Key takeaways
- ChatGPT shopping ads and organic product cards look similar but come from completely different systems: ads are labeled "sponsored" and sit below the answer, while organic product cards are generated from web crawl data or merchant feeds and are, per OpenAI, unsponsored and unbuyable.
- Ads only started appearing on May 27, 2026, and have climbed fast. Promptwatch's data shows ads in roughly 20-32% of search-powered responses by mid-to-late 2026, up from zero earlier that spring.
- You will never see a ChatGPT ad on a Plus, Pro, Business, Enterprise, or Edu account. If you're auditing from a paid seat, you're checking the wrong account entirely.
- Organic product cards are won through feed quality, review counts, and GPT tag presence, not through bidding. Profound's research found cards with a "GPT tag" rank 144% better than those without one.
- Tracking both channels well requires separating them in your reporting. Treat ads as a distinct, bid-driven line item and organic cards as a content/feed optimization problem, because mixing the two in one dashboard hides what's actually working.
Why this got confusing in the first place
For most of 2025 and into early 2026, this was a simple question with a simple answer: ChatGPT shopping had no ads. A January 2026 traffic analysis from Discovered Labs, which inspected over 1.1 million product references pulled through ChatGPT's shopping feature, found zero sponsored placements. Every product card traced back to Google Shopping's underlying catalog, with nothing but a utm_source=chatgpt.com tag to show where the click came from.
That changed fast. OpenAI began testing ads in the US on February 9, 2026. By May, it had opened a self-serve Ads Manager with no minimum spend, and on June 11, 2026, it shipped product feed ads, letting retailers connect an entire catalog and have ChatGPT auto-generate sponsored cards instead of building creative one SKU at a time. If you read an older guide claiming ChatGPT shopping has "zero advertising," it's describing a system that no longer exists.
The result is a shopping response that can now contain two fundamentally different things stacked next to each other: a product your brand earned through data quality, and a product your competitor paid to place. Telling them apart, and tracking each one correctly, matters more than ever if you're trying to understand where your ecommerce traffic is actually coming from.
The core distinction: earned vs bought
OpenAI has been unusually direct about this split. In its Instant Checkout announcement, the company states plainly that "product results are organic and unsponsored, ranked purely on relevance to the user," and that merchants using Instant Checkout get no ranking boost for it. Separately, OpenAI's Ads in ChatGPT help center confirms ads are a distinct system layered on top, visually separated and labeled "sponsored" below the generated response.
| Organic product cards | Sponsored product cards (ads) | |
|---|---|---|
| How you win it | Feed quality, review count, GPT tag, promotional pricing | Bid (CPM or CPC) plus relevance |
| Visual placement | Inside the generated answer | Below the answer, labeled "sponsored" |
| Can you pay for it? | No, not currently | Yes, via self-serve Ads Manager |
| Who sees it | All ChatGPT users | Free and Go plan users only |
| Data source | Web crawl (87.3%) or merchant feed (12.7%) | Product feed connected through Ads Manager |
| Targeting | None, matched to query intent | Conversation context, ad copy, landing page, advertiser context hints |
| Appears in ChatGPT Atlas browser | Yes | Not during current testing |
That 87.3% vs 12.7% split for organic cards comes from Profound's analysis of 812,190 product cards sampled between June 18-25, 2026, and it's a detail worth sitting with. Most organic shopping recommendations aren't pulled from a merchant's tidy product feed at all. They come from ChatGPT crawling the open web and figuring out, on its own, what a product is and whether it fits the query. That's a very different optimization target than feeding a clean XML file into Google Merchant Center.
Four ways to visually tell them apart
- The sponsored label. Ad units carry an explicit "Sponsored" tag next to a clickable favicon, headline, short description, and a square image, per Pacvue's breakdown of the ad unit spec. Organic cards never carry this label because OpenAI legally can't call something unsponsored and then charge for it.
- Position in the response. Ads render below the end of the generated text, after ChatGPT finishes its answer. Organic product cards are embedded inside the answer itself, often with a comparison table or carousel.
- Which account you're using. This one trips up a lot of people doing manual spot checks. Free and Go plan users see ads. Plus, Pro, Business, Enterprise, and Edu users never do, and neither do users under 18. If your "audit" was done from your own paid ChatGPT account, you've been checking a version of the product that structurally cannot show ads.
- Whether it shows up in Atlas. Ads are excluded from OpenAI's Atlas browser during the current testing phase. Organic shopping results are not.
How organic product cards actually get selected
This is worth understanding in detail because it's where most of the actionable optimization work lives. According to Profound's July 2026 research, ChatGPT runs a separate internal mechanism called "product query fanout," distinct from the general query fanout that shapes the text answer, specifically to source candidate products. Once candidates are gathered, ChatGPT pulls commercial details, price and merchant name, from whichever offer ranks first in its internal offer list. If no offer is available, it falls back to ChatGPT's own product search index.
The attributes that separate top-ranked cards from bottom-ranked ones are specific and measurable:
| Product attribute | Top-ranked cards | Bottom-ranked cards | Relative lift |
|---|---|---|---|
| Has GPT tag | 2.39% | 0.98% | +144% |
| Median review count | 787 | 352 | +124% |
| Has promotional price | 1.19% | 1.05% | +13% |
| Has offers available | 63.87% | 60.90% | +5% |
GPT tags aren't something a merchant sets directly, they're machine-assigned labels based on available product data. That means the path to earning one is indirect: complete specs, accurate availability, and content that unambiguously describes what the product is and does. Review count mattering this much is a reminder that review generation and management isn't just a conversion-rate tactic anymore, it's an AI shopping visibility signal too.
How ChatGPT ads actually work
Once a retailer connects a product feed through Ads Manager, ChatGPT can auto-generate sponsored cards from the feed data rather than requiring hand-built creative for each item, and the system can scale to roughly a million SKUs per advertiser according to Passionfruit SEO's reporting on the May 2026 launch. New ecommerce partners have to submit a 100-product sample for review before the full catalog connects, which is a stricter gate than Google Merchant Center or Meta's shop catalog require.
Pricing runs on CPM or CPC, with CPC using a second-price auction, meaning you pay just above the next-highest bid rather than your actual max bid. Pacvue reports most brands start with $10,000 to $25,000 monthly test budgets, though some are already spending far more. Targeting has no demographic or behavioral layer; matching happens through conversation context, the ad's own copy and landing page, and advertiser-supplied "context hints." If your site blocks OAI-SearchBot or ChatGPT-User in robots.txt, your ads simply won't serve, and zero impressions is usually the first symptom anyone notices.
One detail that matters for anyone running paid campaigns: the exact same structured product feed drives three things at once. Google Shopping performance, ChatGPT ad eligibility, and ChatGPT organic citation. Fix your feed data once, and it improves all three channels simultaneously rather than requiring separate work for each.
Ad frequency: how common are these ads, really?
This is the part that's moved the most, and the fastest. Promptwatch's tracking of ad-bearing responses shows ChatGPT served zero ads until May 27, 2026, when they first appeared in 1.69% of search-powered responses. From there the climb was not gradual. Ads crossed 30% of search-triggered responses on July 1, 2026, and have stayed above that level since. Over the 90 days through August 17, 2026, the average sat at 20.1%, but the most recent 7-day average had already climbed to 32.4%, with single-day peaks hitting 43.9%.
The mix of what triggers ads is revealing too. Across the most recent week tracked, 73.3% of ad-bearing responses came from generic organic-style prompts, 18.0% from brand-specific searches, and 8.6% from direct competitor comparisons. If you're a brand being compared against a rival by name, there's a real chance a competitor's ad shows up right where you'd want to be seen.
Separately, Promptwatch's shopping feature usage data shows that shopping cards themselves (organic ones) attach to only a low single-digit percentage of all web-search responses overall, but concentrate heavily on genuinely commercial and transactional queries, where the trigger rate runs far higher than the all-prompt average. In other words, the feature is narrow but sharp, it shows up when someone is actually trying to buy something, not during casual browsing-style questions.
Why you can't rely on a single manual check
I ran into this myself testing a few prompts across accounts and markets, and the inconsistency is the whole story. A prompt that shows an ad on one run might show none on the next, since ads attach per-response through a rotation rather than being locked to a specific query. Market and language matter too, an ad check from a US office doesn't tell you what a shopper in Mexico, Japan, or the UK sees, and OpenAI's rollout went from a US pilot in February 2026 to Canada, Australia, and New Zealand in March, then the UK, Mexico, Brazil, Japan, South Korea, and 31 European countries by August.
Stack all of that with the account-type restriction (no ads on paid seats) and you get a channel that's genuinely hard to audit by hand. A single screenshot from a single account on a single day tells you almost nothing reliable about whether your brand, or your competitors, are showing up as ads.
How to track organic product cards
The practical starting point is treating your Google Shopping feed as the shared foundation, since the same structured data drives organic citation in ChatGPT, Google Shopping performance, and ChatGPT ad eligibility. From there:
- Run the actual commercial and transactional prompts your buyers use, repeatedly and across several days, and log whether your products appear as cards.
- Check review counts and consider whether your review generation strategy needs attention given how strongly review count correlates with card rank.
- Make sure product pages carry complete, specific attributes: material, dimensions, occasion suitability, certifications. These are exactly the fields keyword-era feeds tend to leave blank, and they're what feeds ChatGPT's own product-matching logic when it falls back to web crawl data instead of a merchant feed.
- Watch Promptwatch's shopping feature usage tracking to see whether trigger rates are shifting for your category, since OpenAI appears to actively tune when shopping cards appear at all.
For ongoing monitoring at scale rather than one-off manual checks, Promptwatch tracks product card appearances alongside broader AI citation data, including the content-type breakdown showing product pages now account for 22.4% of all classified ChatGPT citations overall, and roughly 26% on purely commercial prompts.
How to track sponsored ads
OpenAI's own Ads Manager only reports on your own campaigns, it gives you zero visibility into what competitors are running. If you need competitive intelligence on who's buying ad slots against your brand terms or your product category, that requires third-party capture.
Promptwatch's Ads Radar feature extracts every ad found in tracked responses, including the creative, headline, ad copy, which model it appeared on, the advertiser and domain, and the date it was seen. The dashboard shows ad volume over time, ranks top advertisers, and lets you view the data by ad or by prompt, filterable by advertiser domain, ad type, and date range.
When evaluating any ad-tracking tool for this purpose, worth checking for seven things: does it capture the real interface by market (not just an API), does it log the complete ad record (advertiser, domain, creative, copy, landing page, prompt, date), does it collect on a schedule rather than one-off, can you segment by prompt type, does it keep ad data as its own distinct metric rather than blending it into a generic "visibility score," can you export or access the data via API, and is pricing actually published.
A practical tracking setup for 2026
Given how fast this moved, here's a reasonable structure if you're setting up tracking from scratch:
- Separate dashboards for ads and organic. Don't let a blended "AI visibility score" obscure whether you're winning through content quality or through ad spend, they require completely different fixes.
- Sample across days, not a single run. Ad rotation and the fact that Promptwatch's own 7-day average (32.4%) sits well above the 90-day average (20.1%) both tell you the landscape shifts week to week.
- Check from a Free or Go account specifically if you're manually spot-checking ads. Your Plus or Pro account will show you nothing.
- Treat your product feed as a single asset serving three channels. Fixing titles, availability, and attributes in Google Merchant Center improves organic ChatGPT citations, ChatGPT ad eligibility, and Google Shopping all at once.
- Track competitor ad activity if you're in a category where brand-comparison prompts are common, since 8.6% of ad-bearing responses come specifically from competitor-comparison queries.
If you're building out a broader GEO and AI visibility stack rather than just the shopping piece, the directory at bestgeosoftware.com is a reasonable place to compare platforms beyond this specific feature.
What's likely next
OpenAI has been expanding measurement and attribution partnerships fast, adding Hightouch, Tealium, LiveRamp for conversion data, and attribution partners like AppsFlyer, Triple Whale, and Northbeam, which signals the company expects ad spend here to grow substantially. A new visual ad format tied to image generation was announced for October 2026, initially on Free and Go plans. WeightWatchers reported a 15.3% lower attributed CPA on ChatGPT Ads compared to its blended paid-search benchmark, and Portland Leather found 93% of ChatGPT Ads visitors were new to the brand, early numbers, but they explain why budgets are moving here quickly.
The organic side isn't standing still either. OpenAI's own merchant materials describe continued investment in product coverage and relevance, and the gap between brands with clean, complete product data and brands without it is only going to widen as more shopping-intent conversations happen inside ChatGPT instead of a traditional search engine. Whichever side of this you're optimizing for, the one thing that's no longer true is the old assumption that ChatGPT shopping is purely a meritocracy with no paid layer. It's a hybrid now, and treating it as one or the other will leave you missing half the picture.