Does ChatGPT Recommend My Business? How to Check

You check whether ChatGPT recommends your business by asking it the same questions a prospective customer would ask, then recording whether your name appears, how it is described, and which competitors appear next to it. Nothing else gives you a direct answer, because OpenAI does not publish a mention report for brands.

The catch is that a good test has rules. You need a spread of question types, a clean session, more than one run per question, and a look at the sources behind each answer. Skip those and you end up with an anecdote rather than a finding. The steps below take about twenty minutes for a first pass.

If you want a starting point before you open ChatGPT, the free AEO Grader scan gives you an AEO score, an action plan and a competitor snapshot in about a minute. Be clear about what it is: it runs one Gemini check grounded in Google Search, and its per-engine view is an estimate, not a query to ChatGPT. The manual method below is how you see ChatGPT itself. For the content side of the problem, our ChatGPT optimization checklist covers what to change once you know where you stand.

Most owners type their company name into ChatGPT and stop there. That only tells you whether the model knows you exist. Buyers rarely start with your name; they start with a need. A useful prompt set covers four kinds of question, each testing something different, and you need two or three of each.

Category questions ask for the best options without naming anyone. They are the hardest test, because a mention has to be earned. Comparison questions put you next to a named rival. Problem questions describe a situation in the buyer's words and see who gets suggested as the fix. Brand questions ask about you by name and reveal whether the facts are right.

Write them the way a customer talks, including your city if you serve a local market. OpenAI's web search guide notes that the model can choose to search the web or not based on the content of the prompt, and that developers can pass an approximate user location to refine results. Both details matter: a vague prompt may be answered from memory alone, and an answer can change with where the question comes from.

Assistants increasingly remember the person asking. Google announced in August 2025 that the Gemini app can reference your past chats to learn your preferences, with that setting on by default. Anthropic's help center says Claude's memory is on by default for Free, Pro and Max plans. If you have spent months asking an assistant about your own company, it may be primed to mention you in a way a stranger would never see.

So run every prompt twice: once in your normal logged-in account, and once in a logged-out browser window with no history. The logged-out run is closer to what a new buyer sees. The logged-in run shows how much your own history is colouring the answer. If the two disagree sharply, trust the clean session.

Two more controls help. Turn search on, or phrase the question so it clearly needs current information, such as asking for options available this year. And keep the wording identical between runs, because small phrasing changes produce different searches and therefore different answers.

A single answer is a sample, not a verdict. Researchers at Thinking Machines Lab put it plainly in their work on nondeterminism in LLM inference: asking ChatGPT the same question multiple times provides different results, and even with temperature set to zero, LLM APIs are still not deterministic in practice. Your brand can appear in one run and vanish in the next without anything having changed on your side.

Run each prompt at least three times in the clean session and log the result in a simple sheet: the prompt, the run number, whether you were named, where in the answer you appeared, which competitors appeared, and which sources were linked. Then summarise per prompt as a count, such as "named in two of three runs". Counts like that are honest about the variation and easy to compare next month.

Resist the urge to rerun until you get the answer you like. If a category prompt names you once in five tries, that is the finding.

When ChatGPT searches, it shows the pages it used. Those links tell you more than the answer text does, because they show what the model was reading when it decided who to name. Click through every source on your category and problem prompts. You will usually find a short list of review sites, directories, comparison articles and forums that the model leans on again and again.

Three patterns are worth noting. If your own site is never among the sources, check that ChatGPT's search crawler can reach it: OpenAI's crawler documentation states that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. If the sources are third-party lists that leave you out, those lists are your outreach targets. And if a source describes you wrongly, such as an old price or a service you dropped, fix it at the source rather than hoping the model will work it out.

Keep a note of every cited domain. Over a few prompts the same names repeat, and that list becomes a practical map of where your citations have to come from.

Once you have a few runs logged, the results usually fall into one of five patterns. Each points to a different next step, so it is worth naming which one you are in before changing anything. The patterns are about the gap between being known and being recommended: a model can describe you accurately when asked by name and still never suggest you to someone who has not heard of you.

ChatGPT is one engine among several your buyers use, and the others build answers from different evidence. Perplexity runs its own crawler; its documentation says PerplexityBot is designed to surface and link websites in Perplexity's search results. Anthropic warns that disabling Claude-SearchBot may reduce a site's visibility in user search results. Google's Gemini grounding documentation describes connecting Gemini to real-time web content through Google Search. Microsoft, meanwhile, launched AI Performance in Bing Webmaster Tools in February 2026 to show citations across Copilot and Bing's AI summaries.

Different crawlers and indexes mean a business can be recommended by one engine and missing from another on the same day. We unpack the reasons in why AI engines give different answers about your brand. If you repeat the four steps above on every engine, the work multiplies quickly: four engines, a dozen prompts, three runs each. That is the point where most teams either build a spreadsheet routine, as described in our guide on measuring brand visibility in ChatGPT, or hand the asking to a tool.

The $10 AEO Grader report does the asking for you. It sends the same panel of buyer questions to ChatGPT, Claude, Perplexity and Gemini live, with each engine's web search switched on, and shows every engine's actual answer with the sources it cited. The questions cover what you do, whether you are trustworthy, how you compare with competitors, the alternatives to you, and the best options in your category.

Before counting a mention, the report checks that the answer is about your company and not a business with a similar name, and it labels answers that describe a namesake. It then gives you visibility per engine, your AI share of voice against the competitors the engines named, how often each engine cited your own site, and a site audit with a prioritized fix list. One re-scan within 14 days is included, so you can check whether a fix moved anything. If you want the check repeated, monthly tracking at $9 a month runs a fresh four-engine report each month and emails you what changed per engine. No tool can promise that ChatGPT will recommend you; what the report gives you is an accurate picture of where you stand.

Start with the free AEO Grader scan for a baseline, read the methodology to see how the score is built, and if you are still choosing what to fix first, our guide on ranking higher on ChatGPT lays out the content work.