AI shopping visibility: be the product ChatGPT, Gemini and Rufus recommend
AI shopping visibility is being one of the products an assistant recommends when a shopper asks what to buy, before that shopper reaches a store or your site. Asva measures it directly: recommendation share on your category prompts, per assistant and per market, alongside who is recommended instead, which sources decided it, and whether your catalogue is legible enough to be eligible at all.
What AI shopping visibility is, and why it is a separate channel
When a shopper asks an assistant for the best option in a category with a budget and a use case, the answer is a shortlist of three to five products with a sentence of reasoning each. That comparison happens entirely outside your analytics: if your product is not on the shortlist there is no impression to lose and no bounce to see. Recommendation share therefore has to be measured directly, by asking the assistants the questions your shoppers ask and recording what comes back.
It is also a catalogue problem before it is a content problem. Assistants that recommend products need structured facts to reason with: what the product is, who makes it, what it costs, whether it is in stock and what people think of it. Missing identifiers or availability, a feed that disagrees with the page, or product pages that only render those fields in the browser all disqualify a product before quality is considered.
The surfaces covered are ChatGPT Shopping, Gemini and Google AI Mode shopping, Perplexity shopping, Microsoft Copilot and, for brands selling on Amazon, Rufus. This is the retail application of answer engine optimization; the AEO overview and the AEO glossary cover the vocabulary used below.
What the feature tracks
Six views of the same question: are you on the shortlist, and if not, why not.
Recommendation share for category prompts
The headline number. For every prompt in your set, such as "best running shoes for flat feet under $150", we record which products the assistant names, in what order, and why. Recommendation share is how often one of yours is on that shortlist, measured per prompt, per assistant and per market, because a product recommended in one country is routinely absent in another.
Per-assistant breakdown
ChatGPT Shopping, Gemini and Google AI Mode, Perplexity, Amazon Rufus and Microsoft Copilot each build answers from different inputs, so the same product can be a default pick on one and invisible on another. The breakdown shows share by surface, which tells you whether a gap is a feed problem, a retrieval problem or a review-coverage problem.
Competitor products named instead
When you are not on the shortlist, something else is. We log every product and brand recommended for each prompt, how often and at what position, so you get a ranked list of who the assistants prefer for each use case. The products winning your prompts reveal which sources and which product facts the assistant is weighting.
Sources cited for the recommendation
Shopping answers lean on review sites, editorial roundups, community threads and marketplace listings far more than on a brand's own site. We capture the URLs each assistant cites for every prompt and roll them up by domain. That list is your work plan: the review platforms you need a complete listing on, the roundups to pitch, the threads where your category is discussed without you.
Catalogue legibility
Whether an assistant can read your products at all. We crawl product pages and score Product structured data completeness per SKU: name, brand, GTIN or MPN, image, offers with price, priceCurrency and availability, and aggregateRating where reviews exist. Where a feed applies, we flag mismatches between feed, page schema and visible price, plus product pages that only render these fields in the browser.
Trend over time
Assistants change models, sources and shopping features often, so a snapshot tells you little. Every metric is stored per run, so you can see recommendation share move week over week, tie a change to a feed fix or a new review listing, and watch a new SKU from launch to the day it first becomes eligible.
How it works
Five steps, in the order that finds problems cheapest first.
- 1
Connect your catalogue
Point the tracker at your storefront domain and, where you use them, your merchant feed and marketplace listings. We crawl product and category pages, read the Product JSON-LD on each, and build the SKU list the module works from.
- 2
Build the prompt set
Start from category, budget, use-case and comparison questions shoppers actually ask, in the language and currency of each market you sell in. We suggest prompts from your categories and competitor set; you edit them. A typical set is 50 to 300 prompts per market, including the ones you currently lose.
- 3
Run the prompts on a schedule
Each prompt is run against ChatGPT Shopping, Gemini and AI Mode, Perplexity, Copilot and, if you sell on Amazon, Rufus. We record the products named, their order, the reasoning and the sources cited. Runs repeat so the numbers are a trend, not a one-off.
- 4
Score catalogue legibility
In parallel we grade every product page for schema completeness, check that price, availability and JSON-LD are in the raw HTML rather than injected by JavaScript, confirm retrieval crawlers can fetch the pages, and reconcile feed values against the site so a stale price or sell-out is caught before an assistant repeats it.
- 5
Fix in order and re-measure
The dashboard ranks what to do by expected effect: eligibility gaps first, because a product with missing GTIN or availability cannot be recommended however good it is, then source coverage, then on-page product content. Ship the fix; the next run shows whether recommendation share moved.
Building or checking Product JSON-LD? Use the JSON-LD generator to produce a template and the JSON-LD validator to confirm what a page actually serves.
What each assistant reads
The inputs differ by surface, which is why the breakdown is per assistant. These notes reflect each platform's documentation at the time of writing; shopping features change often, and the tracker tells you when behaviour has moved.
ChatGPT Shopping
ChatGPT builds product answers from structured product metadata, which as documented at the time of writing can include a merchant product feed submitted to OpenAI, together with on-page product data and reviews retrieved from the open web. Retrieval is done by OAI-SearchBot and, for pages fetched during a conversation, ChatGPT-User. If those agents are blocked, or your product page only renders price and stock in the browser, the assistant has nothing to reason with. Where a feed exists, its accuracy and freshness matter as much as the page.
Gemini and Google AI Mode shopping
Google's shopping answers draw on its product data, fed by Google Merchant Center listings and by product pages Googlebot indexes with Product structured data. A complete, current Merchant Center feed and matching on-page schema are the inputs that matter. Google-Extended is a training control, not a search control: blocking it does not remove your products from AI Mode, and allowing it does not add them.
Perplexity shopping
Perplexity answers from live web retrieval via PerplexityBot, so review sites, comparison roundups and community threads weigh heavily; a product absent from the sources it trusts is unlikely to be recommended. Perplexity has also offered merchants a way to share product data directly; the details have changed more than once, so check its current documentation.
Amazon Rufus
Rufus answers from Amazon's own catalogue: listing titles, bullets, A+ content, customer questions and reviews. It does not retrieve your website, so it only matters if you sell on Amazon, and the work is listing completeness and review health rather than schema or feeds. We track it separately for that reason; this page replaces the earlier standalone Rufus optimizer tool.
Microsoft Copilot
Copilot grounds answers in the Bing index and, for shopping, in Microsoft Merchant Center product data where a merchant has submitted it. The Product schema that serves Google serves Bing, so most teams get Copilot coverage as a by-product of the first two. Shopping features here move quickly; check current documentation.
Visibility versus agentic checkout
Two different problems get bundled under "agentic commerce". Visibility is whether an assistant recommends your product. Agentic checkout is whether an agent can then complete the purchase on the shopper's behalf, through protocols such as the Agentic Commerce Protocol, the Universal Commerce Protocol or a Model Context Protocol server exposing your catalogue and cart. They are sequential, not alternatives: an agent will not buy a product it never recommended, and a perfect checkout behind a product the assistant cannot read transacts nothing.
This page covers the first half. If you want to know whether your storefront is ready for the second, the agentic readiness scan checks protocol support, feed exposure and crawler access for free and tells you what an agent would find if it tried to transact today.
Who it is for
Ecommerce and D2C brands
Merchandising and growth teams who need recommendation share on the monthly report, and a ranked fix list for the catalogue.
Ecommerce and retail solutionMarketplace and commerce SaaS
Platforms whose merchants' products need to be recommended, with catalogue legibility scored across thousands of storefronts.
SaaS solutionAgencies
Retail agencies who need per-client prompt sets, per-market reporting and a defensible before-and-after for the work they bill.
Agency solutionMetrics you report
Every number is stored per run, per assistant and per market, and exports to the reporting you already use.
Recommendation share: share of category prompts where at least one of your products is named, per assistant and per market.
Average position: where your product sits in the shortlist when it appears.
Competitor share: the same two numbers for every brand named instead of you.
Cited-source share: which domains supply the citations behind shopping answers, and whether you are present on them.
Catalogue legibility score: Product schema completeness and feed consistency across the whole SKU list, not a sample.
Price and availability accuracy: what assistants say you charge and whether you are in stock, against the truth.
Launch-to-eligible time: days from a new SKU going live to its first recommendation.
For brand-level mention rate and share of voice outside purchase prompts, pair this with the Brand Visibility Tracker. For the citation profile behind those answers, see Citation Intelligence.
Five mistakes that keep products off the shortlist
Incomplete GTIN, price or availability
The most common reason a product is never recommended. Assistants need identifiers to match a product across your site, your feed and third-party listings, and price, currency and availability to answer a budget question. A product missing any of these is usually skipped rather than guessed at.
Product pages that only work in the browser
Headless storefronts often inject price, stock and JSON-LD client-side. Retrieval crawlers take the raw HTML, so the fields the assistant needs are simply not there. Fetch a product page without JavaScript and check.
Blocking retrieval crawlers
Bot-management rules written to stop training crawlers frequently catch the retrieval ones too. OAI-SearchBot, ChatGPT-User, PerplexityBot and ClaudeBot are retrieval agents; GPTBot, CCBot and Google-Extended are training controls. Block the second group if you wish, but blocking the first removes you from the answers.
Stale feeds
A feed that still shows last month's price or a sold-out variant as in stock produces a wrong answer with your name on it. Stale data is worse than absent data. Fix the sync cadence so changes propagate within hours, and use identical identifiers in feed, page schema and marketplace listings.
Ignoring review and roundup sources
An assistant asked for the best option cites sources that compare several products, which your product page by definition does not. If you are absent from the review platforms and roundups it cites, on-page perfection will not get you recommended.
Frequently asked questions
What is AI shopping visibility?+
AI shopping visibility is how often an AI assistant recommends your product when a shopper asks it what to buy, before that shopper reaches any store or search results page. It is measured as recommendation share across a set of category prompts, per assistant and per market.
Which AI shopping assistants does Asva track?+
ChatGPT Shopping, Gemini and Google AI Mode shopping, Perplexity shopping, Microsoft Copilot and, for brands that sell on Amazon, Amazon Rufus. Each is tracked separately because each reads different inputs.
How is this different from the Brand Visibility Tracker?+
The Brand Visibility Tracker measures whether an assistant mentions your brand in answers generally. AI Shopping Visibility measures whether it recommends a specific product for a purchase prompt, and adds the catalogue side: Product schema completeness, feed consistency and the sources cited for shopping answers.
How does ChatGPT decide which products to recommend?+
As documented at the time of writing, ChatGPT combines structured product metadata, which can include a merchant-submitted feed, with product information and reviews retrieved from the web through OAI-SearchBot and ChatGPT-User. Complete structured data, accessible product pages and presence on trusted review sources are the practical levers.
Do I need a merchant feed to appear in AI shopping answers?+
Not for every assistant, but it helps on the ones that support one. Google AI Mode leans on Merchant Center, OpenAI has documented a merchant feed for ChatGPT, and Copilot uses Microsoft Merchant Center. Perplexity relies mostly on web retrieval. In all cases the on-page Product schema still needs to be complete and to match the feed.
Does blocking GPTBot stop my products appearing in ChatGPT Shopping?+
No. GPTBot is a training crawler. Shopping answers are built from retrieval, which uses OAI-SearchBot and ChatGPT-User. You can disallow GPTBot and still be recommended, provided the retrieval agents can fetch your product pages.
Is Amazon Rufus relevant if I only sell on my own site?+
No. Rufus answers from Amazon's catalogue and reviews and does not retrieve external sites. If you do not sell on Amazon, ignore it. If you do, Rufus visibility is a listing-quality and review problem and is tracked separately.
What is the difference between AI shopping visibility and agentic checkout?+
Visibility is being recommended. Agentic checkout is an agent completing the purchase on the shopper's behalf through protocols such as ACP, UCP or MCP. They are sequential: an agent cannot buy a product it never recommended, so visibility comes first. Our free readiness scan at /features/agentic-readiness covers the checkout side.
How long until recommendation share changes after a fix?+
Eligibility fixes such as adding missing GTIN or availability, or unblocking a retrieval crawler, typically show within a few scheduled runs once the page has been re-fetched. Source work, such as getting listed on a review platform the assistant cites, takes longer because the third-party page has to exist and be retrieved first.
Can I track several markets and currencies?+
Yes. Prompt sets, catalogue checks and recommendation share are all per market, because the sources an assistant cites differ by country and language. Run the prompts in the local language with a local-currency budget.
Complete your retail AI visibility stack
Dominate AI-Driven Shopping
As consumers increasingly turn to conversational AI for product recommendations and shopping advice, ensure your products are front and center in AI shopping experiences.