A mention is your brand appearing in AI-generated text without a link. A citation is a clickable source link to your site inside or beside that answer. A click is a visit recorded in your own analytics. They measure different things: mentions show presence, citations show source selection, clicks show realised traffic. Track all three separately, because one can move without the others.
Why do mentions, citations and clicks get confused?
Most confusion comes from treating an AI answer as if it were a ranked result. In classic search, a ranking position and a click are closely coupled: you see the link, you click it. AI answers break that coupling. Google's own guidance on AI features describes AI Overviews and AI Mode as experiences that surface relevant links alongside generated responses, and notes that both may use a "query fan-out" technique, issuing multiple related searches across subtopics and data sources to build an answer (AI features and your website, checked 30 September 2026).
That fan-out matters for measurement. A single user question can trigger several underlying retrievals. Your brand might be mentioned in the synthesised text, cited as one of several supporting links, or both, or neither. Each outcome is a different signal.
Definitions worth fixing before you measure
- Mention: your brand name or domain appears in the generated answer text, with or without a link.
- Citation: a clickable link to your page appears as a source in or attached to the answer.
- Click: a user actually visits your site, visible in your own analytics or server logs.
A mention without a citation is visibility you cannot directly attribute. A citation without a click is a source selection that did not convert into a visit. A click without a visible citation is possible through other paths, such as branded search after exposure.
What does the official evidence actually support?
Google's documentation states that the best practices for SEO remain relevant for AI features in Google Search, and that there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimisations necessary (AI features and your website, checked 30 September 2026).
That is a narrow but important claim. It supports continuing to invest in fundamentals: crawlable pages, clear structure, useful content. It does not support claims that a specific formatting trick, schema pattern or "GEO factor" guarantees inclusion. The same source describes AI Overviews as designed to appear on queries where they add benefits beyond existing Search results, and says people have been visiting a greater diversity of websites for more complex questions. That is a directional statement about behaviour, not a measurement recipe.
So the honest position is: the mechanism is documented at a high level, the measurement is not standardised, and anyone selling a precise AI visibility score is filling a gap with their own assumptions.
How should you sample AI answers without fooling yourself?
Because AI answers vary by query, location, device, account state and time, a single screenshot proves very little. Build a small, dated protocol instead.
Step 1: Fix a query set
Choose 10–30 queries that matter commercially and are phrased the way people actually ask. Include a mix:
- branded queries (your name plus a category)
- non-branded category queries
- comparison queries ("X vs Y")
- problem queries where an AI answer is likely
Record the exact wording. Paraphrasing later invalidates comparison.
Step 2: Fix market and tool
Record country, language, device type, whether you were signed in, and which surface you used (for example AI Overviews in standard results versus AI Mode). These are not cosmetic details; they change what you see.
Step 3: Log three columns, not one
For each query and each run, record:
| Field | What to record |
|---|---|
| Mention | Yes/no, plus whether your brand name appeared in the answer text |
| Citation | Yes/no, plus the exact URL cited if it is yours |
| Click | Whether your analytics later shows a visit you can plausibly tie to that query or page |
Clicks are the hardest column. AI referrals are not always cleanly labelled, and some arrive as direct or branded search. Treat click attribution as directional, not exact.
Step 4: Repeat on a schedule
Run the same set weekly or fortnightly, at roughly the same time of day, and keep the raw records. The value is in the trend across runs, not in any single snapshot.
Example: what a two-week log might look like
Hypothetical example, illustrative figures only.
Suppose you track 20 queries twice, one week apart, in the UK on desktop while signed out.
- Week 1: 6 queries show a mention, 3 show a citation to your site, analytics shows 11 visits from AI-adjacent referrers.
- Week 2: 8 queries show a mention, 3 show a citation, analytics shows 9 visits.
Mentions rose, citations held flat, clicks fell slightly. A single-number "AI visibility score" would hide that. The three-column view tells you presence improved while source selection and realised traffic did not, which points you toward checking whether your cited pages are the ones users actually want next.
How do you check tool variability rather than trusting one dashboard?
Third-party AI visibility tools each define mentions and citations differently, and their crawl or prompt methods are usually undisclosed. Rather than assuming one is correct, run a small cross-check:
- Pick five queries from your set.
- Run them manually in the surface you care about, signed out, in your target market.
- Compare what you observed with what the tool reports for the same day.
- Note where they disagree and why it might be (different location, different surface, different definition of "citation").
If a tool cannot explain its method, treat its numbers as a prompt for investigation, not as a KPI. This is also where adjacent reporting changes matter: reports should be assessed on what they actually capture, as covered in our analysis of Search Console multimodal reporting. The same discipline applies: know what a report does and does not capture.
What should you do with the three signals together?
Use them as a diagnostic, not a scoreboard.
- Mentions up, citations flat: your name appears more often in the sample, without more linked citations. This does not establish that your pages were used as grounding. Check whether the cited competitors answer the specific sub-question more directly.
- Citations up, clicks flat: you are being offered as a source but not chosen. Check titles, snippets and whether the landing page matches the query intent.
- Clicks up, mentions flat: traffic may be coming from branded search or other channels. Do not attribute it to AI visibility without evidence.
- All three flat: the query set may be wrong, or the topic may simply not trigger AI answers.
For teams also watching paid surfaces, the same separation logic applies to campaign reporting: announced reporting features should be assessed on what they actually expose.
What remains uncertain?
Several things are genuinely unresolved as of late September 2026:
- There is no official, standard definition of an "AI citation" across surfaces and tools.
- Referral attribution from AI answers is inconsistent, so click data undercounts or mislabels in ways you cannot fully correct.
- Answer variability means small samples are noisy; a two-point change in a 20-query set is not a trend.
- Google's documentation describes mechanisms and confirms that standard SEO practices remain relevant, but it does not publish per-query inclusion rules.
Given that, the defensible approach is a dated, documented sample with three separate columns, plus a periodic cross-check against manual observation. That will not give you a guaranteed visibility metric, but it will stop you from reporting a mention as if it were a click.
Follow-up questions
Can I use one number to report AI visibility to stakeholders?
You can, but you should label it clearly as a composite you defined. A single index hides whether presence, source selection or traffic changed. If a stakeholder needs one figure, pair it with the three underlying columns so the movement is interpretable.
Does appearing in AI Overviews require special optimisation?
According to Google's AI features documentation, there are no additional requirements to appear in AI Overviews or AI Mode and no special optimisations necessary, and standard SEO best practices remain relevant. Treat any claim of a secret AI-specific ranking factor as unverified.