Measuring AI Visibility

You cannot manage AI visibility from your normal analytics, because most AI answers resolve without a click. Measuring it means sampling the answers themselves, on a fixed cadence, and reading a share of voice the way you once read rank. This chapter is the method and its honest limits.

What is share of voice in AI answers?

Share of voice is the percentage of a defined prompt set where an engine mentions, cites, or recommends your brand, measured against your competitors. It is the AEO analog of rank tracking. Instead of one keyword and one position, you sample many prompts and measure two things: whether you show up at all, and how prominently, first among options or buried in a list.

The shift matters because there is no single position to watch. An answer engine names a handful of sources per response, and the set changes from prompt to prompt and run to run. So you stop asking "where do I rank for this term" and start asking "across the questions my buyers actually ask, how often does the answer name me, and how often does it name a competitor instead."

How do you actually measure it?

Build a prompt set and run it on a schedule. The steps are concrete.

• 1. Build 50 to 200 buyer-intent prompts. Cover the real questions: problems, comparisons, "best X for Y," and "alternatives to Z."
• 2. Run each prompt across your target engines on a fixed cadence, so runs are comparable over time.
• 3. Record four things per response: whether your brand is mentioned, whether it is cited with a link, whether it is ranked first among named options, and which URL got cited.
• 4. Compute share of voice per brand across the set, and track deltas and which of your pages earn the citations.

This is a measurement cadence you run alongside your other tracking, not a one-time audit. Prompt tracking has its own tooling and rhythm worth setting up deliberately. Prompt tracking →

Two numbers fall out of this that are worth naming. Citation rate is the share of your prompt set where an engine cites you, computed as prompts where cited divided by total prompts. Share of voice weighs your citations against the whole competitive set, so it answers "of all the naming that happened, how much was mine." Track both over time rather than in isolation, because a rising citation rate while your share of voice falls means competitors are gaining faster than you are. A market of dedicated tools has grown up around this, but verify each one's engine coverage before you quote its numbers, because they sample differently and do not agree.

Can your analytics see any of this?

A little, as a supporting signal rather than the main measure. In GA4 you can segment referral traffic by host, so visits from chatgpt.com, perplexity.ai, and similar domains become visible, and you can group them into an "AI assistants" channel to watch the trend. GA4 for SEO → Cross-reference that against your server logs, where AI search-bot fetches act as a leading indicator: the bots read you before any referral click shows up, if one ever does.

Set expectations low on the absolute numbers. Most AI reads are zero-click, so referral traffic from answer engines is real but small, and the brand mention is the actual prize, not the click. Search Console is still your source of truth for the classic search click that AI answers sit next to. Search Console → Treat AI referral counts as a pulse, not a revenue line.

A single run of an AI prompt is a weather reading, not a rank. The answer shifts with the model, the user, the location, and the hour. Sample repeatedly and report ranges, or you will chase noise.

What can these numbers not tell you?

Handle every figure here with care, because the ground is soft. AI answers are non-deterministic: the same prompt returns different sources on different runs. They are personalized, and they are sensitive to geography and time. So one reading proves nothing. Sample repeatedly, hold the prompt set and cadence steady, and report ranges rather than single points.

Two more honest caveats. Google's new generative-AI reports in Search Console show impressions only, and they are a subset of the picture, so do not present them as new traffic you gained. Citation-rate baselines, the practitioner rule of thumb that competitive visibility runs somewhere around 15 to 25 percent, are early and unstandardized, so quote them as directional, not as a benchmark to hit. The same caution applies to any headline figure about how often AI answers appear or how much they cut clicks: those numbers move month to month and studies disagree, so report them as ranges and refresh them before you lean on them. Measured with that discipline, share of voice tells you whether the on-page, entity, and crawler work is landing. Pull it all together and you have a repeatable playbook. The AEO playbook →