AI Beacon/Citation tracking
Ten blue links became three names.
A search result hands a buyer a page of options. An answer hands them two or three names and stops. AI Beacon records which of those names is yours, on every engine, and separates the sources shown to the reader from the ones that shaped the answer without ever getting credit.
ChatGPT4 / 10
Perplexity9 / 10
Gemini2 / 10
Claude7 / 10
AI Overviews0 / 10
Copilot3 / 10
Grok3 / 10What is AI citation tracking?
AI citation tracking is the practice of repeatedly asking AI platforms the questions your buyers ask, and recording whether your brand is named, how it is named, and which sources the answer credited. It replaces rank tracking for a channel where there is no ranking, only presence or absence in a short list.
Key facts
- Answers name a handful of brands, so presence is closer to binary than to a position
- Responses vary between runs, so one answer is a data point and a pattern across runs is a finding
- A mention, a citation and a recommendation are different outcomes with different value
- The same question produces different answers on different platforms, so coverage must be measured per platform
- Being cited sends measurable referral traffic. Being mentioned without a link often does not
- Tracking tells you where you stand. Working out why takes prompt diagnosis
Last reviewed 24 August 2026 by the SEORCE team.
Asking once tells you almost nothing.
Every team tries this first: open ChatGPT, ask the question, see whether you appear. It feels like data and it is not.
One answer is not the answer
Ask the same question twice and the brands named can differ. Without repetition you cannot tell a real position from a coincidence.
Your session is not neutral
Your own history, location and account shape what you are shown. The answer a prospect gets is not the answer you get.
It does not scale or repeat
Seven platforms across a few hundred questions is thousands of checks. Done by hand it is never finished, never consistent, and never comparable week to week.
Mentioned, cited, recommended.
Most tools collapse these into one number. They are worth very different amounts, and improving one does not necessarily improve another.
Your name appears somewhere in the answer. It might be in a list of six, in a passing comparison, or as the option ruled out. It confirms the model knows you exist, which is not nothing, and is not yet worth much.
One of your pages is credited as a source the reader can click. This is the outcome that sends traffic you can see in analytics, and the one worth optimising a page for.
The answer puts you forward as the choice for the question asked. This is what actually moves pipeline, and it depends less on any single page than on whether the whole picture of your brand supports the recommendation.
Tracking all three separately matters because the fixes differ. A page earns a citation. A coherent, well-evidenced brand earns a recommendation.
Why a number you cannot trust is worse than no number.
Somebody checked ChatGPT once and it looked fine
One reply, one day, one phrasing, one session carrying its own history. That is an anecdote, and it will not survive a second look.
The same questions, the same way, every time
Runs go out across seven platforms on a fixed schedule with identical phrasing, so a movement is a movement rather than the model having a different afternoon.
Mentioned, cited and recommended get counted as one thing
They are worth wildly different amounts and they are earned in different ways. Averaging them hides which one you are actually winning.
Three outcomes, tracked separately
Each is recorded on its own, per question and per platform, because a single page earns a citation while a coherent brand earns a recommendation.
You cannot tell a real change from noise
Answers drift between sessions. Without a stable baseline every wobble looks like a trend, and every trend looks like a wobble.
Enough repetition to make a change mean something
A fixed schedule across enough runs turns session variance into a signal you can read with confidence.
A citation is the end of a chain. We show you the whole chain.
Whether a page gets cited depends on whether an engine can read it, whether it answers the question, and whether anything else already answers it better. Those are three different systems, and they only make sense side by side.
A review site is cited on nine of ten runs. You are cited on two.
Your comparison page is behind a rule that blocks two crawlers.
The page that is readable buries the answer below the fold.
Both pages rank on page one for the same term.
Your strongest page is invisible to two engines, and the readable one puts the answer where a model stops reading. Ranking well and being citable turn out to be two different tests, and you were only ever passing one of them.
Consistent questions, run the same way every time.
Which questions, which platforms
- Your question set run across all seven platforms rather than sampled
- Per-platform results, because they disagree with each other more than people expect
- Position within the answer, since first named and last named are not equivalent
- Which of your pages was credited, where a source was given
How reliably you appear
- Frequency across repeated runs, not a single snapshot
- A brand named in most runs holds a stronger position than one named occasionally
- Variance is itself a signal that a position is contested
Movement worth acting on
- Trends over weeks, which is the timescale answers actually shift on
- Alerts when you drop out of a question you previously held
- Comparison against a recorded baseline, so a fix can be proven
There is no position eleven to climb from.
Rank tracking assumes a list you can move up. An answer has no page two. You are in it or you are not, and the gap between those two states is the entire distance.
Small changes, large consequences
Moving from tenth to eighth in search is marginal. Moving from absent to named in an answer is the whole difference.
No stable ordering to defend
Positions in an answer are not stable enough to defend one at a time. Consistency across runs is the thing to protect.
Traffic is not the only outcome
A recommendation with no link still shapes a decision. Measuring only clicks misses most of what is happening.
Tracking finds it. Something else fixes it.
About tracking citations.
Find out where you actually stand.
Connect your brand and see which questions name you, which name a competitor, and which name nobody at all.