Does ChatGPT recommend your online store?

A traditional search results page gives a shopper ten (or more) links and lets them decide. A shopping question asked inside ChatGPT works differently: the model reads the question, decides what a good answer looks like, and writes a short list of two to six stores or products with a sentence of reasoning for each. There is no scroll, no page two, and no "position eleven" to climb into next month.

That single design choice changes what "visibility" means. On Google, a store can rank on page three and still get found by a persistent shopper. In a ChatGPT answer, a store that is not named in that shortlist is invisible for that question, no matter how strong its organic rankings are elsewhere. Shoppers also often follow up - "what about something cheaper" or "does it come in black" - and each follow-up is a fresh chance for the model to add, drop, or reorder stores based on the same underlying process. The practical question for a shop owner is no longer "where do we rank" but "were we named, and how".

Why shopping answers are not ten blue links

This is the shift worth internalizing before anything else: a shop's job is no longer to climb a results page, but to earn a place in a very short, written answer.

Cold prompts vs brand keywords vs ChatGPT Memory

A cold prompt is a shopping question that never mentions a brand or domain, for example "best waterproof hiking boots for wide feet" or "gift idea for a coffee lover under 50 euros". Cold prompts are the fair test of discovery: they show whether a model reaches for a store on its own, without being pointed at it.

A branded prompt already contains the store's name or domain ("is [store] legit", "[store] return policy"). These are useful for other reasons, but they say nothing about discovery, since the shopper already knew the store existed. That's why branded prompts are excluded from mention rate: mention rate is the share of tracked prompts where a store is named this week, and mixing in prompts that already name the store would inflate it artificially.

Signed-in ChatGPT accounts can use memory - past conversations and saved preferences can quietly shape a later answer. That's convenient for a shopper, but it makes a single logged-in chat a poor way to test visibility: the same cold prompt can return different stores for different people, or different stores for the same person a week apart. Testing what a new shopper would see, without history steering the answer, needs a context where memory isn't shaping the result.

How a model actually picks a store

It helps to separate three mechanics that decide whether a store shows up in a shopping answer.

Retrieval

For most current shopping questions, the model does not rely only on what it learned during training. It runs a retrieval step: it looks up current pages relevant to the question, then compares what several of them say before writing an answer. A store's product pages have to be crawlable and indexed for that retrieval step to find them at all. If a category page returns errors, blocks crawlers, or hides prices and availability behind JavaScript that never renders for a bot, it is competing with one hand behind its back before content quality even enters the picture.

Citations

When an answer names a store, it has often pulled the specifics - price, material, sizing, availability - from a specific page it can point to. That's a hint for shop owners: pages that state facts plainly, with a matching Product and Offer schema, are easier for a model to read confidently and cite. A blog-style page that never states a current price in machine-readable form is harder to trust than a product page with clean structured data.

Recency

Stale pages - discontinued products still listed, prices that don't match schema, sitemaps that haven't updated in months - lower a model's confidence, in the same way an outdated page erodes a shopper's trust. Freshness contributes to what shows up in an audit as a content score.

Behind all three sits a simpler gate: whether AI crawlers can reach the site at all. GPTBot, Google-Extended, and ClaudeBot are among the crawlers that gather the material these systems draw on, and a llms.txt file is one way to state, in a format built for that purpose, which parts of a site are meant to be read. None of this guarantees a mention on its own - it removes reasons to be skipped.

What to read in one answer

A single AI shopping answer carries more signal than a yes/no on visibility. Four things are worth reading closely:

• Named or not. The baseline check: does the store appear at all for this cold prompt this week.
• Order. Stores named first are usually presented as the strongest fit for the question, even without an explicit ranking.
• Citations. Is the store linked or referenced as a source, or just mentioned by name with no link back. The full text of the answer - not just a mention count - is what a tool like an Answer Explorer view is for.
• Tone. Recommended outright, mentioned as one option among several, or flagged with a caveat ("check current stock before ordering"). Tone changes what a mention is worth.

Comparing tone and order across a set of prompts over several weeks is also how a store starts to see its share of voice: which competitors get named on the same questions, and how often.

When Google rank and ChatGPT diverge

It is common to see a store rank well in traditional search and still be missing from ChatGPT's answer to a closely related question, or the reverse: a smaller store with clean structured data outranked on Google but named ahead of larger competitors in an AI answer.

The reasons overlap with the mechanics above. Search ranking rewards backlinks, historical authority, and years of accumulated signals. An AI answer rewards a page a model can retrieve, read with confidence, and cite without extra effort - which a newer store with well-tagged product pages can sometimes do better than an older store with thin or inconsistent product data. A store that has invested in years of link-building but still lists prices only inside an image, or blocks the crawlers that feed these answers, can end up ranking well and being named rarely. Neither system replaces the other; they are simply answering different questions ("what pages match this query" versus "what should I tell this shopper right now"), and a store's job is to be legible to both.

A practical weekly method

A simple, repeatable method beats a one-off check:

1. Start with a baseline. The free Aigely Scan runs ten open shopping prompts against ChatGPT for one domain, once a week, with no account or card required, and returns a shareable scorecard. It's enough to see, this week, whether a store is named at all on cold prompts in its category.
2. Decide if history matters. A single scan shows a snapshot. Tracking mention rate and share of voice over several weeks - and across more than one AI engine - needs a running project rather than a one-off check.
3. Scale up if the store needs weekly tracking. A free account keeps one project with ten prompts on ChatGPT and one manual scan a week, with no automated weekly job. A 7-day trial (a card is required) unlocks fifty prompts across ChatGPT, Gemini, and Google AI Overviews together, though not Claude. Paid plans add automated weekly tracking: Starter covers ChatGPT and Google AI Overviews, Shop adds Gemini, and Business adds Claude on top of the rest.
4. Act on what the answers show. Use the audit to check llms.txt, crawler access, and Product/Offer schema on the pages actually named or missing from answers - by default that's every URL in the store's sitemap, or a specific list of URLs when a shop wants to focus the audit on a smaller scope.

What not to do

• Don't seed the brand name in the test prompt. Asking "recommend stores like [store name]" or including the domain defeats the purpose of a cold prompt and will not show what a new shopper sees.
• Don't judge visibility from a personal, signed-in ChatGPT tab. Memory and prior chats can steer the answer toward or away from a store for reasons that have nothing to do with that store's actual visibility to a new shopper.
• Don't treat one answer as the final word. Model answers vary week to week and prompt to prompt; a mention rate built from a tracked set of prompts over time is more reliable than any single chat.
• Don't confuse a Google ranking with an AI mention. They are measured differently and can move in opposite directions, as covered above.

Checking where a store stands starts with a two-minute look: run the free Aigely Scan against cold shopping prompts on ChatGPT and read the resulting scorecard. For a store that needs to track mention rate and share of voice every week, across more AI engines, and see the full text behind every answer, compare the plans to find the right coverage.