When ChatGPT names your competitors instead of you

It usually happens by accident. A store owner opens ChatGPT to test something unrelated, types a shopping question close to their own category out of curiosity - "where can I buy a good winter jacket" or "best place to order office chairs online" - and the answer names a marketplace, a national chain, or a competitor three times their size. Their own store is not in the list. Not misspelled, not buried at the bottom of a paragraph. Simply absent.

That moment tends to trigger a fast, understandable reaction: panic, then a dozen more prompts typed into the same chat window to see if it was a fluke. It rarely is a fluke, and it also is not proof of anything final. Both things are true at once, and the rest of this article is about telling them apart - so a bad afternoon in a chat window turns into a weekly habit that actually moves the needle.

The answer you did not want

A search engine result page and a ChatGPT answer solve the same shopper problem in structurally different ways. A results page hands over ten or more links and lets the shopper decide. ChatGPT reads the question, decides what a good answer looks like, and writes two to six named options with a sentence of reasoning each. There is no page two, and a store left out of that short list is invisible for that question - regardless of how it ranks anywhere else.

So when a marketplace like Allegro, or a large multi-brand chain, gets named instead of a smaller independent store, it is rarely personal and rarely a sign the model has never heard of the store. It is a sign that, for this specific question, the model reached for something it could confirm confidently and quickly - broad selection, a recognizable name, or a page it could read without friction - and the smaller store did not surface as an equally safe answer. That is a fixable gap, not a verdict. But before deciding what to fix, it is worth checking whether the chat that produced this answer was even a fair test.

Why a personal ChatGPT tab lies to you

Signed-in ChatGPT accounts can use Memory - past conversations, saved preferences, and prior searches can quietly shape a later answer. That is convenient for a shopper, but it makes a single logged-in chat an unreliable way to check visibility. The exact same question can return a different set of stores for a different person, or a different set for the same person a week later, depending on chat history that has nothing to do with the store being tested.

There is a second, less obvious issue: a single answer is one draw from a system that can vary run to run, even without Memory involved. One good or one bad answer does not establish a pattern on its own, in either direction. A store owner who sees their store named once should not relax any more than a store owner who is left out once should panic. What both need is the same thing: a way to ask the question the way a new shopper actually would, repeated often enough to see whether the result is consistent or noisy.

None of this means a personal chat is useless - it is a legitimate first alarm. It just cannot be the instrument used to decide whether a fix worked, or whether there is a problem to fix at all.

Cold questions are the fair test

A cold prompt is a shopping question that never names a brand or a domain - "durable backpack for daily commuting" rather than "is [store] any good". 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, which is why it gets excluded from mention rate - mention rate is the share of tracked prompts where a store gets named this week, and mixing in prompts that already say the store's name would inflate that number for no real reason.

This is exactly what the free Aigely Scan runs: ten open, cold ChatGPT prompts against one domain, once a week, with no card and no account required, returning a scorecard. It is enough to see, this week, whether a store gets named at all on the kind of question a new shopper would actually type - the same kind of question that produced the surprising answer in a personal chat, but asked cleanly, without any history attached. Creating a free account unlocks the full text of those answers, not just the yes-or-no scorecard, which matters for the next step: reading what the answer actually said.

Share of voice, not a single missing mention

One missing mention on one prompt is a data point, not a diagnosis. The more useful number is share of voice: across a set of cold prompts that matter to a store's category, who else gets named, and how often. A store that is missing from one question but named on four others in the same category is in a very different position than a store that is missing everywhere. Share of voice is also what turns "a marketplace beat us once" into something specific - is it the same marketplace every time, is it a specific competitor's product page, or does it change week to week with no consistent winner.

Watching that pattern over several weeks, rather than reacting to a single chat, is what a competitor tracking view is built for - not to prove a store is being wronged, but to show exactly which questions a competitor keeps winning and which ones are still open. For a broader look at how mention rate and share of voice fit together as a weekly measurement, see how to track AI visibility for an ecommerce store.

What to fix when someone else is named

Before assuming the answer is arbitrary, it is worth reading the full text of the reply, not just the list of names. A store's own account unlocks that full answer, and an Answer Explorer view is built specifically for it. Three details are worth checking every time:

• Who is named first. Order tends to signal which option the model treated as the strongest fit, even without an explicit ranking.
• What was said about the named store. A specific reason - "ships nationwide, wide size range, clear return policy" - usually traces back to something the model could read plainly on a page. A vague mention with no detail behind it is a weaker win for that competitor and a smaller gap to close.
• Whether the winning page is even comparable to the missing store's own page. Sometimes the named competitor simply has a category page that states shipping, price, and stock in plain text, while the smaller store's equivalent page is a script-rendered wall of images with none of that stated anywhere a model can read it.

That last point is usually where the real fix lives. A store that never gets named on cold prompts is rarely missing because of one clever trick a competitor used - it is usually missing because a model could not confirm what it needed to confirm on that store's own pages: current price, availability, and a plain description of what the product is and who it is for. The bulk of that work is the same work that has always mattered for search - being crawlable, indexed, and readable. GEO-specific extras, like an llms.txt file or checking that GPTBot and other AI crawlers are not blocked in robots.txt, matter, but they are a small layer on top. They cannot make a page a model can cite if the page itself never states the facts a model needs.

None of this is a guarantee that fixing crawl access and content will flip a specific answer next week. Model answers change on their own timeline, not on command, and a fix made today is tested by asking the same question again in a few weeks - not by refreshing the same chat five minutes later.

A weekly way to watch it

A repeatable method beats reacting to one chat every time a surprising answer shows up:

1. Run the same open questions on a schedule, not on impulse. The free Aigely Scan covers ten cold prompts for one domain once a week with no card required, which is enough to catch a first pattern.
2. Read the full answer, not just whether a name appears. A free account unlocks the complete text behind each result - who is named, in what order, and in what tone - which is what actually explains a competitor's win.
3. Scale up once a store needs more than a weekly snapshot. A 7-day trial (a card is required) runs fifty prompts across ChatGPT, Gemini, and Google AI Overviews together, though not Claude - enough to see mention rate and share of voice across more than one engine at once.
4. Pick a plan that matches which engines a store's shoppers actually use. Starter covers ChatGPT and AI Overviews with automated weekly tracking. Pro adds Gemini. Business adds Claude on top of the rest, for the widest weekly coverage.
5. Fix the pages a competitor's win points to, then wait for the next scan. A schema fix or a rewritten category page is judged on next week's cold prompts, not on the same chat refreshed twice.

The full context behind this - what a cold prompt is, how retrieval and citations decide which store gets named, and why a Google ranking and a ChatGPT mention can diverge - is covered in does ChatGPT recommend your online store. The two-minute way to start is the same either way: run the free Aigely Scan against cold shopping prompts and read the scorecard. For tracking mention rate and share of voice every week, across more engines, and reading the full text behind every answer, compare plans to find the right coverage.