On this page
What counts as an AI citation?
An AI citation is an instance where an assistant used one of your pages as a source when composing an answer. It is not a visit, not a click and not a lead. Conflating citations with traffic is the single most common measurement error in this field, and it produces reporting that falls apart under scrutiny.
A second distinction matters just as much. The query recorded against a citation is usually a grounding query — the assistant's own internally reformulated search — not the words the user typed. A person asking a long conversational question may generate several grounding queries behind the scenes. This is why citation counts can be large relative to any plausible number of human askers, and why they should never be presented as a headcount of people.
Layer 1: first-party citation data (free, and the layer most teams skip)
Bing Webmaster Tools' AI Performance report is the only first-party record of AI citations a site owner can get. It entered public preview in early 2026 and expanded mid-year to add intent classification, topic grouping, citation share against competitors, and period comparison. It reports:
- Citations — how many times your pages were used as a source
- Cited pages — how many distinct URLs were drawn on
- Grounding queries — the reformulated retrieval queries behind those citations
- Citation share — your proportion of citations for a given query or topic
Because Bing's index feeds Copilot and parts of other assistants' retrieval, this is also the best free proxy for visibility you cannot otherwise observe. Start here before paying for anything.
Layer 2: Search Console's generative-AI views
Google added dedicated performance views for AI Overviews and AI Mode in 2026, with a limitation worth stating plainly: they report impressions only. There is no click data, no click-through rate and no query-level breakdown in those views, and the rollout has been gradual rather than universal. They are useful as a directional signal that your pages are being surfaced in generative surfaces — not as a performance measure.
Layer 3: AI referral traffic in GA4
GA4 added a native AI Assistant channel in 2026, so sessions arriving from recognised assistant domains are classified automatically without configuration. Two caveats change how you should read it. The channel recognises a specific set of assistants and does not currently include every major one — Claude and Perplexity traffic lands in Referral unless you write a custom channel group. And the classification is not retroactive, so historical assistant traffic stays wherever it originally fell, usually Direct.
The blind spots nobody adjusts for
Some assistants send no referrer header at all. ChatGPT in particular frequently passes nothing, especially for free-tier users, so a substantial share of genuinely assistant-driven sessions is recorded as Direct traffic. This is structural — it is how the traffic arrives, not a tagging mistake you can configure away.
The practical consequence: treat every AI-referral number as a floor, not a count. If your first-party citations are climbing while GA4 AI referrals stay flat, the traffic is very likely arriving as Direct. Triangulating the two is more informative than trusting either alone.
One more gap worth naming: no major AI-visibility platform currently ships a purpose-built warehouse connector. Getting this data next to your SEO reporting means an API or a scheduled export, so budget a little engineering time rather than expecting a one-click integration.
Building a scoreboard you can trust
Four numbers, reviewed monthly, are enough to run a programme:
| Metric | Source | What it tells you |
|---|---|---|
| Total citations | Bing AI Performance | Direction of travel — noisy, so read the trend not the day |
| Distinct cited pages | Bing AI Performance | Whether visibility is broadening or resting on one page |
| Citation share by query | Bing AI Performance | Where you are winning and where demand is leaking |
| Concentration ratio | Calculated | Share of citations from your single biggest query cluster |
That last one is the metric almost nobody tracks and the one that predicts trouble. If a large majority of your citations come from one query cluster, a single change in how an assistant reformulates queries can remove most of your visibility overnight. Falling concentration is a healthy outcome even when total citations dip.
What to actually report to leadership
Report citations as visibility, not as demand. A defensible monthly summary reads: citations and their trend, how many distinct pages earned them, the queries where share moved, and what was published or changed in the period. Then state the caveat once — that assistant referral traffic is undercounted by design, so on-site numbers understate the real effect.
Resist converting citations into a revenue figure. The attribution chain does not currently support it, and a number you cannot defend costs more credibility than it buys.
Frequently Asked Questions
How do I track AI citations for free?
Bing Webmaster Tools' AI Performance report is free and is the only first-party source of AI citation data available to site owners. It reports citations, distinct cited pages, the grounding queries behind them and your citation share by query and topic. Pair it with Google Search Console's generative-AI views for directional impression data, and a custom GA4 channel group for assistant referrals. That combination covers the essentials at zero software cost.
What is a grounding query and why does it differ from what users type?
A grounding query is the assistant's own internally reformulated search, generated when it decomposes a user's question into narrower retrieval queries. One conversational question can produce several grounding queries. This is why citation counts often look large relative to plausible user numbers, and why grounding queries should be read as retrieval behaviour rather than as a record of what people asked.
Why does my AI traffic not show up in Google Analytics?
Mostly because some assistants pass no referrer header. ChatGPT frequently sends none, particularly for free-tier users, so those sessions are recorded as Direct rather than as AI referrals. GA4's native AI Assistant channel also does not cover every assistant — Claude and Perplexity land in Referral without a custom channel group — and the classification is not retroactive. Treat AI referral figures as a floor.
Do I need a paid AI visibility tool?
Not to begin with. Free first-party data covers measurement for most single-market programmes. Paid platforms earn their place when you need daily monitoring across engines Bing cannot see, competitor share-of-voice tracking, or multi-market scale. Build the content base first and add software when a specific question arises that free data cannot answer.
What is the most important AI visibility metric?
Concentration — the share of your citations coming from your single largest query cluster. Total citations tell you the direction of travel, but concentration tells you how fragile that number is. A site earning most of its citations from one query cluster can lose the majority of its visibility from one change in assistant retrieval behaviour, so a falling concentration ratio is a healthy result even if the headline total dips.
Related reading
- The 10 best AEO tools in 2026, compared
- AEO reporting: the metrics that matter
- The AEO experimentation roadmap
- AEO Agency Singapore
- AI search optimisation services
Want to be the answer, not just a search result?
AI Studio builds AI search visibility for Singapore brands — entity, schema, and the content that assistants actually cite. Start with a free AI Visibility Audit of your own site.