Brand-level AI visibility answers one question: does the AI know who you are? But when a buyer asks “what’s the best espresso machine for a beginner”, the answer that comes back doesn’t recommend a brand. It recommends a model — a specific product, often with a price and somewhere to buy it. You can be mentioned in that answer and still lose it, because the machine being recommended is somebody else’s.
BotScope now tracks that. Today every project gets a Products tab: which specific models AI answers recommend — yours and your competitors’ — across every provider we scan, plus a second surface most brands don’t know exists yet: ChatGPT’s shopping carousel, with the prices it displays and the merchants it hands the click to.
TL;DR
- A product catalog per project — the models you care about, yours and competitors’ (“Telecaster” → Fender, “Les Paul” → Gibson). AI suggestions draft one from your homepage in a click.
- Model-level mentions, every provider. In our own tracking of a retail test project, 18 of 28 answers (64%) named at least one specific model. Each mention is scored for sentiment from its own sentence — so “the K2 is superb but the Uno feels dated” scores each model separately.
- Models-per-prompt — the headline view. For each question you track: which models the answer actually recommends, in how many scans, and whether any of them are yours.
- ChatGPT’s shopping carousel, captured. Product cards with displayed prices, ratings and merchants. In our tracking it appeared on roughly one in four ChatGPT answers to retail-shaped prompts — and of 39 product placements, a single retailer carried 74% of the merchant links.
- Price history — the price the carousel showed, per scan and per country. Not the price on the merchant’s site: the price the buyer actually saw in the answer.
- It ships with history. Model mentions are backfilled across your entire stored answer archive the moment you add a catalog. Carousel capture has been recording since 7 August 2026.
- In reports, the API and MCP — client reports and PDFs carry a Products section, and twelve new endpoints plus eleven MCP tools expose the same numbers.
- Every plan, including trials. No feature gate.
Your model appears — but a competitor’s leads the answer. Present is not the same as winning.
Why model-level matters
The difference between brand visibility and product visibility is the difference between being in the answer and being the answer. A brand mention in a listicle is nice. “Buy this specific machine” is a purchase about to happen.
Model-level tracking also finds a failure mode brand-level tracking structurally cannot see: the prompt where your brand appears but none of your models do. The AI knows who you are, talks about you warmly, and then recommends someone else’s product. At brand level that reads as a win. At product level it reads as what it is — a recommendation you’re losing five scans out of five, with a name attached to what’s beating you.
And because every mention is scored from its own sentence, sentiment finally works at the granularity buyers think in. A response that praises one of your models and damns another produces one useless brand-level score — and two accurate product-level ones.
The carousel nobody is watching
For shopping-shaped questions, ChatGPT doesn’t just write an answer any more. It shows product cards — image, price, rating, and merchants to buy from. This is organic, not advertising (we track sponsored placements separately, and are careful never to mix the two). It’s also, as far as we can tell, completely unmonitored by the brands in it.
BotScope captures the carousel on every ChatGPT scan: which products appear, at what displayed price, with what rating, sold by whom. Because prices are recorded per scan, you also get price history — and the displayed price does drift, which matters when the number a buyer first sees for your product isn’t the one on your own site.
Who captures the click
Every carousel card links to a merchant. Aggregate those links and you get a question retail brands should find mildly alarming: when AI answers your category’s questions, who gets the sale?
39 placements across one project’s scans. Every product card in a carousel links somewhere — and when three quarters of them link to one retailer, that retailer is quietly becoming the default checkout for your category’s AI answers.
In our own tracking the answer was strikingly concentrated — one retailer carried 74% of all carousel placements we observed. If that holds for your category, the AI layer isn’t just deciding which product gets recommended. It’s deciding which store sells it.
The honest limitations
We would rather you hear these from us than discover them.
Mention tracking only counts what’s in your catalog. A model you don’t track is invisible to it. The AI suggester drafts a catalog in seconds, but the craft is in the aliases: answers say “Strat”, not “American Professional II Stratocaster”. Add the short names people actually write — and add sub-brands to your brand aliases, so their models count as yours.
The carousel is ChatGPT-only, with history from 7 August 2026. Carousels only exist where a provider renders them. Capture started on 7 August; answers before that are excluded from carousel metrics rather than counted as empty. Model mentions have no such limit — they backfill across everything you’ve ever scanned.
Prices are observations, not quotes. We record what the carousel displayed at scan time. It is not a live price feed, and we never fetch or click merchant links — a card’s destination stays untouched, exactly as with ad landing URLs.
Serving varies, so every number carries its denominator. The same prompt can produce a carousel one scan and none the next. That’s why prompt-level results read “seen in 4 of 5 scans” rather than a percentage, and why every rate in the tab is paired with the count it was measured across.
In your reports, API and MCP
Client reports, share links and PDF exports carry a Products section — the model leaderboard and models-per-prompt, then the carousel and merchants when there’s data. Projects with neither simply don’t get the section; no empty panels. When a new product appears in your carousel for the first time, a marker drops onto your trend charts automatically, next to the new-advertiser ones.
Twelve endpoints join the public API — the catalog, four mention views, and seven carousel views including per-product price history — with the same set as MCP tools, so you can just ask Claude which models beat yours last month.
FAQs
Do I need to set anything up? For the carousel: no — if your project scans ChatGPT, it’s being captured now. For model mentions: add a catalog in project settings (two minutes with the AI suggester), and your entire scan history is backfilled the moment you save.
Does this cost extra credits? No. Both surfaces arrive on scan results you already pay for. Reading them costs nothing, on every plan including trials.
Does this affect my visibility scores? No. Product mentions and carousel placements are recorded separately and never counted as brand mentions — mixing them would quietly corrupt every score and historical comparison you have.
My brand owns several sub-brands. How do I track their models as mine? Add the sub-brand names to your project’s brand aliases. Model attribution resolves against your brand, its aliases and your domain — so once “Squier” is an alias of Fender, a Squier model counts as yours everywhere.
Why does a model I know gets recommended show zero mentions? Almost always aliases. We match the model names in your catalog verbatim against answer text, word-boundary safe — if answers say “the Classic” and your catalog only has “Fairbank Classic Series II”, it won’t match. Add the short form as an alias and re-check; history re-scores automatically.
For a year, AI visibility has meant knowing whether the machines mention your name. But buyers don’t purchase a name — they purchase the specific thing the answer told them to. Open the Products tab and find out what that is.