ChatGPT ads creative best practices: 30-char headlines, 1:1 images, and the mistake most advertisers make
ChatGPT ads work on a strict creative spec: a 30-character headline, 60-character body, and a 256×256 pixel square image. Inside those limits, the ad that wins is the one that reads like a useful reply — specific, benefit-forward, and matched to the conversation your context hints describe. The biggest mistake advertisers make is importing the Google Search or Meta voice. This guide walks through the specs, four rules for the headline, an actual before-and-after rewrite, and the coverage strategy that lowers your clearing price.
The short version
- Specs: 30-char headline, 60-char body, 256×256 square image, one destination URL.
- Aim shorter: OpenAI's latest guidance says aim for ~16 characters in the headline and ~32 in the description — the limits are ceilings, not targets.
- Voice: complete the user's thought. Write the sentence a helpful expert would say next in the conversation.
- Coverage: five to eight distinct variations per ad group, each with a genuinely different angle.
- Match: landing page answers the exact question the headline implies, in the first paragraph.
- Relevance is a discount. The auction is second-price and relevance-weighted — better creative usually pays less per click.
The short answer
OpenAI's creative unit is small on purpose. Ads render below the model's response, and the format has to disappear into the conversation rather than dominate the screen. That constrains every choice you make: what fits in 30 characters, what an image at thumbnail size can communicate, and what promise a first-paragraph landing page can deliver. The advertisers who treat those constraints as guardrails — not as a Google Ads unit reformatted — outperform the ones who fight them.
The creative specs, in one place
| Element | Spec | What matters |
|---|---|---|
| Headline | Up to 30 characters — aim for ~16 | Complete the user's thought; specific, concrete, benefit-forward. |
| Body | Up to 60 characters — aim for ~32 | Add the missing detail the headline hinted at — price, format, or qualifier. |
| Image | 256×256 pixels, 1:1 square | Recognizable at thumbnail size; single subject, high contrast, no small text. |
| Destination URL | One URL per ad | Must answer the exact question the headline implies, in the first paragraph. |
| Context hints | Ad-group level, short phrases | Describe the conversations where the ad is relevant — in the user's words, not the marketer's. |
Two of these — the image and the destination URL — are settled once and rarely rewarded for iteration. The three that move performance are the headline, the body, and the context hints. Spend your creative time there.
Writing the 30-character headline: four rules
Thirty characters is the ceiling, and OpenAI's own creative guidance says to aim for roughly sixteen — see our ad anatomy guide for the element-by-element breakdown. There is no room for warm-up, no room for a value-add adjective, and no room for the phrase "the leading." What fits is a specific answer to a specific question. Four rules do most of the work.
- Complete the user's thought. If the conversation is "best CRM for a 10-person team," a good headline finishes that sentence — "CRM for 10-person teams" — not restarts it with "Revolutionize your sales stack."
- Lead with the noun. Ads that start with the thing being sold outperform ads that start with the promise. "CRM for founders" beats "Finally, a CRM that."
- Name a number. Prices, seat counts, plan names, and timeframes all cost characters and buy trust. "Starts at $29 / seat" is more useful than "Affordable pricing."
- Cut every abstraction. "Revolutionize," "unlock," "transform," "next-generation," "AI-powered" — none of these survive the 30-character constraint or the relevance auction. Delete them and use the space for a fact.
Writing the 60-character body: fill the gap the headline made
Sixty characters is the ceiling; OpenAI's guidance says to aim for roughly thirty-two. The body's job is not to sell twice — it is to add the one detail the headline had to leave out. If the headline names the product, the body qualifies who it's for; if the headline names the audience, the body qualifies the outcome. Two patterns work reliably:
- Product + qualifier. "CRM for 10-person teams" (headline) + "No admin, no per-user fees, live in 5 minutes" (body).
- Audience + outcome. "For seed-stage founders" (headline) + "Track every deal, forecast in one click, from $29" (body).
The body is also where you insert the concrete qualifier that filters non-buyers before the click: seat count, industry, price band, or feature. A click that self-disqualifies before landing is a click you did not pay for.
The 1:1 image: a recognition device, not a hero visual
256×256 pixels is small — smaller than a browser favicon on many screens. Treat the image as a recognition mark, not as a hero photograph or a screenshot. High contrast, one subject, no small text. A clean logo on a solid background outperforms a busy product screenshot at that size. If you must use imagery, a single object shot on a plain field will read; a scene will not.
Two moves lift results without touching the creative brief:
- Ship a light and a dark variant. The ChatGPT surface renders both, and ads that look right in one theme frequently disappear in the other.
- Match the ad image to the landing page's hero visual. Continuity between the small square and the page above the fold reinforces the promise-payoff match that the auction scores.
Before and after: a rewrite of a typical bad ad
A concrete example, using a mid-market B2B SaaS ad the shape most accounts default to.
Before — imported Google Search voice:
- Headline (28 chars): "Revolutionize Your Pipeline"
- Body (52 chars): "AI-powered CRM built for the modern sales team"
- Landing page: homepage, "The category-defining CRM"
Nothing in that ad answers a question. "Revolutionize" is a hedge, "AI-powered" is table stakes, and "the modern sales team" describes no one. In a conversation the user is having about picking a CRM for a specific team size, the ad has nothing to say.
After — completes the user's thought:
- Headline (26 chars): "CRM for 10-person teams"
- Body (58 chars): "No admin, no per-user fees, live in five minutes."
- Landing page:
/for/ten-person-teamswith a lede "CRM for 10-person teams. From $29 per seat."
Every element is now testable, matches a real conversation, and can be truthfully answered on the destination page. The auction will score this higher against context hints such as "small team looking for a lightweight CRM" or "cheap CRM for a startup," and — because the auction is second-price and relevance-weighted — the effective CPC will typically drop even at the same max bid.
Coverage: build five to eight variations, not two
One of the durable lessons from the pilot: two carefully polished ads underperform six adequate ads with genuinely different angles. Conversations are heterogeneous, and the auction rewards you for having something to say in more of them. Aim for five to eight variations per ad group, each varying by:
- Angle: price, use case, audience, integration, outcome, or feature.
- Qualifier: team size, industry, seniority, or spend level.
- Proof shape: price anchor, timeframe anchor, or comparison anchor.
Do not vary punctuation or word order and call it a variation. If two ads could be summarized in the same sentence, they are the same ad. Kill the duplicates and use the ad-group slot for a real alternative.
Landing-page match: the second half of the ad
The relevance-weighted auction scores four inputs together: your context hints, ad title, ad copy, and landing page. Getting three right and skimping on the fourth costs you materially. The rule that concentrates the effort: your landing page must answer the exact question your headline implies, in its first paragraph, in the same phrasing. If your headline says "CRM for 10-person teams," the page's first line should not be "Welcome to Acme, the category-defining sales platform." It should be "Acme is a CRM for 10-person teams." Everything else follows.
For most accounts, this means building small dedicated pages for the top three or five conversation types, rather than routing every ad to the homepage. The lift is usually larger than any single copy iteration.
What review will reject, and what will slip through and hurt you
Ad policy on OpenAI Ads Manager broadly follows the shape of established platforms — no misleading claims, no unverifiable superlatives, no unsupported medical or financial guarantees, and no competitor trademarks used to trigger placement. A few edges worth naming:
- Testimonials as headlines are usually rejected as unverifiable. Move the quote to the landing page.
- Comparative "better than X" copy is either rejected in review or approved and then rejected on relevance — the auction penalizes ads that name competitors in the creative.
- Regulated categories (health, finance, legal, alcohol, political) have surface-specific restrictions that shift; verify against current OpenAI policy before writing to those verticals.
- All-caps or heavy punctuation passes review but underperforms; treat it as self-imposed cost.
Frequently asked questions
What are the ChatGPT ads creative specs?
OpenAI's Ads Manager accepts a headline of up to 30 characters, body copy of up to 60 characters, and a square 1:1 image asset at 256×256 pixels — with official guidance to aim for roughly 16 characters in the headline and 32 in the description. Every ad also requires a destination URL and, at the ad-group level, a set of context hints describing the conversations where the ad is relevant.
How many ad variations should I run per ad group?
Build for coverage. Ship five to eight distinct headline and body combinations per ad group, each introducing a different angle rather than rewording the same idea. The relevance-weighted auction favors variations that match a wider set of conversation types, and the winning creatives usually cost less per click than the ones you thought would win.
What is the single biggest mistake advertisers make with ChatGPT ads creative?
Importing their Google Search or Meta ad voice. Google Search rewards keyword-stuffed benefit language; Meta rewards pattern-interrupt hooks. ChatGPT rewards specific, useful, conversational answers that complete the user's thought. Vague benefit lines like "Revolutionize Your Sales Pipeline Today" get skipped in favor of specific answers like "CRM for 10-person teams — from $29 per seat."
Does the ChatGPT ad image really need to be 1:1?
Yes. OpenAI requires a square 256×256 pixel image asset. Ads render below the answer at small size, so the image is a recognition device rather than a hero visual. Keep it clean, high-contrast, and readable at thumbnail size — logo marks, a single product hero, or a bold single-word visual work; screenshots and dense compositions do not.
Should my landing page match my ad copy?
Yes, and matching goes further than in traditional PPC. Because the relevance-weighted auction scores your landing page as one of four inputs, promise-payoff mismatch costs you twice: it tanks conversion and it raises what you pay per click. The landing page should answer the exact question the ad headline implies, in its first paragraph, with the same phrasing.
Can I use emojis, quotes, or brand names in the headline?
Emojis are allowed but count toward the 30-character limit and often reduce clarity at small render size. Direct quotes and testimonials in headlines are typically rejected in review as unverifiable claims. Brand names are fine for your own brand; competitor brand names in headlines are usually disallowed as trademark infringement, mirroring Google Ads policy.
How long does creative review take in the Ads Manager?
Account verification is manual and typically takes a few business days on first setup. Individual ad review is faster, usually within 24 hours of submission once the account is approved. Plan a two- to four-day buffer for the first campaign and a one-day buffer for iterations.
Should context hints match the ad headline word-for-word?
No. Context hints describe the conversations where your ad is relevant, in the user's language, while the headline addresses the user directly with your offer. Overlap in intent is expected, but writing them identically wastes the coverage that separate inputs provide. For structured guidance, see our writing Context Hints guide.
Sources and further reading
- OpenAI Help Center — Create ads for ChatGPT (specs and creative guidance).
- OpenAI Help Center — Ads in ChatGPT: the basics.
- Just Global — ChatGPT Ads best practices guide & B2B playbook.
- StackAdapt — How to advertise on ChatGPT.
- Context Hints — writing Context Hints that convert, Ads Manager setup, and how ChatGPT ads work.
Want your ad creative reviewed?
30 minutes with Tarun. Bring your top ad group and we will rewrite two headlines, one body line, and one landing-page lede to move relevance — and clearing price — in the same session.
Book a discovery call