Key takeaways
- ChatGPT ads run as two API creative types,
chat_cardandproduct_ad_template, but render in at least six visible layouts, from a single text card to a scrollable product carousel. - Ads now show up in roughly 20-30% of ChatGPT web-search responses as of mid-2026, up from zero before late May, according to Promptwatch's tracking of ad-bearing responses.
- Product feed carousels, launched August 6, 2026, let OpenAI (not the advertiser) decide whether a response gets one product card or a multi-item carousel.
- Only logged-in adults on ChatGPT's Free and Go plans see ads. Plus, Pro, and Business users don't.
- Google AI Mode has gone further on one dimension: it now places a sponsored product carousel inline, inside the generated answer itself, not just below it, something ChatGPT hasn't done yet.
Why ChatGPT ads look nothing like banner ads
If you're picturing AdSense rectangles bolted onto a chatbot, drop that image. ChatGPT ads are built to look like the thing they're competing with: organic citations. Product pages are already the single most-cited content type in ChatGPT answers, at 22.4% of all citations in late September 2026, with landing pages close behind at 16.8%, according to Promptwatch's citation share by content type data. On commercial and comparison prompts that split tilts even further toward product pages. OpenAI built its ad units to mirror that mix rather than interrupt it, which is why a sponsored product card and an organic one are nearly indistinguishable at a glance, aside from a small "Sponsored" label.
That resemblance is deliberate. A user asking "what's a good waterproof hiking jacket under $200" doesn't want a banner, they want a answer that looks like an answer. OpenAI's ad formats are an attempt to monetize that moment without breaking the illusion.
The three formats you'll actually see
1. The chat card (text ad)
This is the original and still most common format. It's the one advertisers build by hand, API name chat_card. It carries:
- Advertiser name and favicon
- A headline (up to 50 characters)
- A short description line (up to 100 characters)
- A single image
- A landing page link
- A "Sponsored" label
It sits below the ChatGPT response, same position every time. Early sightings from February 2026 showed double chat cards from a single advertiser (two Expedia ads stacked under one answer), which told researchers OpenAI's layout engine was built from the start to stack multiple units, a hint at what came later with carousels.
2. Product feed ads (sponsored product cards)
Launched May 12, 2026, this format pulls from a merchant's product catalog rather than a hand-built creative. It looks closer to an Amazon sponsored listing: product image, title, price, and star rating. OpenAI's technical spec requires at minimum an item_id, title, description, product URL, brand, seller name, and price, and it accepts Google-compatible Shopping feeds directly. During the current beta, feed products are only eligible for paid placements, not organic citations, though OpenAI has said that may change.
3. The product carousel
The newest format, live since August 6, 2026. Multiple products from a single retailer render side by side in one scrollable unit, in the same bottom-of-conversation slot a single ad would normally occupy. Digiday verified the format through screenshots and reported that OpenAI, not the advertiser, decides whether a given auction win renders as a single card or a full carousel.
Right now carousels only pull from one retailer per unit. But the mechanics, according to Digiday's reporting, are already built to support more: multiple products surfaced by relevance, and eventually competing retailers appearing side by side in the same carousel, the way Google Shopping has worked for years. If that cross-retailer version ships, it changes the competitive dynamic entirely, since a user comparing jackets could see three different brands bidding into the same scroll.
A format comparison table
| Format | API type | What it shows | First seen | Who decides layout |
|---|---|---|---|---|
| Chat card | chat_card | Headline, description, image, link | Early 2026 rollout | Advertiser builds it |
| Double chat card | chat_card x2 | Two cards, same advertiser | Feb 23, 2026 | OpenAI's ranking |
| Single product feed ad | product_ad_template | Image, price, rating from catalog | May 12, 2026 | OpenAI |
| Product carousel | product_ad_template | Multiple products, one retailer, scrollable | Aug 6, 2026 | OpenAI |
How often do these ads actually show up?
Ads appeared in an average of 20.1% of ChatGPT web-search responses over a 90-day window from May 20 to August 17, 2026, per Promptwatch's tracking of ChatGPT ad frequency. That average hides a lot of movement: the rate was 0% until May 27, spiked past 30% on July 1, and has bounced between the low 20s and high 30s since, peaking as high as 43.9% on individual days. In the most recent 7-day window the research covered, the rate sat at 32.4%, up 12.5% from the week before.
The prompts carrying ads lean organic. About 73% of ad-bearing responses come from generic, non-branded queries, 18% from brand-specific prompts, and roughly 9% from competitor-comparison prompts. That split matters for advertisers: you're not just buying placement on "best X brand" searches, you're buying into the much larger pool of plain "what's a good X" conversations.
Shopping features specifically (the product card and carousel formats) trigger on a smaller slice of overall responses, but Promptwatch's data on ChatGPT shopping feature usage shows the rate moves in step-changes rather than a slow climb, roughly doubling overnight at points, which points to OpenAI actively tuning how often shopping units appear rather than letting the auction decide on its own.
Pricing and how the auction works
OpenAI runs what it calls a relevance-weighted, second-price auction. For CPC campaigns, OpenAI recommends starting bids of $3-5 per click. CPM campaigns set a max CPM bid instead. Third-party benchmarks put typical 2026 ranges at $25-60 CPM and $3-5 CPC overall, with e-commerce running cheaper ($3-6/click) than software ($8-18) or finance and legal ($15-25), though these figures aren't OpenAI-confirmed.
OpenAI dropped its original $50,000 minimum spend requirement when self-serve Ads Manager opened to the US, and daily minimums now start around $25/day in the US, with minimums set across 30 currencies globally.
Targeting runs on conversation context and intent, the ad's own landing page and copy, and advertiser-set "context hints" at the ad group level (not exact-match keywords). If ad personalization is on, select signals from a user's broader ChatGPT history can factor in too. OpenAI has flagged this as the easiest place for advertisers to mess up: treating context hints like Google Ads keywords. A hint matching "CRM software" can fire even when a user is describing a tool they want to avoid, because ChatGPT conversations are long and contextual rather than discrete search queries.
Who actually sees these ads
Only logged-in adults on ChatGPT's Free and Go plans. Plus, Pro, Business, Enterprise, and Edu accounts don't see ads at all, and neither does any account flagged as under-18. That's a meaningfully narrower audience than it sounds, but ChatGPT reaches 1.2 billion people weekly by OpenAI's own count, so the Free-tier slice alone is enormous.
Brand safety guardrails try to keep ads out of "emotionally vulnerable, sensitive, or otherwise unsuitable" conversations, and advertisers can set negative phrases to exclude specific contexts. OpenAI launched brand-suitability evaluation pilots with DoubleVerify and Integral Ad Science in October 2026 to assess those safeguards independently, without accessing private conversations.
What's coming next
On October 5, 2026, OpenAI announced a new visual ad format appearing during ChatGPT's image-generation flow itself, not just below text answers, testing with US advertisers later that month. The same announcement expanded attribution and measurement partnerships considerably, adding Hightouch, Tealium, and LiveRamp for conversion data, plus AppsFlyer, Triple Whale, Adjust, and several others for attribution. Early performance numbers OpenAI cited: WeightWatchers' attributed CPA on ChatGPT Ads ran 15.3% lower than its blended paid-search benchmark per DV Rockerbox, and Portland Leather reportedly saw 93% of its ChatGPT Ads visitors arrive as new customers per Triple Whale. Those are vendor-reported numbers worth treating as early and self-selected, but they're the first real signal that OpenAI is trying to prove ROI, not just impressions.
How this compares to Google AI Mode and Perplexity
Google moved faster on one specific front: inline ad placement inside the generated answer. Google AI Mode formalized shopping ads on February 11, 2026, and by August 31, 2026 it was showing a sponsored product carousel directly inside the AI-generated response text, between the answer and the follow-up questions, plus a separate standalone ad unit at the bottom of the page. That's the first confirmed instance of an ad rendering inside the answer itself across any major AI search product. ChatGPT's carousels, by contrast, still sit below the response in the same slot a single ad would occupy, not woven into the generated text.
Perplexity went the opposite direction. It tested sponsored follow-up questions (appearing on roughly 40% of queries at one point) and sponsored video, then appears to have walked away from advertising entirely as of February 2026. Reports on whether that's a full shutdown or a pause are mixed, so treat Perplexity's ad status as unsettled rather than confirmed one way or the other.
| Platform | Ad placement | Inline with answer? | Status as of late 2026 |
|---|---|---|---|
| ChatGPT | Below response, chat card or carousel | No | Expanding fast, 20-30%+ of responses |
| Google AI Mode | Below and inside response | Yes, since Aug 31, 2026 | Active, expanding carousel placements |
| Perplexity | Follow-up questions, video (historical) | No | Reportedly paused/abandoned Feb 2026 |
What this means if you're trying to get seen without paying
The organic route still matters more than the paid one for most brands. One analysis found that roughly 83% of products appearing in ChatGPT's product carousel today come from organic Google Shopping feeds, not paid placement, and that products get reshuffled by an average of five positions from their Shopping rank once ChatGPT reorders them. That means feed quality, schema markup, and pricing data still carry more weight than most marketers assume.
If you're trying to understand whether your brand shows up in these citations and carousels at all, before you spend a dollar on ads, tools like Promptwatch track which of your pages get cited across ChatGPT, Google AI Overviews, and the other major AI search surfaces, along with crawler logs showing whether AI bots are even reaching your product pages in the first place.
Common mistakes to avoid
A few patterns show up repeatedly in practitioner reports from 2026:
- Running context hints like exact-match keywords instead of intent signals, which misfires when a user mentions your category while describing what to avoid
- Sending clicks to generic landing pages instead of ones matched to the conversational intent that triggered the ad
- Launching a campaign and never iterating the creative, treating it like a set-and-forget Google Ads group
- Ignoring the FTC's existing disclosure guidance; "Ad," "Sponsored," and "Paid Advertisement" are all recognized as sufficient disclosure language under current native-advertising rules, even though there's no AI-chat-specific FTC rule yet
The bottom line
ChatGPT's ad formats are evolving in a straight line toward looking exactly like what they're surrounded by: product listings and citations. The chat card was the training wheels version. Product feed ads brought in real catalog data. Carousels brought the multi-item browsing experience people already expect from Amazon and Google Shopping. The next real shift, cross-retailer carousels where competitors bid into the same scroll, hasn't shipped yet, but the mechanics are already built for it according to Digiday's reporting. Whichever direction it goes, the gap between "organic citation" and "sponsored placement" inside ChatGPT keeps getting narrower, which is good for advertisers chasing attention and a little uncomfortable for anyone trying to tell the two apart.