ChatGPT / 4 min
ChatGPT ad copy prompts: separate channel, limits, CTA, and review rules
A practical ChatGPT ad copy prompt guide for drafting Google Ads, Meta ads, banners, and landing-page CTAs with channel limits, evidence, claims, and review rules.
Watch on YouTube
BananaNL in 1 minute
A quick look at how selected prompts move into NotebookLM and AI Chat input fields.
Watch on YouTubeA generic “write ad copy” prompt usually misses channel limits and claim checks
ChatGPT can quickly draft search ad headlines, Meta ad primary text, banner copy, landing-page CTAs, and campaign variations. But a vague request like “write ad copy” often produces generic hype because the audience, placement, action, proof, and compliance boundary are still unclear.
Google Ads Help lists text ad limits such as up to 30 characters for each headline and up to 90 characters for each description, and recommends using the description to highlight details plus a call to action. Meta Business Help recommends short text that makes the desired action obvious, with common guidance around 125 characters for primary text, 40 for headline, and 25 for description.
OpenAI prompting guidance recommends being specific about context, desired output, length, format, style, and constraints. For ad copy, add one more layer: do not let the model invent results, prices, rankings, comparisons, or regulated claims. Saving channel-specific prompt templates in BananaNL keeps these review rules close to the draft every time.
Five inputs to define before asking for ad copy
- Name the channel and placement. Search ads, paid social, banners, and landing-page hero copy need different length and structure.
- Choose one reader and one problem. Write whether the audience is new, comparing options, returning, or already aware of the product.
- Provide source facts and proof. Product details, price, offer conditions, allowed claims, and banned wording should be supplied instead of guessed.
- Specify each output field. For Google Ads, separate headlines and descriptions. For Meta ads, separate primary text, headline, description, and CTA. Keep Japanese, Korean, and Chinese copy shorter when character counting is stricter.
- Add review rules. Ask ChatGPT to flag unverified numbers, “No.1” claims, guaranteed outcomes, and sensitive categories such as health, finance, employment, housing, politics, or education.
Prepare these four inputs first
What changes by ad format
| Search ads | Use search-intent language, short headlines, concrete benefits, and one CTA. Ask for many variants that do not repeat the same meaning. |
|---|---|
| Meta / social ads | Separate primary text, headline, description, and CTA. The opening should be scannable because people see the ad while scrolling. |
| Banner / image ads | Keep on-image text short: main headline, small support line, and CTA. Separate copy guidance from visual direction before sending it to an image tool. |
| Sensitive categories | For health, finance, politics, employment, housing, and education, avoid guaranteed results or targeting assumptions and check platform policies before publishing. |
Copy-ready ChatGPT ad copy prompts
Draft search ad headlines and descriptions
Create Google Search ad copy for this product: “A Chrome extension for AI prompt management. Users can save reusable prompts and insert them into ChatGPT or Gemini input fields. It does not auto-submit.” Audience: freelancers and small teams starting to use AI at work. Provide 10 headlines under 30 characters, 4 descriptions under 90 characters, and one natural CTA. Do not invent numbers, rankings, customer claims, or guaranteed outcomes. End with a list of copy that needs human verification.
The prompt provides facts, placement, field limits, audience, and no-guess rules, so the result is easier to review before entering an ads manager.
Create Meta ad text by field
Using the same product facts, create three Meta ad variations. For each variation, include primary text under 125 characters, headline under 40 characters, description under 25 characters, CTA, and review risks. Use three angles: time saved, reusable prompt library, and manual review before sending. Avoid hype, fear-based urgency, and unverified numbers.
Field-level output prevents one long paragraph from being pasted into every ad field.
Review claim and policy risk before publishing
Review the following ad copy before publication. Return a table with: copy line, possibly too-strong wording, evidence needed, platform-policy check needed, and safer revision. Treat health, finance, employment, housing, politics, and education wording as needing extra review.
This turns the generated copy into a review checklist instead of treating it as ready-to-publish text.
Save reusable ad copy prompt templates in BananaNL
Ad copy prompts often repeat the same rules: channel, limits, CTA, banned claims, and review criteria. Saving templates for search ads, Meta ads, banners, and claim review helps you start from a safer structure instead of retyping the rules.
BananaNL is a Chrome extension that inserts saved prompts into AI Chat input fields such as ChatGPT, Gemini, and Grok. It never auto-submits, so you can review company names, prices, campaign conditions, and ad policies before sending manually. Under the current pricing rules, NotebookLM can be started for free. In AI Chat, prompt viewing, insertion, and saving are paid features, while Image Collections, image saving, image editing, and video conversion are free.
FAQ
Can ChatGPT run my ads automatically?
This guide covers drafting and reviewing ad copy. Campaign setup, submission, bidding, budget changes, and policy responses still need the tools and rules of each advertising platform.
Can I trust the character counts in the draft?
Use them as a starting point, then confirm in the ads manager. Double-width languages such as Japanese, Korean, or Chinese may be counted differently, so asking for shorter copy is safer.
Can ChatGPT write claims for regulated products?
It can draft candidates, but publishing requires human review. Health, finance, employment, housing, political, education, and performance claims should be checked against evidence and platform policies.
If searching for prompts is the hard part, use BananaNL
Prompts become useful when they are close to the input field. Use BananaNL to carry them there, then adjust before sending.