONE]
Delivery: [DAYS, areas]
Payment options: [UPI, cards, COD? only what you really offer]
Returns/exchange: [YOUR ACTUAL POLICY]
Proof we really have: [REVIEW COUNT, press, certificates, or NONE]
Customer: [WHO, where, what they were doing before they found you]
Top 3 objections we hear: [LIST]
Words/claims we must not use: [LIST]
"Proof we really have: NONE" is allowed and common for new stores. The model should then write ads that work without social proof, not fake it.
Prompt: find angles before writing copy
An angle is the reason someone would care. Different angles beat different wording of the same angle.
Using the OFFER SHEET, list 10 distinct ad angles for this product. For each: the angle name, the customer's situation in one sentence, the objection it overcomes, and what evidence from the offer sheet supports it. Mark any angle that would need proof we don't have as NEEDS PROOF. Types to cover: problem/solution, comparison with the usual alternative, price/value, convenience, gifting, use-case specific, myth-busting, and a skeptic's angle.
[PASTE OFFER SHEET]
Pick three to five. Cut anything marked NEEDS PROOF unless you can supply the proof. I'd also cut any angle you wouldn't say to a customer's face.
Prompt: hooks for Meta ads and reels
The first line, or first three seconds of a video, decides if someone stops scrolling. Ask for hooks by type so you do not get ten rewrites of one idea.
For angle "[ANGLE]", write 12 hooks: 3 questions, 3 specific-situation statements, 3 numbers-based hooks that use ONLY numbers from the offer sheet, 3 contrarian statements. Each at most 90 characters. No exclamation marks, no 'Are you tired of', no claims about results that aren't on the offer sheet.
[PASTE OFFER SHEET]
For video, ask for the hook as a spoken line plus a visual direction: "What is on screen in the first three seconds?" A hook that describes a scene you can actually film beats a clever line you can't shoot.
Meta copy that fits the limits
Meta's Ads Guide page for Facebook Feed image ads, as I read it on 2026-10-09, lists a headline of 27 characters and primary text of 50 to 150 characters. I couldn't open other placement pages (they returned errors), and third-party references give different figures, for instance about 125 characters before truncation behind "See more" and up to 40 characters for headlines. I wouldn't trust any single number. Limits vary by placement, and Stories and Reels are tighter. The practical rule: write to the short number, then check the preview for each placement in Ads Manager.
Write 6 Meta ad variants for angle "[ANGLE]" using the OFFER SHEET.
Each variant: primary text (the first line must work alone, max 125 characters total for the visible part), headline (max 27 characters), and a one-line call to action from: Shop Now, Learn More, Order Now.
Return a table with columns: variant, primary text, headline, primary text characters, headline characters. If a draft is over the limit, rewrite it before returning.
Do not mention a discount, deadline or guarantee unless it is on the offer sheet.
Models miscount characters. Paste the table into a spreadsheet and use =LEN() on each cell rather than trusting the column the model gave you.
Google responsive search ad headlines
Google's help page on responsive search ads says you can supply 3 to 15 headlines (up to 30 characters each) and 2 to 4 descriptions (up to 90 characters each), plus two display path fields of up to 15 characters. Headlines and descriptions can appear in any order, and only some are guaranteed to show, so each line must make sense alone. Pinning a line to a position is possible but reduces combinations and can lower Ad Strength, so Google doesn't recommend it for most advertisers (page checked 2026-10-09).
Write 15 RSA headlines (max 30 characters each) and 4 descriptions (max 90 characters each) for the keyword "[KEYWORD]" selling [PRODUCT] in India.
Mix headline types: 4 keyword-led, 3 benefit, 3 offer (only from the offer sheet), 3 trust/policy (only real policies), 2 calls to action.
Each headline must make sense alone and next to any other headline. No repeated wording across more than two headlines. No exclamation marks, no price unless on the offer sheet.
Return a table with the text and the character count.
[PASTE OFFER SHEET]
Check counts again in your spreadsheet. Remove any headline that only works if a specific other line appears with it.
Prompt: a creative brief for the image or video
Copy and creative need to say the same thing. A short brief keeps your designer, freelancer or image tool aligned.
Create a one-page creative brief for a static image ad and a 15 second vertical video for angle "[ANGLE]".
Include: single message, who it is for, the first three seconds (visual + spoken line), shot list of 4 shots using only things we can film at home with the product, on-screen text (max 6 words per card), what the product must look like (list of details that must not change), CTA, and what to avoid (claims, imagery).
Keep it under 250 words.
[PASTE OFFER SHEET]
If an image tool will draw the visual, follow the process in the AI product photo workflow: real product photo in, scene around it, details checked. Don't ask a model to render text in the image; add copy in a design tool.
Audit before you launch
Run this check on every line the AI wrote.
| Check | Question |
|---|---|
| Truth | Is every number, offer, delivery time and guarantee on the offer sheet? |
| Limits | Does it fit the platform's character limit and the preview of each placement? |
| Policy | Does it make health, income or result claims, or use before/after framing? Check the platform's current ad policies, and for regulated categories ask a professional |
| Distinct | Are the variants different angles or just reworded? |
| Landing page | Does the page show the same offer, price and product as the ad? |
| Voice | Would your store actually say this? |
Mismatch between ad and landing page is the most common avoidable waste. The ad promises free shipping and the page shows a shipping fee. For the product page side, see Shopify product description prompts.
A small, honest testing plan
I can't tell you what click-through or conversion rate to expect, and neither can a prompt. A plan that works with any numbers:
- Round 1, angles. Pick 3 angles. Use the same image, same audience, same budget per ad. One variable only.
- Decide the metric first. Cost per purchase if you have enough purchases; otherwise a leading metric like cost per add to cart. Write it down before you look.
- Set a minimum before judging. For example, don't call a winner until each variant has spent at least a few multiples of your target cost per purchase. Pick the number from your own economics.
- Round 2, hooks. Take the winning angle, test 3 hooks.
- Round 3, creative. Winning angle plus hook, test image versus video.
- Log everything. A sheet with date, angle, hook, creative, spend, result, and the decision.
- Retire fast, scale slowly. Pause clear losers. Don't change the budget of a winner by large jumps in one go; follow the platform's guidance on learning phases.
For a fuller treatment of testing logic and why small samples mislead, see the A/B testing guide. It's written for prompts, but the sample-size reasoning carries over. More general marketing prompts are in prompting for marketing.
What AI is not good for in ads
- Choosing the winner. Only spend data does that.
- Knowing your margin. If your free-shipping offer loses money, the model won't tell you.
- Compliance. It can't tell you what a platform will approve this week.
- Your customer's real words. Paste real reviews, chat messages and support questions into the offer sheet; it will beat anything invented.
Keep the loop short: facts in, options out, you cut, you test, you log what you learned, and feed that back into the next offer sheet.
