To check your AI visibility, ask four buyer-shaped questions across ChatGPT, Perplexity, Gemini, Claude and Grok, then record two things for each: whether your brand is named in the answer text, and whether your domain appears in the citation row. The gap between those two columns is your actual problem.
You do not need a tool for the first check, and running it by hand once will teach you more than a month of dashboards. The only mistake that matters is the one almost everyone makes on their first attempt.
Because nobody researching a purchase types a brand they have already chosen. Asking an engine about your own company returns a description, not a recommendation, and it will almost always be flattering. It tells you the engine knows you exist, which was never the question.
The question that decides whether you get the customer is the one your buyer actually types, and that question describes a situation rather than a company.
This is not a quirk of one engine. Google's own documentation describes AI features selecting from pages that answer the query rather than pages about a brand, and OpenAI's crawler docs make clear the same retrieval step runs before an answer is written. The engine searches the question first. Your brand name never enters the process.
Run a category prompt, a comparison prompt, a safety prompt and a how-to prompt across five engines. Between them they cover the four moments a buyer reaches for an answer engine. Swap the bracketed parts for your own category and the situation your best customers are in.
Four prompts across five engines is twenty results, and three pulls each because these systems are not deterministic. That is the twenty minutes.
Start with ChatGPT, Perplexity and Gemini, which carry most buyer research. Add Claude if you sell to technical or enterprise buyers, and Grok if your category lives on X. Both are worth checking even at low volume, because they fail differently: Claude hedges when your trust footprint is thin, and Grok swings with whatever is being discussed that week.
| Engine | Why it matters | Shows citations? | Watch out for |
|---|---|---|---|
| ChatGPT | Largest share of buyer research | Yes | Answers vary most between pulls. Run it three times. |
| Perplexity | Most transparent sourcing | Yes | Favours recent pages, so a fresh competitor post can distort one pull. |
| Gemini | Grounds in Google's index | Partial | Blocked entirely if Google-Extended is disallowed in your robots.txt. |
| Claude | Skews technical and B2B buyers | Yes | Conservative on unverified claims, so a thin trust footprint shows up here first. |
| Grok | Weighted toward live X discussion | Partial | Moves fastest and reverts fastest. Useful as a sentiment read, poor as a baseline. |
Split every answer into two columns. Left column: which brands are named in the prose. Right column: which domains appear in the citation row underneath. Being in the right column only means you supplied the research while somebody else got recommended.
Named and cited. You are in good shape. Keep the pages that earned it fresh and monitor for slippage.
Cited but not named. The most common outcome and the most expensive. Your content is good enough to source and your entity is not strong enough to recommend. Work on third-party presence and category positioning rather than more blog posts.
Neither. You are not in the candidate set at all. Find the roundups that rank for your category question and get onto them. That is the whole job until it changes.
Automate once you need the answer to be comparable month over month. Doing it by hand tells you where you stand today. It cannot tell you whether last quarter's work moved anything, because you will not have run identical prompts under identical conditions.
The manual check has one real weakness. Answer engines sample, so your twenty results are a snapshot taken under conditions you cannot reproduce next month. That is fine for a diagnosis and useless for measuring progress.
This is the point where a tool earns its cost, and there are several good ones. We compare eight of them, including our own with the conflict disclosed, in the best AI visibility tools guide.
citeOS is the crypto-specific option. It runs 20 buyer-shaped prompts across five engines for 100 observation points per audit, scores mention at 0.60 against citation at 0.20 because being named is worth roughly three times being cited, and publishes every weight in the methodology. If you are not in crypto, one of the horizontal tools will serve you better and the guide above says which.
The free scan runs the same check automatically across all five engines and takes about a minute. No signup, no card.
Run the same check across five engines, automatically.
Scan my brand freeA way of testing whether AI answer engines name your brand when someone asks a buying question in your category. It can be a paid tool or, as described here, four prompts run by hand across three engines in about twenty minutes.
Monthly is enough for most brands. Answer engines move faster than that, but the underlying sources they draw on move slowly, so weekly checking mostly measures noise unless you have just shipped something.
Answer engines sample rather than look up, so identical prompts return varying responses. Run each prompt three times and treat the pattern as the signal. A brand named in two of three pulls is genuinely in the candidate set.
Less than most teams expect. Your site needs to be crawlable and clearly written, but in our own measurement 89% of citations pointed at domains the brand does not own. Third-party pages carry most of the weight.
Yes. The method on this page costs nothing but your time. citeOS also runs a free scan across five engines if you want the same diagnosis produced automatically.
Growth marketer working in Web3 since its early days, and the person who built citeOS. The figures on this page come from our own measurement work: 4,969 AI citation events across 10 audited crypto brands, 5 engines and 979 source domains.
See your citation share across ChatGPT, Perplexity, Gemini, Claude and Google AI Mode. Free scan, about a minute, no signup.
Scan my brand free →