How to Check If Your Website Shows Up in AI Search

There is a version of this question that has no answer, and it is worth naming before we start.

"Does my website show up in AI search" implies there is a single place to look and a single yes or no waiting there. There is not. Answers change with the wording of the prompt, with what the assistant remembers about the person asking, with the country they are in, and with which surface they opened.

So the honest goal is not a verdict. It is evidence, from more than one direction, that a machine can actually read you.

Method 1: ask the engines, then distrust the answer

Open ChatGPT, Perplexity, Gemini, and Claude. Ask each one the question a real buyer would ask, phrased the way a real buyer would phrase it. Not "tell me about acmeplumbing.com." Something like "best emergency plumber in Leeds" or "cheapest CRM for a two person agency."

Then note two separate things: whether your business is named at all, and whether any of your pages are cited as a source. Those are different results and they move independently.

Now the trap. If you paste your URL and ask "can you read this page," a model can produce a confident, fluent summary of a page it never fetched, drawn from training data or a stale memory of the web. The summary sounds like proof and is not.

The cheap way to catch this: make a small, distinctive edit to the page, wait, then ask again. If the summary still describes the old version, you learned that the model answered from memory. That is a genuine finding, not a failed test. Our step by step version of this probe is in how to test if ChatGPT can read your page.

Repeat the prompt three or four times, in fresh chats. One answer is noise. A pattern is the signal.

Method 2: confirm you are even eligible

Before asking whether an engine chose you, confirm it could have.

Google's Search Central documentation puts the bar plainly: "To be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements. There are no additional technical requirements" (Google Search Central).

So run the boring checks:

• A site:yourdomain.com/your-page probe in Google, to see whether the page is indexed at all.
• Your robots.txt, for rules blocking the crawlers you want.
• The raw HTML the server returns, with scripts not running. If your main content only appears after JavaScript executes, a machine may receive a shell.
• Any nosnippet or noindex directives you forgot were there.

A page failing these is not being skipped by AI. It is being skipped by the index, which is upstream of everything else.

Method 3: read Search Console the way Google says to read it

People go looking for an AI Overviews report and conclude there is no data because they cannot find one.

Google's own documentation says otherwise: sites appearing in AI features "are included in the overall search traffic in Search Console," and "they're reported on in the Performance report, within the 'Web' search type."

There is no separate line. AI feature impressions are already inside the number you have been reading for years, which is exactly why the shift went unnoticed on most sites.

What that gives you is limited but real: impressions and clicks for the queries where you were eligible, on Google surfaces only. It gives you nothing about ChatGPT, Perplexity, or Claude, because none of them report to Search Console.

Method 4: look for the referral, and understand why it is small

In GA4, filter acquisition by session source and look for the assistant domains you care about. A visit that starts inside an AI answer and ends on your site is the cleanest possible evidence that the engine both read you and pointed at you.

Two honest caveats.

First, this is a floor, not a measure. AI referral traffic counts the people who clicked. It cannot count the people who got their answer, saw your name, and never left the chat. SparkToro's zero-click study found that in 2024, 58.5% of American Google searches ended without a click. Answers that resolve in place are the norm, not the exception, and they leave no referral row.

Second, this method only ever shows you the past. It tells you nothing about the page you published last week.

Old way, new way

The old way was one instrument: the rank. It was a single number, it updated daily, and it answered the question by itself.

The new way is four partial instruments, none of which is complete. Prompts sample behavior. The index check proves eligibility. Search Console covers Google surfaces only. Referrals count only the clicks. You triangulate.

That is more work, and it is also why "just check it manually every so often" quietly stops happening after week three.

What the free scan does, stated exactly

Here is the part where most tools get vague, so we will be specific.

Citedon's scan does two things on any URL you give it.

It reads the page the way a machine does and reports readiness: whether the content is reachable and in real text, whether the structured data is present and valid, whether the headings and answer are extractable, whether crawlers are blocked. That is a state of your page, and it is fully in your control.

Then it infers the realistic queries that page should be relevant for, runs them across ChatGPT, Perplexity, Gemini, and Claude, and reports which engines named or cited you on those queries. That is a citation measurement on a sample.

The damaging admission

That second number is a sample, and we will not dress it up as a census.

We query the four engines through their APIs with web search enabled. That is a real read of the live web by a real model, and it is not the same thing as the AI Overview box on a Google results page, or as what one specific logged-in person with three years of chat memory sees at 9pm on their phone. We report the surface we measured, on the queries we ran, on the day we ran it.

Nobody can tell you everywhere you appear in AI answers. Anyone who claims to is selling certainty that does not exist. What is genuinely measurable is readiness, because readiness is a property of your page rather than a property of a model's mood.

Also: if you are not on WordPress, the scan still works and still diagnoses what is missing. The automated fix layer, with a preview and a per-fix approval, runs only through the connected plugin on WordPress. Elsewhere you apply the changes yourself.

And if you run a three page site for a local shop, the manual method above is genuinely enough. Check it twice a year and get on with your life.

What to do with the result

Whichever method you use, write the result down with a date. Readiness is not a photograph, it is a trend line, and a single reading has nowhere to sit.

Then re-check after the things that break it: a redesign, a platform migration, a plugin change, a batch of new pages that shipped without structure. Those are the moments when a page that read cleanly in spring stops reading cleanly, with nothing visibly wrong on screen.

Run a free scan on the page you would most want an engine to quote, and you will have your first dated reading in about a minute.