How often do people click a citation in an AI answer?

Rarely, on the one surface with a public measurement. In a Pew Research Center study of 68,879 Google searches during March 2025, users clicked a link inside an AI summary in 1% of visits. Treat the citation as visibility, not as a traffic channel.

Does a citation mean the model actually read my page?

In a retrieval-based answer, usually yes: the system fetched candidate documents and the model wrote from them. In an answer generated purely from trained parameters there is no retrieval step and no page fetch, so any URL that appears is a reconstruction from memory rather than a record of a visit, and can be wrong.

What is the difference between a citation and a mention?

A citation attaches a source to a statement, so a reader can check it and follow it. A mention names you inside the prose with nothing attached. Both count as visibility; only the citation gives the reader a route to you, which is why they should be counted separately.

Can I mark up my content to be cited more often?

There is no citation markup. Google states plainly that there are no additional requirements to appear in AI Overviews or AI Mode and no other special optimizations necessary; eligibility comes from ordinary indexing and snippet eligibility. What you can control is whether crawlers may fetch the page and whether the page contains statements specific enough to be worth attributing.

Why do assistants cite different sources for the same question?

Because they run different retrieval systems over different indexes with different crawlers, and each vendor documents its own crawler and its own robots.txt policy. Two assistants asked the same question are querying two different pools of documents, so agreement between them is a signal rather than an expectation.

Source citation - AI SEO & GEO Glossary

A source citation is the attribution an AI system attaches to something it has said: the link, footnote or supporting-link card that connects a statement in generated prose back to a document. It is the closest thing AI search has to a ranking, and it behaves very differently from one.

Why it is its own outcome

A citation is not a click, and on the one surface where the behaviour has been measured publicly the gap is large. The Pew Research Center analysis of 68,879 Google searches during March 2025 found that users clicked a link inside an AI summary in 1% of visits. The citation is therefore a visibility outcome rather than a traffic channel, and reporting it inside a sessions chart guarantees it looks like a failure.

It is also not a mention. A mention names you in the prose. A citation attaches you to a claim, which means a reader can verify it and follow it. Counting them together loses the distinction that actually predicts whether anyone arrives.

Where citations come from

The mechanism differs by how the answer was produced, and this is the single most useful thing to understand about them.

Retrieval-based answers. The system fetches candidate documents at query time and the model writes from what it read. Citations here correspond to documents that were genuinely retrieved. This is retrieval-augmented generation, and it is why crawler access is upstream of everything: a page that the retrieval crawler cannot fetch is not in the candidate pool at all. Vendors run separate agents for this job — OpenAI documents OAI-SearchBot as the one used to surface websites in ChatGPT's search features, and Perplexity documents PerplexityBot as designed to surface and link websites in its results and explicitly not used to crawl for foundation models.

Parametric answers. The model answers from what it learned during training, with no fetch. There is no document to attribute, so any URL that appears is reconstructed from memory and may not exist. This is one origin of a fabricated citation, and it is why an unverified URL in an AI answer is not evidence that a page was read.

The difference is set out in AI training vs AI retrieval, and it is the reason the same question can produce a solid citation on one assistant and a broken one on another.

What you can and cannot control

There is no citation schema and no opt-in file. Google's AI features documentation states that there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary, and that a page must be indexed and eligible to be shown with a snippet to appear as a supporting link.

What is genuinely in your hands is narrower and more concrete:

Access. Whether the retrieval crawler may fetch the page at all, decided in robots.txt per user agent.
Serving rules. Whether a nosnippet or restrictive max-snippet rule removes the passage from use. Google states the max-snippet limit applies to AI features too.
Attributability. Whether the page contains statements specific enough to be worth attributing. A sourced number, a named mechanism or an explicit comparison can be cited; a paragraph of positioning cannot, because there is nothing in it that needs a source.

How to measure it

Citations cannot be read out of your own logs, since most cited visits never arrive. The workable method is sampling: fix a prompt set that matches the questions your buyers ask, run it on a schedule across the assistants they use, and record for each answer whether you were cited, merely mentioned, or absent. Track those three states separately and watch the distribution move. Attensira does this and, where the gap is something a page could fix, opens a pull request against your repository for you to review.

Expect disagreement between assistants. Each runs its own crawler over its own index, so two of them answering the same question are drawing from two different pools.

Frequently asked questions

How often is a citation clicked?

In Pew's March 2025 dataset, a link inside a Google AI summary was clicked in 1% of visits.

Does a citation prove the model read my page?

Only for retrieval-based answers. A parametric answer can produce a URL it never fetched.

Is there markup that makes citation more likely?

No. Eligibility follows from ordinary indexing and snippet eligibility.

Why do assistants cite different sources?

Because each runs its own crawler over its own index.

Source citation