When a buyer asks ChatGPT “what’s the best tool for this?”, the answer is a shortlist. A few products are named, usually with a sentence on why each one fits. If your brand is on that list, you’re part of the decision. If it isn’t, the buyer may never hear of you in that conversation.
Checking where you stand sounds simple: open ChatGPT and ask. In practice, the obvious way to ask gives a misleading answer. This guide covers a check you can trust: which questions to ask, what to record, why ChatGPT and Gemini can disagree, and how much a handful of answers can actually tell you.
Why AI recommendations matter for buyer discovery
A search results page lists ten links and lets the reader choose. An AI answer does the choosing first. It reads sources, decides which options are worth naming, and presents a short list with reasons attached. The buyer sees the conclusion, and often never opens the sources behind it.
That makes the shortlist the moment that counts. A brand that’s left off isn’t ranked lower, it’s simply not in the conversation, however strong its website or its search rankings are. And because the answer arrives as a recommendation, the brands it names start with credibility the others have to earn back.
So the useful question isn’t “does ChatGPT know my brand?” It’s “when a buyer asks for options in my category, does the answer recommend me?”
Recommended, mentioned, or absent
An answer can include your brand without recommending it. For a buyer, the difference matters, so it’s worth recording each answer in one of three states:
- recommended
- The answer puts the brand on its shortlist for the buyer's question: named as an option to choose.
- mentioned
- The brand appears in the answer, but only in passing, not as an option the answer recommends.
- absent
- The brand doesn't appear in the answer at all.
- Cited source
- A web page the answer used as evidence. A source is not an outcome: a brand's own site can be cited in an answer that recommends a competitor.
A mention often sounds like a qualification: “X is another option, but it’s mainly built for enterprise teams.” The brand is visible, but the answer is steering the buyer elsewhere. Counting that as a win hides the real problem. An answer that names you as one of the options to choose is a different result, and it’s the one that moves a buyer.
The fourth row matters too. AI answers that search the web cite the pages they drew on. A cited page is evidence the answer used, not a verdict on any brand. Your own site can be cited in an answer that recommends a competitor.
Start from buyer questions, not your brand name
The most common way to check is to ask “What do you know about [brand]?” or “Is [brand] any good?” Both name your brand, so the answer is about you by construction. That tests whether the model can describe you, not whether it would put you forward when a buyer hasn’t heard of you yet.
Buyers rarely start with your name. They start with a problem, a category, or a competitor they already know. Useful questions come in a few shapes:
| Question type | Example | What it tests |
|---|---|---|
| Category | “What’s the best scheduling tool for a small marketing agency?” | Whether you make the default shortlist |
| Alternatives | “What are good alternatives to [a competitor]?” | Whether you’re seen as a substitute for a known name |
| Use case | “Which tool can schedule posts across Instagram and LinkedIn from one calendar?” | Whether you’re linked to a specific job to be done |
| Segment | “What do agencies with a small team use to plan social content?” | Whether you’re recommended for the buyer you actually serve |
Write them the way a buyer would type them: no brand name, plain language, and the context a real buyer adds, such as team size, market, or the constraint that matters to them. Five well-chosen questions cover more ground than twenty variations of the same one.
How to check it yourself
You can run a small check by hand in under an hour:
- Write five buyer questions across the shapes above.
- Use a clean session. ChatGPT’s memory and your past conversations shape its answers. A temporary chat, or a logged-out session, comes closer to what a new buyer sees.
- Ask each question once, exactly as written, and copy the full answer, including any sources it cites.
- Record three things per answer: the state (recommended, mentioned, or absent), which brands were recommended instead, and which sources appeared.
- Repeat the same questions in Gemini.
- Date everything. Answers change as models and the web change, so an undated result can’t be compared with anything later.
The manual method works, but it’s slow, hard to keep consistent, and easy to bias. You know your own brand, so it’s tempting to read a passing mention as a recommendation.
Why ChatGPT and Gemini can give different answers
It’s natural to expect two assistants to agree on the best options in a category. Often they don’t. They use different models, different search systems, and different sources, so the same question can produce different shortlists.
Crendo’s public sample audit shows this on a small scale. It asked five buyer questions about social media scheduling tools, and Postiz was recommended in four of the five ChatGPT answers and three of the five Gemini answers:
Audit snapshot · measured Oct 18, 2026
Postiz was recommended in 7 of 10 answers to 5 buyer questions.
- ChatGPT 4 recommended · 0 mentioned · 1 absent
- Gemini 3 recommended · 0 mentioned · 2 absent
A one-answer difference across five questions isn’t evidence that one assistant favors a brand. What it does show is that checking a single assistant leaves part of the picture out. Buyers use both, and a brand can be on one shortlist and missing from the other for the same question.
Why cited evidence matters
When an answer searches the web, the sources it cites show where it drew its information from: comparison articles, review sites, forum threads, documentation, sometimes your own pages. They’re the most concrete thing you can look at when trying to understand an answer.
They need careful reading, though. A cited source tells you what the answer used, not why it chose the brands it chose. It’s tempting to conclude “this review site is cited, so getting listed there will get us recommended”. The data doesn’t show that. The sources are a place to start investigating: which pages describe your category, which competitors they feature, and whether what they say about you is accurate and current.
Repeatable measurement and the limits of a small sample
The value of a check comes from repeating it the same way. Same questions, same assistants, the same way of classifying answers, so that when a result changes, you know the answer changed and not your method.
This is also why consumer apps are a shaky baseline. The ChatGPT app adds memory, personalization, and session context, so two people can get different answers to the same question. Measuring each provider’s grounded answer through its API, with web search on, gives a more stable baseline. It still won’t match what every individual sees in their own app.
Be honest about scale. Five questions across two assistants is ten answers. That’s enough to spot clear problems: you’re absent from the category question in both assistants, or the same competitor is named every time you’re not. It isn’t enough to compute a meaningful percentage or to read a one-answer change as a trend. Describe results as what they are, “recommended in 4 of 5 answers”, rather than as a rate.
How Crendo checks your brand
Crendo turns this method into a guided audit:
- Add your website. Crendo reads it to infer your category, your offering, and who you sell to.
- Choose your competitors. Review the suggested competitors or add your own.
- Confirm the buyer questions. Crendo proposes questions across category, alternatives, use-case, and segment intents. You pick the five that matter and can edit them before anything is measured.
- See where you stand. Crendo asks each question to ChatGPT and Gemini and shows, for every answer, whether your brand was recommended, mentioned, or absent, which brands were recommended instead, and the sources each answer cited.
The report shows where the two assistants differ and suggests what to look at first, based on the answers and their evidence. Each answer is stored with its sources, so the same questions can be checked again later. The first audit is free and needs no signup.
Crendo measures what AI answers say. It doesn’t change them, and no tool can promise that an assistant will start recommending a brand. What a measurement gives you is a clear, dated starting point and evidence to work from.
Frequently asked questions
Can I just ask ChatGPT whether it recommends my brand?
You can, but naming your brand in the question tells the model what the answer is about. It shows whether ChatGPT can describe you, not whether it would put you on a buyer’s shortlist when they haven’t heard of you. Questions without your brand name are the better test.
How many questions do I need?
Five is enough for a first snapshot that shows clear gaps, as long as the questions cover different buyer intents. More questions give more coverage, but they don’t turn a small sample into a statistic.
If my website is cited, does that mean I'm recommended?
No. A citation shows the answer used your page as evidence. The same answer can still recommend a competitor. Check the outcome and the sources separately.
Why does my ChatGPT app give a different answer?
The app layers memory, personalization, and session context on top of the model. A measurement through the API, with web search on, gives a repeatable baseline instead of one person’s session.