AI visibility should be measured in two layers: what answer engines say before the click, and what users do after the click. Prompt coverage, brand mentions, citations and sentiment show visibility; GA4's AI Assistant channel, lead events and CRM progression show whether that visibility creates commercial value.
A traffic report is not an AI visibility report
The easiest measurement mistake is to count visits from ChatGPT and call that GEO performance. Referral traffic only records the journeys that produced a click. A brand can be recommended repeatedly without receiving a visit, or can receive visits from an informational citation that never creates buying intent.
That is why the measurement model needs a pre-click layer and a post-click layer. The first tells you whether you are present in the answer. The second tells you whether the presence matters.
The pre-click layer: measure the answer itself
Build a prompt set that represents the questions your market asks at awareness, comparison and decision stages. Hold the wording stable enough to compare results over time. For each prompt, capture whether the brand is named, the position or prominence of the mention, the description used, the sentiment, the cited source and the competitors included.
Track this separately by model. ChatGPT, Gemini and Claude do not produce identical answers or use identical retrieval systems. A blended "AI visibility score" is useful for a board view, but it should always be possible to drill down to the model and prompt that moved.
The post-click layer: GA4 now makes this easier
Google Analytics added dedicated AI Assistant traffic measurement in May 2026. Recognised visits from assistants such as ChatGPT, Gemini and Claude can be categorised in the AI Assistant channel, with an "ai-assistant" medium. This is a useful operational improvement because teams no longer need to bury all AI referrals inside generic Referral or Unassigned buckets.
Use the channel as a starting point, not the end of attribution. Compare engaged sessions, key events, form submissions, assisted conversions and account quality against Organic Search, Paid Search and other channels. A small AI channel with a high qualification rate may deserve more attention than a larger source with weak progression.
Connect analytics to CRM evidence
The revenue question is not whether an AI visitor filled a form. It is whether those accounts became qualified opportunities and whether AI appeared anywhere in the journey. Capture self-reported attribution where practical, preserve source data through the CRM, and review AI-touched accounts separately from simple last-click referrals.
For complex B2B sales, the buyer may discover a firm in an assistant, return through Direct two days later, read a case study through Google, and then book a meeting from a branded search. Last click will miss the discovery event completely.
The scorecard to review monthly
At minimum, report prompt coverage, recommendation share versus named competitors, cited pages, sentiment, AI Assistant sessions, engaged-session rate, key-event rate, qualified leads, pipeline created and AI-assisted opportunities. Add a notes field for meaningful answer changes, because a stable score can hide an important shift in how the brand is being described.
The point of measurement is to decide what to do next. If visibility is low, inspect access and content. If mentions are strong but descriptions are wrong, fix entity clarity and corroboration. If traffic is healthy but pipeline is weak, the problem is conversion or prompt selection rather than visibility.
Frequently asked questions
Does GA4 track ChatGPT traffic?
Yes. Google Analytics introduced an AI Assistant channel in May 2026 for recognised traffic from popular AI assistants, including ChatGPT, Gemini and Claude.
Is AI referral traffic the same as AI visibility?
No. Referral traffic measures clicks. AI visibility also includes non-clicked mentions, recommendations, citations and how the brand is described inside generated answers.
What is a good AI visibility score?
There is no universal benchmark because prompt sets, markets and models differ. The useful comparison is your own baseline, named competitors and coverage of the prompts that have real buying intent.
How often should AI visibility be measured?
Monthly is a practical minimum for most B2B firms. Faster-moving categories may benefit from weekly checks on a smaller set of priority prompts.