Case study / measuring_the_invisible.txt
Attribution when nobody clicks
Problem. AI answers create influence without clicks. The classic analytics stack goes blind exactly where the market is moving.
System. I measure in layers. (1) A 600-prompt share-of-voice system tracks brand mention rate across ChatGPT, Claude, Gemini, Perplexity, and AI Overviews daily — 53% → 87% in eight months. (2) I separated “AI Search” into its own post-purchase survey source, turning anecdote into a trend line tied to signups. (3) A GSC misspelling-divergence method: people who discover you through AI type the brand the way the AI spelled it — the gap between clean and divergent branded queries becomes a proxy for AI-driven discovery.
Why it matters. When the channel hides the click, you measure the echoes. This is the methodology briefed at board level.
Historical mention-rate result: (commercial tooling — since audited and replaced by the in-house platform above). See “The vendor data was synthetic, so I built the real thing” for the replacement platform.
When the click disappears,
the evidence doesn’t.
Complementary measurements, not a claim that every AI-influenced conversion can be individually identified.