AI visibility, citation and traffic checker

Use the checker when you want a first, honest read of where your site stands in AI search and you do not yet have a baseline. It suits a business owner who has heard that buyers now ask ChatGPT or Google AI Mode for recommendations and wants to know whether their own site is part of that picture. It also suits a marketing lead who must decide whether to spend on a project and wants numbers from their own data before talking to anyone.

It works best at three moments. First, before you change anything, so you have a dated starting point. Second, about four to six weeks after a round of fixes, to see whether page readiness and crawler access moved. Third, once a quarter as a light check, because engines, interfaces and your own pages all change. Each run takes under an hour if you do the citation test on 15 searches.

It is not the right tool for everything. It does not track competitors, it does not monitor your brand every day, and it cannot read what other engines say to other people. If you need a recorded comparison across several engines on a fixed set of buyer prompts, that is the job of an LLM visibility audit. The checker is the free first pass that tells you whether such an audit is worth doing.

• You have a Search Console property for the site
• You can export the Queries file as CSV
• You can spend about 10 minutes per engine on the manual test

The citation chance signal does not ask any AI engine anything. It reads the wording of each query in your export and scores it. A query that starts with a question word, that sounds informational, or that compares options scores higher. Longer queries of five words or more score higher than short ones, because long conversational searches are where AI answers tend to appear. A query that is clearly a purchase request and not a question is scored down. The score becomes High, Medium or Low.

Brand searches are removed first. The tool matches your domain and your extra brand words against each query, with and without spaces, and marks those as Brand. They are shown in the summary as a share of impressions, but they are left out of every score.

Two measures are then built from what remains. The first is the share of question style impressions that come from positions 1 to 10. It reflects a pattern many practitioners report, that AI answers often draw on pages that already rank well, but treat that as a working assumption and not a law. The second is a click check. If a non brand search sits in the top 10, has at least 30 impressions and earns less than half the clicks a typical click curve would predict, the tool flags it. A click shortfall on a question search can be a sign that an answer box is taking the click, or that your title is weak. The tool cannot tell which.

The priority list is the 25 searches with the best mix of impressions and likelihood, drawn from positions 1 to 20 with at least 10 impressions. If fewer than eight searches qualify, the tool widens the filter so you still have something to test.

The guided test is the slowest part and the most valuable. It is the only signal that looks at what an engine actually shows. Done carelessly, it produces noise. Follow a fixed routine.

Open each engine in a private or incognito window so that your own history does not shape the answer. Stay in the same country for every run, because results differ by place. Use the search exactly as typed by the tool. Do not rephrase it on one engine and not on another. For Gemini the tool cannot open a prefilled link, so you copy the prompt and paste it.

Then read the sources, not just the answer text. Mark Cited if your domain appears as a linked source. Mark Mentioned if your brand is named in the answer without a link. Mark Absent if neither happens. Be strict. A page from a competitor that merely sounds like you is Absent.

Do at least five checks, because the tested citation score only appears from five logged results. Fifteen searches across the engines your buyers use is a better target. Use more engines for a Dubai or London audience than a narrow list, and note the date. The links that open each engine are unofficial and can stop working when an interface changes, so if one fails, paste the search by hand.

Each of the 15 page checks carries a weight, and a pass earns full weight, a warning half and a fail none. Three checks carry the most. Can be indexed looks for a noindex rule or a canonical that points to another page. Text is in the HTML counts the words in the raw page and fails when there are very few words and the page depends on scripts. Structured data looks for JSON LD on the page.

A fail on the HTML text check is the one to take seriously first. Many AI crawlers do not run JavaScript, so a page that builds its content in the browser may look empty to them. If you see it, read our notes on JavaScript and headless sites. The fix is usually server side rendering or pre rendering, which is template work and not a copy edit.

The remaining checks are lighter: a clear title, a meta description, one H1, a direct answer after the H1, headings written as questions, an organization or person entity in schema, sameAs profile links, an author or byline, a visible date, internal links, image alt text and enough depth. Treat them as hygiene. They make a page easier to read and quote. They do not decide a citation on their own.

For schema details, our guide to schema markup for AI search explains what to mark up and what to leave out.

The page already lists the main limits. A few more are worth knowing before you act on a report.

The Search Console interface limits what an export contains. As we understand Google's documentation, the Performance export from the interface is capped at 1,000 rows, so a large site sees only its top queries. If you need the long tail, use the Search Console API or a connected tool, and check Google's current documentation for the limit, since we have not re checked it this session.

The tool scores query wording, not real AI answer data. A high likelihood means the search looks like the kind that gets an AI answer, as of the engines we have seen. It does not confirm that Google shows an AI Overview for that search today. Use the Generative AI report in Search Console and your own manual check to confirm.

The page scan fetches one URL with a short time limit and reads a capped amount of HTML. A slow server, a bot challenge page or a login wall can make the scan fail or read the wrong page. If a scan reports an HTTP error on a page you can open yourself, a firewall may be treating the scanner as a bot. That is itself useful information.

Finally, the tool is not a measure of share of voice. It does not count how often competitors are cited. For that you need a repeated, recorded test, which is what our guide to measuring GEO performance describes.

If the report shows blocked crawlers, read AI crawlers and robots.txt: what to allow and use the robots.txt generator to build a corrected file. If the AI traffic signal is empty, AI search traffic in Google Analytics shows how to set up the export. If your citation test is weak, how to run an AI visibility audit goes deeper than the free test can.

The checker finds problems. It does not fix them. The GEO Pilot is where we do the fixing: a fixed price of $999, one time, delivered inside 30 days. Week 1 is a baseline across five AI engines on up to 20 buyer prompts, a crawler access check and a technical and entity audit. Week 2 implements the technical fixes. Week 3 rewrites up to 10 priority pages to be answer ready. Week 4 tests the same prompts again and hands over a before and after report and a 90 day roadmap. It covers one domain and one language, on any platform, and we promise no ranking, citation or traffic outcome. You can read the details on the GEO Pilot page.