Product card impressions are a ChatGPT Ads metric that counts how many times an individual product card in a multi-product carousel becomes viewable to a user. They are tracked separately from billable ad impressions, which count the whole sponsored unit instead of each product inside it.
If you run product feed campaigns, these counts tell you which catalog items actually earn attention inside the carousel, so you can decide which products and feed attributes deserve investment. Ignore them and you judge performance by the unit alone, missing that products in later carousel slots are often served but rarely seen.
A product card impression records a single viewable render of one product tile within a ChatGPT Ads carousel, logged the moment that card enters the user's view. OpenAI reports these counts in Ads Manager and exposes product-level fields through its public Insights API, giving advertisers attention data at the level of each item in a feed campaign. OpenAI makes complete carousel card data available beginning August 20, 2026, on a rolling 30-day basis.
For the metric to register, a product feed campaign has to serve in the multi-product carousel format, and an individual card has to become viewable on screen. Because OpenAI decides whether a single-card unit or a carousel appears, card impressions only accrue when the system chooses the carousel and the shopper scrolls far enough to expose each tile.
This sits one layer below the billable ad impression, which counts the sponsored placement as a whole. Card impressions are diagnostic signals you are not charged for, so they read as a free view into product-level attention within paid placements. AirOps helps teams connect these feed-campaign signals to the broader picture of how AI engines surface their products.
Product card impressions accumulate through a fixed sequence, from feed upload to the viewability count that lands in your report.
Upload feed. You send a structured product catalog to ChatGPT Ads Manager over SFTP, and the system generates product ads from your titles, prices, images, and availability.
Match the moment. OpenAI matches eligible products to a conversation's purchase intent and decides whether to show a single card or a multi-product carousel.
Render carousel. When the carousel serves, several items from your catalog occupy one sponsored placement below the answer.
Count viewability. Each time an individual card becomes viewable on screen, the system logs one product card impression, kept separate from the billable ad impression.
Report and expose. Ads Manager reports card-level impressions and clicks, and product-level fields flow through the public Insights API.
The count tells you which products in a carousel actually reached a shopper's eyes and in what volume. It stays silent on whether the shopper clicked, converted, or even registered the product, so pair it with click and conversion data before you judge a product's pull.
Product card impressions change how you allocate budget and feed effort across a catalog, because they show attention at the item level that billing never reveals. For a buyer deciding where to concentrate spend, that item-level view is the difference between funding products shoppers see and funding products the carousel buries.
They expose weak carousel slots. Products placed in later positions can be served constantly yet rack up almost no card impressions, and treating the billable unit as one number hides that dead weight.
They separate attention from cost. Because you are never billed for a card impression, the metric gives you an attention signal that carries no media penalty when you read it.
They tie feed quality to visibility. A card with a missing attribute or a stale title can lose eligibility, and watching card impressions by product surfaces which feed fields are costing you reach.
SEO managers use product card impressions to spot which catalog items draw attention inside ChatGPT carousels and prioritize those pages for feed and content work.
Growth marketers use product card impressions to reallocate feed campaign budget toward products that earn views and away from items the carousel rarely surfaces.
Demand gen leads use product card impressions to brief merchandising teams on which products need better titles, images, or attributes to stay eligible.
A product card impression only fires when a card actually becomes viewable on screen, so a product that serves in the carousel but never scrolls into view records no impression. This is why serving volume and card impressions can diverge sharply for the same product.
The billable ad impression counts the whole sponsored unit and drives your cost, while the card impression counts each product tile and stays out of billing entirely. Reading them together shows how much of the unit's paid exposure each item is actually capturing.
Product-level data flows through the public Insights API, so pulling card impressions by item means working with feed identifiers and the fields OpenAI exposes for each product. Clean, consistent identifiers in the feed are what make that item-level reporting usable.
Reveal which products in a ChatGPT carousel earn real viewability.
Diagnose feed problems by spotting items with high serving but low card impressions.
Separate attention from spend, since card impressions carry no billing.
Pull product-level attention data through OpenAI's public Insights API.
Prioritize budget toward catalog items that consistently draw views.
Benchmark carousel positions to see where attention drops off across your catalog.
Read card impressions beside clicks. Pair each product's views with its clicks so you can tell genuine interest from mere exposure.
Segment by carousel position. Group card impressions by slot to learn where attention falls off, because later positions often go unseen.
Keep feed attributes complete. Fill titles, images, prices, and availability for every item, since missing fields can drop a product out of eligibility.
Pull item-level data from the Insights API. Use the public Insights API to analyze card impressions per product instead of eyeballing the Ads Manager table.
Refresh stale catalog entries. Update outdated titles and images regularly so products stay eligible and keep earning views.
Compare against billable impressions. Track the gap between the unit's billable impressions and each card's impressions to see how much paid exposure every item captures.
Avoid judging a product feed campaign by billable ad impressions alone. Competent marketers do this because the unit-level number looks clean, but it masks products that serve often and get seen rarely, and it leaves budget flowing to items the carousel quietly buries.
AirOps: connects your ChatGPT Ads feed-campaign signals, including product card impressions, to a broader view of how AI engines surface and cite your products.
ChatGPT Ads Manager: OpenAI's native console where card-level impressions and clicks for carousel campaigns are reported, with CSV export and the public Insights API.
Google Merchant Center: manages the structured product catalog that feeds campaigns, so the titles, images, and attributes driving card eligibility stay accurate.
Audit your product feed. This week, review your existing catalog file for complete titles, images, prices, and availability on every item you want eligible. No budget approval needed.
Launch a product feed campaign. Set up a feed campaign in ChatGPT Ads Manager and upload your catalog over SFTP so products can serve in the carousel.
Confirm carousel serving. Check Ads Manager to verify your campaign is serving in the multi-product carousel format, since card impressions only accrue there. If OpenAI serves a single-card unit instead, the metric will not populate.
Pull card-level reporting. Export card impressions by product from Ads Manager and the public Insights API to see which items earn views. Segment the pull by carousel position so you can see where attention drops.
Act on the gaps. Reallocate budget toward products with strong card impressions and fix feed attributes for items that serve but go unseen.
Product card impressions count how often each product tile in a ChatGPT Ads carousel becomes viewable to a shopper inside the conversation.
They register per card on viewability and are reported in Ads Manager and the public Insights API.
The metric only appears when OpenAI serves the multi-product carousel instead of the single-card unit.
Relying on billable ad impressions alone hides products that serve often yet are rarely seen, which skews budget decisions.
The leverage sits in feed quality: complete, accurate attributes keep products eligible and earning views.
Product card impressions and billable ad impressions measure different things on the same placement. A billable ad impression is the count OpenAI charges you for when your sponsored unit appears below a ChatGPT answer. A product card impression is a non-billed count that fires separately each time one individual product tile inside a carousel becomes viewable to the shopper. One carousel placement produces a single billable impression but can produce several card impressions, one for every tile that scrolls into view. That difference matters because the billable number tells you the unit was shown, while the card numbers tell you how far the shopper engaged with the products inside it. If you only watch billing, you treat a carousel as one event and lose sight of which items actually captured attention. Watching both numbers gives you the exposure you paid for alongside the product-level attention it produced.
Product card impressions update frequently in Ads Manager, though not in true real time. OpenAI's reporting for impressions, clicks, and click-through rate generally appears quickly, often within minutes of activity, so you can watch campaigns through the day instead of waiting for an overnight batch. Spend figures can lag behind those counts, which means a momentary zero in the spend column does not prove that no charges have accrued. For carousel card impressions specifically, OpenAI makes complete reporting available on a rolling window, so the freshest hours are still settling while older data is final. The practical takeaway: check card impressions often enough to catch trends, but make budget and feed decisions on data that has had time to settle instead of on the last few minutes. Treat the newest numbers as directional and the stabilized window as your source of truth for reporting.
Product card impressions vary between products in the same carousel mainly because of position and relevance. OpenAI decides the order of tiles based on how well each item matches the conversation's intent, and tiles placed earlier in the carousel are far more likely to become viewable than ones a shopper has to scroll to reach. A product that is technically eligible can still sit in a low slot and collect almost no card impressions. Relevance also shifts by conversation, so the same product can lead the carousel in one context and trail in another. Feed quality feeds into this too: a weak title, a missing image, or an out-of-stock flag can suppress how often an item is chosen and shown at all. Because matching is contextual and re-decided each time, you should expect card impressions to form a distribution across your catalog, with a handful of items capturing the bulk of attention and a long tail capturing very little.
You cannot set product card impressions directly, but you can influence them through the inputs OpenAI uses to serve carousels. You do not control which format appears or which slot a product lands in, because the system makes those calls from conversation context, relevance, your bid, and the landing page. What you do control is the raw material: a clean, complete feed with accurate titles, images, prices, and availability gives OpenAI more reason to consider and show your products. Strong bids and well-chosen context hints raise the odds your feed campaign serves in relevant moments. Fixing out-of-stock flags and thin product data removes the obvious reasons an item gets suppressed. So the honest answer is that card impressions are an outcome you steer indirectly. Treat them as a scoreboard that responds to feed quality, bidding, and relevance work, and improve the inputs instead of expecting a dial that sets the number.
There is no published industry benchmark for product card impressions, so good is best defined against your own catalog instead of an external number. The channel is new, OpenAI has not released standard figures, and card impressions depend heavily on how often your products match relevant conversations, which varies by vertical. A more useful read is internal and relative: look at the ratio of card impressions to the billable impressions of the units they appear in, and watch how that ratio trends as you improve the feed. A healthy pattern is a rising share of your catalog earning meaningful card impressions over time and your priority products consistently landing in viewable slots. A warning sign is a tiny set of items absorbing nearly all impressions while the rest stay invisible. Judge progress by movement in your own numbers week over week, and by whether the products you most want seen are the ones actually accumulating card impressions.