How to check AI visibility: a guide for ecommerce brands

For a store, AI visibility has four parts, and each depends on the one before it. An assistant has to be able to reach your product pages, read what they say, choose your product when a shopper asks for something it fits, and send that shopper to you.

• Reach: the assistant's crawler is allowed in and gets a product page back.
• Read: the page states the facts a shopper's question turns on, in text and in structured data.
• Recommend: the assistant names your product, not only cites your page as a source.
• Visit: the shopper clicks through, and your analytics can tell where they came from.

A mention rate, the share of answers that name your brand, covers only the third part, and on its own it cannot tell you why a product is missing. Checking the four parts in order turns a low score into a cause you can fix, and the guide to how AI shopping systems choose products explains why each part matters to the choice.

Open yourstore.com/robots.txt and look for the crawlers each assistant reads with:

• OAI-SearchBot, for ChatGPT. OpenAI says sites that opt out of it will not be shown in ChatGPT search answers. GPTBot is OpenAI's training crawler, and blocking it does not take you out of search.
• PerplexityBot, for Perplexity, which recommends allowing it so your site appears in its results.
• Googlebot, for Google Search and its AI Overviews and AI Mode. Google-Extended governs Gemini training and grounding: Google says it does not affect inclusion in Search, but blocking it can keep your pages out of answers in the Gemini app.

Then check that the file is a file of rules at all. When Sighted scanned a UK sports equipment retailer in October 2026, its robots.txt address without www redirected to the homepage, so a crawler asking there for rules got a web page instead. A site can also turn automated visitors away whatever robots.txt says: when Sighted's scan tried to reach one large sports retailer that month, its address without www timed out and its www address refused the request. The AI crawlers glossary entry lists the rest.

An assistant can only recommend a product for a need its page states. Pick ten product pages across your range and read each one as a shopper's question would: what is it made of, how does it fit, what size, which model is it compatible with, is it in stock, what does delivery cost, can it be returned.

Then check the structured data behind the page. Google says no special schema is needed for its AI features, but its merchant listing structured data still carries price, availability and identifiers in a form systems read reliably, and it has to match what the page shows.

When Sighted's scan read a UK online department store's product pages on 30 October 2026, and the findings were checked against the pages that day, the pages carried no reviews, returns or delivery policy, item condition, GTIN, FAQ or specification table for an assistant to read. The AI product readiness checklist lists each fact to look for.

Write 15 to 25 questions the way your shoppers ask them, by need rather than by brand: "a waterproof running jacket for winter under £100", not your product's name. Ask each one in ChatGPT, Perplexity and Gemini, in a new chat, and record for every answer:

• whether your brand or product appears;
• whether it is recommended, or only cited as a source;
• the reason the assistant gives for each product it names;
• which stores and products it names instead of yours;
• the date you asked.

Ask the same set again on a later date before you read a change as a trend, and keep the questions fixed so the two dates compare. In Sighted's pilot with Onboard Sportswear, which consolidated duplicate URLs, submitted the site's first sitemap, addressed indexing gaps and enriched the products, AI citation coverage went from 0% to 33% of tested responses. The ChatGPT Shopping ranking factors help you read the reasons an answer gives.

Visibility that never becomes a visit is hard to value, so finish in your analytics.

• GA4 added an AI Assistant channel on 13 May 2026, which groups visits whose referrer is a recognised AI assistant. Under Google's channel definitions, visits from its own AI Overviews and AI Mode stay in Organic Search.
• Search Console's generative AI performance reports, open to every site since 31 August 2026, show your impressions and pages in Google's AI features.
• Your landing pages report shows which product pages those visits arrive on, which tells you whether the products you checked in step 3 are the ones being found.

Some influence never shows as a click: a shopper can read an answer and type your address later. Read the channel as the least that assistants other than Google's send you, and set it beside the answers you recorded. The guide to measuring AI shopping performance builds these signals into one dashboard you can keep.

1. Open your robots.txt and confirm OAI-SearchBot, PerplexityBot and Googlebot can reach your product pages, and that the address returns rules, not a web page.
2. Read ten product pages against the questions your shoppers ask, and list the facts each one leaves out.
3. Check that the structured data on those pages matches the price, stock and identifiers the page shows.
4. Ask ChatGPT, Perplexity and Gemini 15 to 25 of your shoppers' questions, record who they name and why, and ask again on a later date.
5. Read GA4's AI Assistant channel and Search Console's generative AI reports, and note the date of every change you make to your product data.

The free AI visibility audit runs the first two steps for you: it reads your robots.txt, reports which AI crawlers are let in, and scores up to 50 of your product pages on the facts and structured data an assistant reads, ranking what to fix first.

On paid plans Sighted runs the third step on a schedule, asking ChatGPT, Perplexity and Gemini a set of real shopper questions about the kind of products you sell and recording whether each answer names your store. The gaps it finds are filled by enriching the catalogue at scale from your own data, the technology behind SO-ME.