Understand the official card fields and creative requirements, then compare your copy with character-length distributions computed from real captured ChatGPT ads.
OpenAI currently documents a single sponsored unit with six core fields. Format availability can evolve during the beta.
This real captured example shows how the documented parts appear together. Open it for the associated advertiser, niche, and prompts.
These figures describe canonical creatives in the library. They are not official maximums, recommendations, or validation rules.
| Field | Average | Middle 50% | Observed range |
|---|---|---|---|
| Title | 29.7 | 22 to 36 | 4 to 236 |
| Description | 70.5 | 46 to 86 | 7 to 5000 |
These comparisons describe captured creatives. They are not official character-limit validation.
The questions people actually ask.
OpenAI's current format documentation lists advertiser name, favicon or logo, title, copy or description, landing page, and an image creative. The unit is clearly labeled and appears separately below a relevant response.
OpenAI's public creative guide recommends concise titles but does not state a numeric maximum on the public page. In 16,639 captured creatives with titles, the observed median is 28 characters and the range is 4 to 236. Observed values are not official limits.
Across 16,610 captured descriptions, the observed median is 63 characters, the average is 70.5, and the observed range is 7 to 5000. These are descriptive statistics, not a platform limit.
OpenAI's public best-practices page says images should be simple, relevant, easy to interpret, and aligned with the title and copy. It does not currently publish a numeric image dimension or file-size requirement on that page.
OpenAI says the destination should be relevant, valid, reachable, consistent with the creative, and accessible to OAI-AdsBot and OAI-SearchBot. Advertisers can add their own UTM tracking parameters.
No. OpenAI reviews the advertiser, creative, image, and landing page for policy compliance. Misleading, inconsistent, inaccessible, or disallowed content can be rejected.
Browse captured titles, descriptions, images, advertisers, and prompts.
Questions about the data, a brand you expected to see, partnerships, or access to the intelligence layer. We read every message.
