How to check whether AI cites your site
You can check it by hand and without paying anything. What you cannot do is check once and call it settled.
Being named, being cited and being recommended are three things
Only the third one brings you a customer, and adding all three into a single "visibility" number is the most common way to inflate a report.
Before you check anything it helps to know what you are looking for in the answer, because each of the three is detected differently and none of them implies the other two.
| What it is | Where you see it | What it tells you | |
|---|---|---|---|
| Mention | The answer writes your name | In the text | That the model knows you exist |
| Citation | The answer links or attributes something to a page of yours | In the source list | That your content feeds the answer |
| Recommendation | The answer proposes you as an option for the user's case | By reading the whole sentence | That you influence the buying decision |
You can be mentioned and not cited, cited and not recommended, or named as an example of what does not fit. The metrics that come out of separating these three states are in AI visibility metrics.
ChatGPT
Ask it, in a temporary chat, what a customer would ask, and watch whether it searched the web before answering.
- Open a temporary chat. No history, no memory. If you ask from your usual account the model already knows who you are, and the answer stops working as a measurement.
- Ask what a customer would ask, not about your brand. "Best employment lawyer in Valencia" works; "what do you know about my company?" measures nothing, because naming the brand in the question guarantees it shows up in the answer.
- Check whether sources appear. If it searched, you will see links. If not, it is answering from memory and may write a URL of yours it has never read.
- Classify the answer into the three states in the table above, and note which competitors share the list.
Whether it searched or not changes the whole result. In an internal measurement over the same 90 questions, presence was 0% without web search and 22% with it. It is in the methodology, along with the rest of the counting decisions.
That is also why we do not publish a ChatGPT citation rate, even though we have it calculated. The reason is explained on the Perplexity page.
Google AI Mode
Check it signed out, and do not trust a single snapshot: this is the engine whose behaviour has changed most this year.
Between 24 and 31 August 2026, over the same 990 weekly questions, Google AI Mode went from citing 12.22 sources per answer to 4.08, and its answers got 32% shorter. The Share of Answer of the eleven brands on the panel fell from 23.4% to 15.9% in that same week, without any of them having changed anything. The following week it was still at 3.98 citations per answer, so it was not a blip. The detail is in the finding on the citation cut.
If you are only going to check one engine, this is the one that weighs most. Over 17,710 measured answers from a closed panel of eleven brands, Google AI Mode accounts for 53% of every time an AI search engine names the brand, with a 95% confidence interval between 50.9% and 55%. Per brand it ranges from 34% to 78%: the full breakdown here.
Microsoft Copilot
It is the only engine that tells you itself which of your pages it has cited, in the AI performance report of Bing Webmaster Tools.
You do not have to guess by asking: with the domain verified, Bing publishes the number of citations and the cited pages, day by day. It is the cheapest of the four checks and almost nobody looks at it.
An uncomfortable example of our own: between 8 and 15 September 2026 that report returned 0 citations and 0 cited pages for vilnux.com. All eight days at zero. And this is a domain that scores 93 out of 100 on AI readability. It makes the point we make in why AI does not mention you: the technical part is fixed in an afternoon and is almost never what is failing.
Perplexity
It is the easiest engine to check, because it always cites.
In the 8,281 Perplexity answers in our corpus, 100% carried at least one source, with an average of 15.9 sources per answer. Google AI Mode cited in 96.5%, averaging 9.1. You just read the source list and look for your domain. Getting chosen is the hard part: that is in how to appear in Perplexity.
A single check proves nothing
Engines do not repeat themselves: we asked 269 questions three times in a row to the same engine, and in 12 of the 27 cases where the brand appeared it did not appear all three times.
That is 44.4% instability, with a 95% confidence interval between 27.6% and 62.7%. It has two consequences worth swallowing whole. If you look once and you appear, you do not know whether you are recommended. If you look once and you do not appear, you do not know whether you are invisible either. The measurement is in the finding on stability.
What does mean something is the frequency over many runs. With zero appearances in n attempts, the real rate can still be as high as 3/n: a zero over three runs only tells you that you have not measured enough.
Checking one engine tells you nothing about the others
Two AI search engines answering the same question share between 4% and 13% of their sources.
| Engine pair | Pairs compared | Sources in common | Not a single source in common |
|---|---|---|---|
| Google AI Mode and Perplexity | 5,235 | 13.4% | 11.0% |
| Google AI Mode and ChatGPT | 1,873 | 5.4% | 68.7% |
| ChatGPT and Perplexity | 2,004 | 3.9% | 60.3% |
In six out of ten comparisons between ChatGPT and Perplexity there is not one source in common. Winning in one and staying invisible in another is the ordinary case: the full figures.
Traffic arriving from ChatGPT
You can measure it in Google Analytics 4, but part of it will be missing and it helps to know which part.
GA4 added a channel of its own called AI Assistant in May 2026, which recognises ChatGPT, Gemini, DeepSeek, Copilot and Grok. Three warnings before you trust the number:
- Perplexity is not in it, so it still falls into
Referral. If you want it counted you need your own channel group with a rule by source, placed above Referral. - It does not reclassify history. Everything before stays where it was.
- Part of it arrives with no referrer and shows up as direct traffic. There is no way to recover all of it.
And a caveat about what you will read regarding that traffic: the claim that it converts nine times better does not hold up. The published evidence puts it between 13% worse and 31% better than organic, and the largest peer-reviewed study, with 973 sites and 164.8 million transactions, is the one that finds 13% worse. We take it apart in this finding.
How often to repeat it
Weekly if you are going to make decisions with the number, monthly if you only want to know where you stand.
The frequency is set by the noise floor of your sample. With a fixed set of questions, any variation smaller than the margin of that sample is noise and means nothing. Over 90 questions, for example, a 10% presence actually means somewhere between 5% and 18%. Looking daily only adds scares.
What is always worth doing is keeping the same questions. Changing the set and comparing against the previous snapshot is comparing two different things.
What this check does not tell you It does not tell you why. Knowing that you do not appear is the diagnosis, not the cause: the cause is usually outside your own site, in which sources the engine reads when it talks about your sector. It does not tell you what it is worth either: AI visibility does not turn into revenue by itself, and anyone promising you a return figure before measuring is guessing.
Frequently asked questions
How can I tell whether ChatGPT recommends my company?
Open a temporary chat, with no history, and ask the questions a customer would ask, without naming your brand. Repeat each one several times and note in what percentage of answers it proposes you as an option, not just whether it names you. A single screenshot is worthless: when we asked the same question three times, in 12 of the 27 cases we measured the result was not the same.
How do I check whether Google AI Mode cites my site?
Run the questions signed out and review the links that come with the answer. Bear in mind this engine changes a lot: between 24 and 31 August 2026 it went from citing 12.22 sources per answer to 4.08 over the same 990 questions, so a one-off check can tell you more about the engine than about your site.
How do I find out which of my pages Copilot is citing?
In the AI performance report of Bing Webmaster Tools, with your domain verified. It is the only engine that publishes directly how many times it has cited you and which of your pages it used, day by day, without you having to ask it. It is the cheapest check and the one fewest people look at.
How do you check whether a model cites your site as a source?
By looking at the source list of the answer, not the text. Writing your name is a mention; linking a page of yours is a citation, and they are different things. Perplexity cites in 100% of the answers we have measured and Google AI Mode in 96.5%, so with those two you just read the sources and look for your domain.
Can you measure the traffic arriving from ChatGPT to your site?
Yes, in Google Analytics 4, which since May 2026 has a channel of its own called AI Assistant. Part of it will be missing: Perplexity is not in that definition and falls into Referral, earlier history is not reclassified, and sessions arriving without a referrer are counted as direct traffic.
How often should you measure AI visibility for it to be worth anything?
Weekly if you are going to make decisions with the number, monthly if you only want to know where you stand. What matters is always keeping the same questions and not reading every rise as a trend: over 90 questions, a 10% presence actually means somewhere between 5% and 18%.
Check it with your own site
The free audit runs the questions from your sector across several engines and shows you the full answers, with their sample and interval. No signup, no card.
Run my free audit →Home · Plans & pricing · Methodology · Research · Perplexity · Metrics · How to choose · Use cases · Changelog · Agencies · Alternatives · Diagnosis · Noise and margin of error · Four-step method · Optimizing for agents · X · Product Hunt