A chat card (chat_card) is the single text-and-image ad unit that OpenAI's Ads API renders beneath a relevant ChatGPT answer, carrying a title, body copy, an image, a favicon, and a destination URL. It is the advertiser-written format you build by hand, separate from the product_ad_template that OpenAI assembles automatically from a merchant feed.
If your buyers ask ChatGPT purchase questions, the chat card is the one paid slot where your brand can answer them at the moment of intent. Skip it and a competitor's card can occupy that slot below the exact answer your customer just read.
The chat_card is the baseline creative type in OpenAI's advertising system, the object you create under an ad group when you want a native paid unit inside ChatGPT. Every chat card you launch shares the same shape: an advertiser name, a favicon, a short title, a line of body copy, a square image, and a UTM-compatible landing URL.
OpenAI's Ads API caps the title at 3 to 50 characters and the body at 100 characters, and recommends far shorter copy because the card visually truncates well before those limits. You upload the image first, receive a file_id, then attach it to the ad, and a chat card will not serve without both a target URL and that image reference. Delivery is contextual, so OpenAI's relevance model decides which conversations your card fits, reading your context hints and the card's own wording.
A chat card sits one level below the campaign and ad group in the account hierarchy, and it differs from the product_ad_template, which OpenAI builds for you from a catalog feed. For marketers planning where their brand should appear across AI answers, AirOps helps map the questions and conversations worth targeting before a single card goes live.
A chat card moves from a draft object to a served impression through OpenAI's Ads API and relevance system. The steps run in a fixed order, and the card cannot deliver until each one is complete.
Upload creative. Send your image to the API and store the returned file_id, which every chat card must reference before it can run.
Build the ad. Create the chat_card under an ad group with its title, body copy, target URL, and that file_id; new ads default to paused.
Set delivery. Attach context hints and bidding at the ad group and campaign level so OpenAI knows which conversations and what price to target.
Match to conversation. OpenAI's relevance model reads your hints and the card's wording, then places the card below answers it judges a fit, with no keywords involved.
Serve and measure. The card renders as a labeled Sponsored block beneath the answer, and the API reports impressions, clicks, and spend back to you.
Those metrics tell you how often your card won a relevant answer and what each click cost. They do not tell you which specific conversation triggered the match, because OpenAI keeps the underlying chat private.
Deciding whether to run chat cards is really a decision about whether you want a paid presence inside the tool your buyers increasingly use to make choices. The format is narrow, and the business case rests on buying intent more than on raw reach.
Intent at the decision point. A chat card can appear right after ChatGPT answers a buying question, putting your brand in front of someone who is actively comparing options at the moment they decide, well before a retargeting pool would reach them.
A contested slot. ChatGPT caps paid units tightly on any answer, so ignoring the format lets a competitor claim the single card below an answer your prospect just read and trusted, and you never see that lost impression.
Measurable spend you control. Organic AI visibility is hard to steer, while a chat card gives you direct control over the title, image, landing page, and budget, with clicks and conversions you can tag and track through your own UTMs.
Growth marketers use chat cards to capture high-intent clicks from ChatGPT users asking product-comparison questions and route them to a tightly matched landing page.
SEO managers use chat cards to claim paid presence on AI answers where their organic pages have not yet earned a citation.
Demand gen leads use chat cards to test ChatGPT as a new acquisition channel with a small budget before committing to the ad-tech partner path.
The card renders small below the answer and visually cuts your title and body well before the API's hard limits, which means the opening words of each field carry the entire message for most readers and everything past them is unlikely to be seen.
Context hints are the short descriptive phrases you attach to an ad group to tell OpenAI's relevance model which conversations your card belongs in, and because the system runs no keyword targeting, these hints plus your creative wording are effectively the targeting.
A chat card references its image through a file_id returned when you upload the creative, so the image is a mandatory separate step, and the ad can neither be created nor served until both that file_id and a destination URL are attached.
Reach buyers inside ChatGPT at the moment they ask a purchase question.
Control every creative element, from the title and image to the landing URL.
Launch with no platform minimum through OpenAI's self-serve Ads Manager.
Track clicks, spend, and conversions directly through the Ads API and UTM tags.
Front-load a short title that can serve the instant your card wins a relevant answer.
Write the title for its opening characters, since the card truncates and the first words do the work.
Match the landing page to the card's promise so the click resolves to exactly what you offered.
Attach specific context hints, three to six words each, so OpenAI places the card in genuinely relevant conversations.
Tag every card with a unique UTM so you can separate chat card traffic from other channels in analytics.
Start on a clicks bid, then move to conversions once your tracking is live and returning data.
Keep the image readable at thumbnail size, because the unit renders small below the answer.
Avoid treating the chat card like a display banner stuffed with brand slogans. Competent marketers waste the slot by writing a headline that shouts instead of answering the question the user just asked, and the card loses to one that reads like the obvious next step.
AirOps: Maps the AI-search questions and conversations your brand should target, so your chat card context hints and creative angles are grounded in real demand before you spend.
OpenAI Ads Manager: The native self-serve console and Advertiser API where you build, upload, and launch chat_card ads and pull delivery metrics.
Google Analytics 4: Measures the UTM-tagged clicks and conversions your chat cards drive once users land on your site.
Audit intent. List the buying questions your customers already ask ChatGPT and note which ones your brand should answer. You can do this in an afternoon with no budget approval.
Request access. Apply for the OpenAI Ads Manager beta and create an ad account, which gives you the API key you will need to build anything.
Prepare creative. Write a title that lands in its opening characters, draft body copy under the truncation point, and produce a square image sized for a small card.
Upload and build. Upload the image to get a file_id, then create a paused chat_card under an ad group with your title, body, target URL, and hints.
Launch and read results. Set a clicks bid, turn the card live, and watch impressions, clicks, and cost per click before you scale or rewrite.
A chat card (chat_card) is the advertiser-written text-and-image ad unit that OpenAI renders below a relevant ChatGPT answer.
You build it through OpenAI's Ads Manager or Advertiser API by uploading an image, attaching a title, body, and URL, then bidding on clicks or conversions.
The format is tight: a short title and short body that both truncate on screen, with no video or carousel in self-serve.
The main risk is a competitor claiming the single paid slot below an answer your prospect just read while you sit out the channel.
The leverage is intent, since the card reaches people at the exact moment they ask ChatGPT a buying question.
A chat card is the ad you write yourself, while a product_ad_template is the creative OpenAI assembles from a product catalog you connect. With a chat card you control the title, body copy, image, and landing URL, and you create it under an ad group through the Ads API. The product_ad_template works differently: you supply a merchant feed, and OpenAI generates the layout, pulling in fields like product image, title, and price automatically based on conversation signals. Practically, the chat card suits any advertiser with a single offer or message to promote, from software to services, because it does not depend on a structured catalog. The product template suits retailers with many SKUs who want OpenAI to match individual products to shopping questions. Both run inside the same account hierarchy and both appear as labeled sponsored units below a ChatGPT answer, but only the chat card gives you line-by-line authorship of what the unit says.
You can start a chat card with no platform minimum through OpenAI's self-serve Ads Manager, which is the key difference from the ad-tech partner path that has carried a steep pilot commitment. Spend is set by your bid and budget, and OpenAI supports both cost-per-click and cost-per-thousand-impression bidding, so you decide how much each result is worth before anything serves. A sensible way in is to fund a small test budget, launch on a clicks bid, and let the numbers tell you whether the channel converts for your offer. Because the system has no keyword auction you can research in advance, early spend is partly discovery: you are learning which conversations your card wins and at what cost per click. Once conversion tracking is live, you can move to a conversions bid and let OpenAI optimize toward the action you actually care about. Treat the first weeks as calibration before scaling.
Your chat card appears on some answers and skips others because delivery is contextual, and OpenAI's relevance model decides match by match whether your card fits the conversation. There is no keyword list that guarantees placement; the system reads your context hints and the card's own title and body, then judges relevance to each specific answer in real time. Variation also comes from audience eligibility, since ads show only to signed-in adults on the Free and Go tiers, and paid tiers see no ads at all, which shrinks the pool of answers where your card can ever appear. Competition matters too: when more than one advertiser fits an answer, OpenAI favors the most relevant card, so a stronger match can crowd yours out. Finally, commercial intent is required, so purely informational questions with no buying signal will not surface any card. The practical takeaway is that sharper hints and tighter creative widen the set of answers you qualify for.
You can influence it, but you cannot target it the way you would in a keyword platform. The direct levers are your context hints, a handful of short descriptive phrases per ad group, and the wording of the card's title and body, which the relevance model reads as part of the signal. Writing specific, situation-led hints and creative tends to place you in tighter, more qualified conversations, while vague inputs scatter your card or suppress it. You also control geography and audience lists, and you choose the bid that decides how aggressively you compete for eligible answers. What you cannot do is name the exact questions or pick individual conversations, because OpenAI withholds the underlying chat for privacy and makes the match itself. So the honest answer is partial control: shape the inputs well and you steer delivery meaningfully, but the final placement decision stays with OpenAI's system.
A good chat card title lands in roughly the first 16 to 24 characters, which is also what OpenAI recommends, even though the API technically allows a longer string. The reason is truncation: the card renders small below the answer and visually cuts the title well before the hard limit, so anything past the first couple of dozen characters is unlikely to be read. Treat the opening words as the whole headline and put the offer or the hook there, and leave brand names or throat-clearing phrases out. Body copy follows the same logic, with a recommended range well under its maximum. Beyond length, a good title reads like the natural next step after the answer the user just received, phrased as an invitation to click. If you are unsure, write three distinct titles, keep each front-loaded, and let click data decide which length and angle actually perform for your offer.