How to check what ChatGPT, Copilot and Perplexity say about your store — and fix it

More of your customers ask an assistant before they search, and the assistant's answer about your store gets written whether or not you have looked at it. Checking takes about an hour. Most of the fixes are the same things that help human shoppers, and none of them need tricks. This is the audit we run for clients, laid out so you can run it yourself first.

Step 1: ask the assistants what your customers ask

Use ChatGPT with search on, Microsoft Copilot, Perplexity, Gemini and Google's AI Mode. If you sell in the US or Canada, add Meta's Muse app, a personal shopping agent that recommends products and can buy them for the user once they approve. Ask each one the same questions and save every answer with the date, because next month's comparison is the whole point.

Set up so the results mean something

• Use a private window or a logged-out session where you can. ChatGPT says it tailors product results to each shopper's preferences and past conversations, so your own history skews what you see.
• Ask each question twice. Answers vary between runs; a fact that is wrong once is worth noting, a fact that is wrong twice is worth fixing first.
• Record the sources each assistant cites, not just the answer. The sources tell you what to fix.

The questions

Adapt these to your store. Use your customers' wording, not your marketing's.

• Trust: “Is [brand] legit?”, “[brand] reviews”, “is [brand] a good company to buy from?”
• Discovery: “best [category] in [your country or city]”, “where can I buy [product type] that [key attribute]?”
• Comparison: “[brand] vs [main competitor]”, “alternatives to [competitor]”
• Policies: “what is [brand]'s return policy?”, “how long does [brand] take to ship to [city]?”, “does [brand] ship to [country]?”
• Product detail: “[product] size guide”, “is [product] compatible with [thing]?”, “what is [product] made of?”

Score every answer

1. Accurate? Prices, policies, shipping regions, what you actually sell.
2. Cited? Does it link to you, or to a third party talking about you?
3. Recommended? On the generic discovery questions, do you appear at all, and next to whom?

Wrong facts are the priority, because they cost sales from people who already wanted you. Absence from generic questions comes next.

Step 2: know where the answers come from

Each assistant builds its answer from a mix of search indexes, its own crawlers, product feeds and third-party sites. Knowing the mix tells you where a wrong answer started.

Search indexes

Copilot runs on Bing. ChatGPT's search uses OpenAI's own crawler and index together with third-party search providers, Bing among them. So if your store is thin in Bing's index, it is thin in two assistants. Set up Bing Webmaster Tools next to Google Search Console; it takes about twenty minutes, and Bing supports IndexNow, which tells it about a changed page the same day.

Crawlers, and what blocking each one does

• OpenAI runs four, and each is controlled separately: OAI-SearchBot decides whether you appear in ChatGPT search, GPTBot collects training data, ChatGPT-User fetches a page when a user asks, and OAI-AdsBot only checks pages submitted as ads (OpenAI). Blocking GPTBot does not remove you from ChatGPT search; blocking OAI-SearchBot does.
• Perplexity uses PerplexityBot to surface sites in its search results (not for model training) and Perplexity-User for fetches a user asks for, which Perplexity says generally ignores robots.txt (Perplexity).
• Anthropic uses Claude-SearchBot for search, separately from its training crawler.
• Google builds AI Overviews and AI Mode from ordinary Googlebot crawling. The Google-Extended setting does not affect them (Google).

Check robots.txt, and also any firewall or CDN setting labelled something like “block AI bots”. Those toggles often catch the search crawlers too, which removes you from answers without protecting anything.

Product feeds

Shopping answers lean on structured product data more than on page text.

• ChatGPT. OpenAI says shopping in ChatGPT is live in the US. If you sell through Shopify or Etsy, your catalog is already integrated with no application needed; other merchants apply to share a product feed. There are no fees on purchases that start in ChatGPT, and OpenAI says it is moving away from a standalone checkout inside ChatGPT, so shoppers complete the purchase on your site or app (OpenAI). A feed is optional if ChatGPT already crawls your site, but it gives you control over how products appear.
• Google. Shopping results, including in AI Mode, draw on Google's Shopping Graph, which your Merchant Center feed feeds. Disapproved products in Merchant Center are, in practice, invisible products.

Structured data

Product, Offer, Organization, FAQ and return-policy markup give an assistant price, availability, identifiers such as GTINs, and your policies without guessing. Missing markup does not stop you being mentioned; it makes the mention vaguer. Google is explicit that there is no special markup for its AI features, so ignore anyone selling “AI schema”.

Third parties

Review platforms, marketplaces, press, forums and directories are cited constantly. An assistant will happily quote a two-year-old complaint if it is the most concrete thing it can find about you. Look at which third-party pages each assistant cited in step 1; those are the pages shaping your reputation.

On Shopify, the platform itself

Since June, agentic storefronts have been on by default for eligible Shopify stores, so AI channels can read your catalog. Since August 5, every Liquid storefront has also exposed WebMCP tools that let a browser agent search the catalog and manage a cart, and on September 28 Shopify added tools for reading, updating and submitting checkout, Shop Pay included (Shopify). The buyer still authorises the final submission, and the agent hands control back for anything that needs a person, such as a Shop Pay login or a payment challenge (Shopify). Nothing needs installing, and today only Chromium-based browser agents can use them (Shopify). The practical meaning: your product data is, directly, the agent's experience of your store. Titles, variants, weights and policies all count.

Meta, which powers Muse, is one of those agentic channels and has its own two switches under Sales channels > Agentic > Meta: Catalog access decides whether Meta can find your products, and Direct checkout decides whether a buyer can pay without leaving Meta. If Allow Shopify to manage for me is on, turn it off first or the settings are not yours to change. We covered which to leave on in an AI agent can now buy from your store: what to actually decide.

Step 3: fix in this order

1. Make sure you can be read. Search crawlers allowed, pages indexed in both Google and Bing, and key facts (prices, policies, stock) in the page's HTML rather than only in images or scripts. Nothing else matters if this fails.
2. Correct wrong facts at the source. If an assistant says you don't ship to Australia and you do, the shipping page is unclear or not indexed. Fix the page, then the markup, then submit the URL in Bing Webmaster Tools or through IndexNow. You cannot edit an assistant's answer directly; you can only fix what it reads.
3. Make policies answer the question in the first sentence. Returns, shipping times, warranty. Assistants lift the direct sentence and struggle with a policy that opens with three paragraphs of preamble.
4. Complete the product data. GTIN or MPN, brand, weight, dimensions, materials, variant attributes, and availability that is actually true. The same work fixes Merchant Center disapprovals and feeds ChatGPT's and Shopify's agent channels, so it pays three times.
5. Make your business facts consistent everywhere. Same legal name, address, phone and description on your site, your Google Business Profile, your app or partner listings, LinkedIn and review platforms. Mark them up with Organization schema and link the profiles together. Assistants trust facts that agree with each other.
6. Publish the questions people really ask as an FAQ with schema, in the customer-service team's wording.
7. Answer the comparison questions yourself. An honest “[you] vs [competitor]” or “who this is not for” page gives an assistant something concrete to cite instead of a forum thread.
8. Add an llms.txt file if you like. It is a convention, not a standard, and no assistant promises to read it; Google says you do not need one. It costs an hour and does no harm, but do not expect it to change citations.
9. Earn a few good third-party mentions. A real review on a real platform, an industry directory listing, a partner page. Not volume, corroboration.

Step 4: measure it every month

A year ago the only measure was the prompt test. There are now real reports, each covering part of the picture.

• Your prompt log. Re-run the step 1 questions on the same day each month and note what changed. It is still the only measure that covers every assistant.
• Google Search Console. The Generative AI performance report shows how often your pages appeared in AI Overviews and AI Mode, by page, country and device. Impressions only; no clicks or queries (Google).
• Bing Webmaster Tools. The AI Performance report, in public preview since February 2026, shows how often your pages are cited in Copilot and in Bing's AI summaries, which URLs, and the search phrases the AI used to find them (Microsoft). It does not cover ChatGPT.
• Google Analytics. GA4's default channels now include AI Assistant, for visits from ChatGPT, Gemini, DeepSeek, Copilot and Grok. Perplexity is not in Google's list and lands in Referral, so add it to a custom channel group. Clicks from Google's own AI Overviews still arrive as ordinary organic traffic and cannot be separated (Google).

Expect movement in weeks for fixes on your own site that crawlers pick up quickly, and in months for anything that depends on third parties or on a model being refreshed.

What this is not

There is no way to buy a place in an unpaid answer. ChatGPT sells ads, but running them does not change what the assistant says about you when nobody is paying; the two are separate programs, and we would tell you if a vendor claimed otherwise. Visibility here is earned the way it is earned in search: by being clear, consistent and cited.

This audit is the first step of our AI search visibility service, and it usually feeds straight into growth & ads work. Related: is your Shopify catalog ready for AI agents? and ChatGPT ads for ecommerce. Or book a free consultation and bring your answers from step 1.