Attribution / 10 September 2026 / 4 min read

You cannot prove AI visibility drove revenue, and you should stop trying

The honest version of attribution is weaker than what vendors promise and considerably more useful.

Citarra field note03Good measurement begins by stating what the data cannot prove.

Every practitioner working on AI visibility eventually gets the same question from a CFO or client: what did this produce? It is a fair question and the usual answers are dishonest.

Why the attribution genuinely breaks

Someone asks an assistant which tool to buy. It names three, including you. They do not click a link because there often is not one. Days later they type your name into a browser or go straight to your site.

Your analytics records direct traffic or branded search. The AI answer that created the demand is invisible, and no amount of modelling recovers a touchpoint that was never instrumented.

This is not a tooling gap that better software fixes. The event happened somewhere you have no access to.The missing touchpoint

What vendors claim anyway

Some tools show revenue attributed to AI visibility. Ask how it is calculated and the answer is usually a correlation between a visibility score and a revenue line during a period when many other things changed.

That is not attribution. It is two charts on the same axis.

The honest claim, which is still worth having

You can establish, with sample sizes stated:

  • Whether visibility moved, on which engines, and in which specific questions.
  • What you changed and when.
  • Whether connected questions and sources moved differently from those that were not connected.

That is co-occurrence, not causation, and it should be labelled as such. It is also enough to decide what to do next.

A better question than “what did it produce?”

Ask instead: are we named in the answers our buyers get, more often than last month, in the questions that matter most?

That is measurable, honestly. Revenue attribution for this channel is not, yet. Pretending otherwise costs credibility the first time someone checks the working.

What to carry forward.

The original AI touchpoint often never reaches analytics.
Correlation is not channel attribution.
Measured change can still guide the next action without proving causation.
Next field noteReturn to all Citarra field notes