Why analytics won't tell you
Most AI influence never produces a click. A buyer asks an assistant which tools to consider, reads an answer that names four brands, and arrives at one of those sites later — via a branded search or direct visit. Your analytics records that as direct traffic, and the AI answer that actually decided the shortlist is invisible.
That is why AI visibility has to be measured by querying the engines, not by reading referral reports. The starting method is in how to audit your brand's AI search visibility.
The five things worth measuring
- Inclusion — is your brand named in the answer at all? This is binary and it is the first gate.
- Citation share — how often you are cited versus the competitors named alongside you. The clearest competitive metric in AI search.
- Source mix — which domains the answer pulls from. If Reddit and review sites dominate, your on-site work alone will never move it.
- Description & sentiment — how you are characterised. Being named inaccurately can be worse than being absent.
- Downstream behaviour — what AI-referred visitors do once they arrive.
Before you measure: make sure you're visible at all
There is a failure mode that measurement will otherwise mask. If you are missing from the index an engine retrieves from, you are not under-performing — you are absent, and no amount of content tuning will help. Check that first: why ChatGPT can't see your website.
The same applies at the entity level. If Google has not resolved who you are, AI has nothing stable to attach your expertise to — see Knowledge Panels & entity authority.
Building a repeatable baseline
AI answers vary between runs, so method consistency matters more than frequency. Fix a query set that reflects real buying questions, record the exact query, engine, date, answer, cited sources, and how your brand was described. Re-run monthly against the same set. Comparability is the whole point — a beautiful one-off snapshot tells you nothing about direction.
Record what you changed between measurements, too. Without that, you have data but no attribution.
Valuing what you find
Do not judge this channel on raw referral volume. AI-referred visitors convert far better than traditional organic, because much of the evaluation already happened inside the answer. The numbers and how to make that case internally are in measuring AI search ROI.
Turning measurement into work
An audit is only useful if it produces a queue. Gaps in the source mix point to off-domain work — usually Reddit and community authority. Gaps in how you are described point to cross-source consistency. Gaps in extraction point to citability. The sequenced version of all of it is the 90-day roadmap.
