Writing chat card creative for ChatGPT ads

Published 17 September 2026 · Updated 17 September 2026 · Based on the OpenAI Ads API documentation

A chat card has exactly four parts: a title, a body, an image and a destination URL, each with a hard limit that OpenAI enforces on the object itself. There is no video and no sitelinks to fall back on, so those four parts have to do the whole job. This guide sets out the exact limits, what happens in review, how a product ad template differs, and worked examples, all using the fictional shops Oostloop, Wielwerk and Twaalf Bonen as illustrations, never a real brand.

What are the exact chat card limits?

Title: 3 to 50 characters. Body: up to 100 characters. Image: JPEG, PNG or WebP, at least 640 by 640 pixels. Destination URL: up to 2,048 characters, and it has to be a page OpenAI's crawler can actually reach, not one blocked by robots rules or a login wall. All four sit on the ad object itself and are checked when the ad is created.

Ads Manager's own "Create Ad" form repeats the title and body limits exactly, but describes the image requirement more loosely, suggesting a square image "at least 256 by 256 pixels" for best quality, with a note that copy may be truncated in some placements. Where the two disagree, build to the API's stricter 640 by 640 floor: it is the number in the developer documentation, and a smaller image risks looking soft wherever the card renders larger than the UI's own minimum implies.

What happens in review, and what sends a card back into it?

Every chat card and product ad template is created with a review_status of in_review, then moves to approved or rejected with a reason. Editing the creative, title, body, image or destination URL, re-triggers review from scratch, so a live card cannot be nudged without that edit waiting for review again. There are also two account-level reviews, a brand review and an account integrity review, which apply to the account rather than to each ad.

Because any edit reopens review, plan chat card changes as replacements, not touch-ups: write the new version, create it as a new ad, let it clear review, then archive the old one, which cannot be undone. A preview, valid for 24 hours, shows how the card renders but is not the same as a review pass and does not confirm the ad is eligible to serve.

How does a product ad template work?

A product ad template is the second, and only other, creative type. It reads its image, its destination URL and its price straight from a feed item, so there is no image to upload or price to type by hand. The template can print the live price using a placeholder, {{product.price}}, so a price change pushed into the feed shows in the ad without editing the creative.

A feed-based ad group carries at most one non-archived product ad template. Because the price is inserted at serve time, count the placeholder toward the character limit using the longest price you expect to show, including the currency symbol or code, not the shortest one; a template that fits at EUR 9.00 can overflow at EUR 1,299.00.

Product ad template, illustration only (Twaalf Bonen, a fictional coffee roaster): title Twaalf Bonen house blend, price line rendered as Now {{product.price}}, image and destination URL supplied by the feed item. Nothing here is a real product, price or brand.

What angles make a chat card worth clicking?

The card has to earn its place in a conversation, so it reads best when it answers the situation the ad group's context hints describe, not when it lists product features. A body that states a concrete, checkable detail, a delivery promise, a guarantee, a specific service, tends to read better within 100 characters than a body that tries to summarise the whole offer.

Three checks before an ad goes to review. Does the title alone make sense without the body, since Ads Manager warns that copy may be truncated in some placements? Does the body add one new fact rather than repeating the title? And does the destination URL land on the exact page the card promises, not a category page or the homepage? A card that fails the third check can pass review and still under-perform.

What do worked chat card examples look like?

The three examples below are illustrations built for this guide using fictional shops, Oostloop (a running shop), Wielwerk (a bike repair shop) and Twaalf Bonen (a coffee roaster). None of them names a real business, and none touches health, financial or legal claims, categories that reportedly face extra review.

• Oostloop, illustration: title Oostloop running shoes (22 characters); body Free gait check in Utrecht and Amsterdam, plus a 90-day exchange if the shoe is not right. (90 characters). The body states two concrete, checkable facts rather than "great shoes for every runner".
• Wielwerk, illustration: title Wielwerk e-bike repair (22 characters); body Same-week motor and battery diagnostics for e-bikes, drop off in Utrecht or ask about pickup. (93 characters). This targets a service outcome and a location rather than a generic "bike repairs".
• Twaalf Bonen, illustration: title Twaalf Bonen coffee (19 characters); body Small-batch beans roasted weekly, ground to order for filter, espresso or French press. (87 characters). The body names the format options a buyer actually decides between.

Each of these pairs a title that survives on its own with a body that adds one new, specific fact, and each stays comfortably inside the 50 and 100 character ceilings with room for a slightly longer variant if the first version under-performs.

How should I structure creative for measurement?

Because there is no query report and no rotation control across multiple cards in one placement, the ad group is the practical unit for measuring an angle, since its hints and creative are reported together. Put one chat card, or one small set of close variants, in an ad group built around one context and one set of hints, rather than a single ad group carrying every angle you can think of. That mirrors the advice in the context hints guide: one narrative, one ad group, one measurable result.

When a new version replaces an approved card, keep the old ad group's history rather than editing the ad in place if you want a before-and-after comparison; a fresh ad within the same ad group inherits the group's hints and bid but starts its own review and its own reporting row. Building the whole creative step into a first campaign, in order, is covered in launching your first ChatGPT ads campaign.

Frequently asked questions