What is AI search visibility, and how do you measure it?

Why does this need its own definition?

Because the familiar instruments stop working. For twenty years “visibility” in search meant position on a results page: rank first, get the click. AI answers break that chain in two places. First, there is no list of ten: the engine writes one answer and attaches a few citations, so you are either a source or invisible. Second, the sources it picks do not line up with classic rankings. The same question put to Google’s AI Overview, Google’s AI Mode and ChatGPT will often return three different sets of citations.

Google’s own documentation on AI features in Search sets the floor: a page must be indexed and eligible to show a snippet before it can appear in an AI Overview or AI Mode, and there is no separate AI index or special markup to apply. That is the part classic SEO still owns. What it does not tell you is whether, given a hundred eligible pages, yours is the one that gets quoted. That is what AI search visibility measures.

How does an AI answer engine pick its sources?

In three steps: retrieve, rank, cite. A question, often long and specific, is expanded into several sub-queries. Each sub-query retrieves candidate passages from the engine’s index (Google’s for its AI features, Bing’s index in ChatGPT’s case, Perplexity’s own index for Perplexity). A re-ranking step discards most candidates on quality, freshness and relevance. The model then writes an answer from the survivors and attaches citations to the passages it leaned on.

Three consequences follow for anyone trying to be cited:

• The sub-queries are the real targets. A page that only answers the head question misses the cost, comparison, legitimacy and timeline questions the engine quietly asks alongside it.
• The passage, not the page, is the unit. A 40 to 60 word paragraph that directly answers a question is what gets lifted. A page that builds to its point over a thousand words rarely survives the re-rank.
• Trust comes from outside the page. Engines favour brands they have seen named on other sites they already trust. Mentions, linked or not, feed that.

OpenAI’s bot documentation and Perplexity’s crawler guide both separate the crawler that builds the search index from the one that fetches a page on a user’s behalf, which is why robots.txt decisions have to be made per bot, not per vendor.

What does “visible” actually mean in a measurement?

Five things, scored separately for each engine:

1. Mention rate: the share of runs in which the brand is named at all, with or without a link.
2. Citation rate: the share of runs in which one of your URLs is attached as a source.
3. Position: where in the answer the mention or citation appears. First is worth more than fifth, roughly on a harmonic scale.
4. Sentiment: whether the engine describes you neutrally, favourably or with caveats.
5. Accuracy: whether what it says about you is true. Engines confidently repeat stale pricing and discontinued features.

Mention and citation need different fixes. Citation follows topical authority and passage structure on your own site. Mention follows off-site presence: press, reviews, forums, video. They overlap less than people expect, and a brand can be heavily mentioned while rarely cited, or the reverse.

How do you build the instrument?

Write a prompt library before you change anything on the site, so you have a baseline. Our working recipe:

• 25 to 50 prompts in the words a buyer would use, 70 words if that is how they ask. Roughly a third category questions (“best [category] for [situation]”), a third comparisons (“[A] vs [B]”, “[A] alternatives”), a third use-case questions (“how do I [outcome] with [constraint]”).
• Run each prompt three to five times per engine on each cycle, because answers are non-deterministic. Score the aggregate, not a single run.
• Weekly for priority prompts, monthly for the rest. Citations swing 40 to 60 percent month over month in our experience, so look at quarterly trends before reacting.
• Report Google AI Overviews and Google AI Mode as separate lines. They usually reach the same conclusion but rarely cite the same pages. A blended score hides which one is moving.
• Fix analytics attribution. Most AI referral sessions arrive without a referrer and land in “Direct”. A channel group matching chatgpt.com, perplexity.ai, copilot.microsoft.com, gemini.google.com and claude.ai recovers some of it. Server logs show which AI crawlers actually fetch your pages; analytics never will, because bots do not run scripts.

Tooling ranges from a spreadsheet and a disciplined assistant to paid trackers. The method matters more than the tool.

What moves the number?

The same handful of things across engines, in roughly this order of leverage:

1. Eligibility. Indexed, fast, server-rendered, not blocked for the search bots that matter. Nothing else works until this does.
2. Answer structure. Question-shaped headings; a direct 40 to 60 word answer under each; about 150 words per section; specific numbers; a source link on every claim; clean H1 to H2 to H3 hierarchy.
3. Fan-out coverage. A page or section for each sub-question, especially cost, comparison, legitimacy and timeline, which most sites skip.
4. Freshness. Recently updated pages are cited far more often, particularly by Perplexity. Updates have to be substantive; changing a date is detectable and earns nothing.
5. Off-site mentions. Trade press, YouTube, Reddit and review platforms. Being named where the engine already looks is the strongest single lever for mention rate.

Structured data helps Google’s own AI features understand your entity, and is worth doing, but third-party engines read visible text. Any fact you want quoted has to be in the body copy, not only in a script tag. And one tested tactic reliably backfires: keyword stuffing reduces visibility in AI answers, in the one controlled study that measured it.

Where does this leave classic SEO?

As the foundation. Google’s guidance is explicit that its AI features draw on the same systems as Search and reward the same fundamentals described in its SEO starter guide. What changes is the scoreboard: positions and clicks on one line, citations per engine on another, and enquiries, the only line that pays, at the top. That is how we report AI search visibility work, alongside the SEO that makes it possible.