AI Beacon/Attribution
Someone has to ask whether this worked.
Measure what changed. Know what you can prove. AI influence does not always produce a click, which makes it hard to count rather than impossible. AI Beacon records the baseline before the work starts, then separates what you can observe from what you can only infer.
Referral sessions, citation counts, and the pages earning them.
Branded search and direct traffic moving after visibility work started.
A buyer reads the answer, forms a view and never visits. Claiming a number here would be invention.
What is AI citation attribution?
AI citation attribution connects changes in how AI platforms cite your brand to changes in the traffic and pipeline that follow. Some of it is directly measurable, because assistants pass a referrer when a reader clicks through. The larger part is not, because a recommendation that shapes a shortlist produces no session at all.
Key facts
- Referral traffic from assistants is visible in analytics and is the smaller half of the effect
- A recommendation that leads to a direct or branded visit later leaves no attributable trail
- Branded search volume is the most useful proxy for influence that produced no click
- Attribution here is correlation over time against a recorded baseline, not a deterministic path
- Measuring only sessions systematically understates the channel
- Any tool claiming exact revenue attribution for an AI recommendation is overstating what is knowable
Last reviewed 24 August 2026 by the SEORCE team.
The conversation that decided it left no record.
Research happens off your site
A buyer shortlists three vendors inside an assistant. Two are eliminated without ever being visited. Nothing about that appears in anyone's analytics.
The visit arrives later, and looks direct
They come back a week afterwards by typing your name. Your analytics records direct traffic. The reason they knew the name is invisible.
Absence has no signal at all
Being left out of an answer produces nothing to measure. No impression, no lost click, no entry anywhere. It is the most expensive outcome and the least visible.
What can be attributed, and what cannot.
This distinction matters more than the dashboard. A number presented with more certainty than it deserves is worse than no number, because it will be relied on.
Directly observable
Traffic where an assistant passed a referrer, and behaviour you can watch on your own site.
- Sessions referred from assistants that send a referrer
- Which pages receive them, and what those visitors do next
- Conversion rate of that traffic against other sources
- Movement in branded search volume after a visibility change
- Citation counts before and after a specific published change
Real, and not directly countable
Influence that shaped a decision without producing a session you can tie to it.
- Recommendations that led to a direct visit days later
- Shortlists you were included in, or removed from, silently
- Deals lost because a competitor was named and you were not
- The revenue value of a single citation, which nobody can isolate
The honest method is a recorded baseline, a deliberate change, and a comparison of both columns over time. That is weaker than deterministic attribution and considerably stronger than a screenshot.
The numbers you can defend are worth more than the ones you cannot.
The decision happened somewhere you cannot see
A buyer researched inside a conversation, built a shortlist, and arrived days later looking exactly like direct traffic.
A baseline recorded before the work starts
Citations, referral sessions and branded search are captured at the beginning, so later movement has something real to move against.
Every figure in the deck gets challenged
Attribution claims that cannot be evidenced damage the ones that can, and the whole programme loses credibility with it.
Countable and uncountable separated, openly
What can be observed is reported as observation. What cannot is named as influence and deliberately left without a number.
Absence leaves no trace at all
When an answer never names you, nothing arrives, nothing logs, and no report anywhere shows a loss.
Presence measured directly, not inferred from traffic
Citation share is recorded whether or not anybody clicks, so being absent is visible as a fact rather than as a missing line.
One source proves nothing. Four sources agreeing prove something.
Attribution is only credible when the citation change, the pages that earned it, the fixes that were shipped and your own search data all move in the same direction, on the same dates.
Citations across tracked questions went from 14 to 39.
Eleven URLs were unblocked in week two.
Four pages were rewritten against the questions asked.
Branded search and direct traffic rose from week four.
The citations rose on the exact questions whose pages were unblocked and rewritten, in the weeks that followed each change, and branded search moved after them rather than before. Any one of those is arguable. Together they are a timeline.
Enough to defend a budget line.
Referrals you can see
- Sessions arriving from assistants, separated from ordinary organic
- Landing pages receiving them, which is often not the page you expected
- How that traffic behaves compared with search visitors
Change against baseline
- Citation movement plotted against traffic and branded search over the same period
- Before and after comparison for a specific published change
- Enough signal to tell a real improvement from a seasonal one
Both channels together
- Search and AI presence in one view rather than two reports
- Pages earning citations and rankings, which are usually the same pages
- Where one channel is carrying the other