How to measure AI search visibility. The part most agencies skip.
Almost every agency now claims to “do AI search”. Far fewer can prove it moved anything. This is the exact method we use to measure whether ChatGPT, Perplexity, Gemini and Google AI Overviews are finding, reading, citing and recommending you, so AI visibility becomes a number you can track, not a promise.
Five metrics, measured together.
You measure AI search visibility across five linked metrics: AI referral traffic (visits that arrive from AI engines, isolated in GA4), share of AI voice (how often you appear in the answer for a fixed set of buyer prompts), citation tracking (which pages and mentions the engines quote), AI crawler coverage (which pages the AI bots read, from server logs), and answer accuracy (whether the AI describes you correctly). Reported together and tied to enquiries, they turn being cited by AI from a claim into something you can actually move.
Why it matters now: by late 2025, 51% of B2B software buyers said they start their research with an AI chatbot more often than with Google, up from 29% earlier that year (G2). If half your buyers begin inside an AI answer, you cannot afford to guess whether you are in it.
AI referral traffic: the visits AI already sends you.
GA4 does not separate AI traffic by default, so most of it disappears into Referral or Direct. The fix is a custom channel group that matches the AI engine domains by regex. Once it is live, every visit that began inside an AI answer is grouped, attributed and tied to enquiries and revenue, exactly like organic or paid. Small numbers today, but the highest-intent traffic on your site and growing fast.
Share of AI voice: how often you are the answer.
We test a fixed set of the real questions your buyers ask across ChatGPT, Perplexity, Gemini and Google AI Overviews, then track how often you show up, in what position, and against whom. This is the metric that tells you whether GEO and AEO work is actually landing, because a page can rank on Google and still never make it into the AI answer that buyers now read first.
Citation tracking: which pages earn the mention.
Being in the answer and being named as the source are different things, and the second is what compounds. Citation tracking shows the pattern: the formats, the data, the coverage and the entities the models reuse. That tells us which content to expand and which digital PR angles to chase next, instead of publishing on a hunch.
AI crawler coverage: which pages the engines actually read.
Before AI can cite you, its crawler has to read you. Server log analysis shows exactly which pages the AI bots fetch and how often, the leading indicator of citation. It also lets us set robots.txt deliberately: allow the search crawlers that build the answer indexes (OAI-SearchBot, Claude-SearchBot, PerplexityBot) while you decide separately about the training crawlers, because OpenAI and Anthropic let you control those independently.
Answer accuracy: is the AI describing you correctly?
Visibility is worthless if the answer is wrong. Models stitch your description together from sources across the web, and stale or conflicting information produces a confident, wrong answer that costs you the deal. We track accuracy and consistency, then tighten the structured data, profiles and citations the engines draw from so there is one true story everywhere.
What you actually get.
On tooling: we combine GA4, raw server logs and dedicated AI-visibility platforms (a fast-moving category that includes tools like Profound, Peec and Otterly). The method matters more than the logo on the dashboard. The point is that being found and cited by AI becomes a measured, repeatable number, not a line in a pitch.
The three mistakes that hide the truth.
Mistake one: judging AI search by clicks. AI answers often resolve without a click, so click-only reporting makes a winning brand look invisible. Measure presence in the answer, not just traffic out of it.
Mistake two: testing prompts once. Answer engines change weekly. A single snapshot is noise. Share of AI voice only means something tracked as a trend on a fixed prompt set.
Mistake three: optimising without reading the logs. If the search crawlers never fetch a page, no amount of on-page work gets it cited. Crawler coverage comes first.
Built into the SEO & AI search engagement.
This measurement system is not an add-on. It is wired into our SEO, AI Search & Digital PR engagement from week one, so every optimisation has a before and after. We set the GA4 channel group, baseline your share of AI voice and crawler coverage, then report the movement every cycle against the only metric that pays the bills: qualified enquiries.
Want the two-minute version first? Run the AI Visibility Scorecard to see where you stand, or read the 2026 AI Search Report for the data behind the shift.
Measuring AI search, answered.
How do you measure AI search visibility?
You measure it across five linked metrics: AI referral traffic (visits that arrive from ChatGPT, Perplexity, Gemini and Copilot, isolated in GA4), share of AI voice (how often you appear in the answer for a fixed set of buyer prompts), citation tracking (which of your pages and mentions the engines actually quote), AI crawler coverage (which pages the AI bots read, taken from server logs), and answer accuracy (whether the AI describes your brand correctly). Tracked together, these turn 'being cited by AI' from a claim into a number you can move month over month.
Can Google Analytics 4 track traffic from ChatGPT and Perplexity?
Not by default. GA4 lumps most AI referrals into 'Referral' or 'Direct'. You fix this by building a custom channel group that matches the AI engine domains by regex (chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, copilot.microsoft.com). Once that is in place, every visit that started inside an AI answer is grouped, attributed and tied to enquiries and revenue like any other channel.
How do I know if ChatGPT or Google AI Overviews is citing my website?
Two ways. First, server log analysis shows whether the AI search crawlers (OAI-SearchBot, PerplexityBot, Claude-SearchBot, Google-Extended) are reading your pages at all, which is the precondition for being cited. Second, prompt testing and citation tracking show whether you actually appear and get named as a source in the answers buyers see. We run both, because a page can be crawled and still never cited.
What is share of AI voice?
Share of AI voice is the percentage of relevant buyer prompts where your brand appears in the AI-generated answer. Test 100 of the real questions your buyers ask across ChatGPT, Perplexity, Gemini and Google AI Overviews, count how many name you, and that is your share. Appearing in 28 of 100 is a 28% share of AI voice. It is the AI-era equivalent of keyword rankings, measured against named competitors.
What is a good AI search visibility score?
There is no universal benchmark yet because the answer engines change weekly, so the score that matters is your own trend against your named competitors. A useful target is to be cited in the majority of your highest-intent buyer prompts and to grow your share of AI voice every month while AI referral traffic and enquiries climb alongside it. Relative progress beats any vanity absolute number.
How is measuring AI search different from measuring SEO?
Classic SEO measures rankings, clicks and organic sessions. AI search often resolves without a click at all, so a page can rank and still lose the visit to an AI answer. AI search measurement therefore tracks presence inside the answer (share of voice and citations) and the smaller but high-intent referral traffic the answer sends, not just position on a results page.
Which AI crawlers should I allow in robots.txt?
If you want to appear in AI answers, allow the search crawlers that build the answer indexes: OAI-SearchBot (ChatGPT search), PerplexityBot and Claude-SearchBot. Whether you also allow the training crawlers (GPTBot, ClaudeBot, Google-Extended) is a separate, deliberate choice, because OpenAI and Anthropic let you control search visibility and model training independently. We set this up so you stay visible without giving away anything you did not intend to.
The references behind this method.
- G2, “New G2 Research: Half of B2B Software Buyers Now Start Their Research With AI Chatbots” (2025). prnewswire.com
- Search Engine Land, “8 GEO metrics to track in 2026”. searchengineland.com
- No Hacks, “The AI User-Agent Landscape in 2026: A Complete Reference” (GPTBot, OAI-SearchBot, ClaudeBot, Claude-SearchBot, PerplexityBot, Google-Extended). nohacks.co
- Databox, “How to Track AI Search Visibility With GA4 and Search Console”. databox.com
See whether AI is
recommending you.
We will baseline your AI search visibility across all five metrics and show you exactly where you are winning, where you are invisible, and the order to fix it in. Free, no obligation.