Attribution is a shrinking measure for AI search, argues Kevin Indig
Attribution was built to divide advertising budgets using cookies and clicks, and it was never a good fit for non-paid marketing, writes Kevin Indig in a piece co-authored with George Bonaci, VP Growth at Ramp. As AI Overviews and ChatGPT answer questions without sending a visit, companies lose the referral, click and conversion data that once showed how customers found them.
The result is a measurement gap that keeps widening. Consent restrictions, Apple's 2021 cross-app tracking opt-out and multi-device journeys have pushed platforms towards modelled conversions, so attribution now explains a shrinking share of demand. Graphite data cited in the piece puts AI underattribution at up to 10x.
Bonaci calls attribution a crutch that replaces critical thinking, and still runs a multi-touch model with caveats, discussing its limits constantly and measuring channel impact separately. Indig proposes two alternatives for non-paid work: triangulation, combining an exposure metric such as AI Share of Voice with self-reported discovery or CRM tags and conversions, and incrementality testing through randomised holdouts or geo experiments.
Referral traffic remains one of the few AI search signals brands can observe directly, even though it cannot show what happened before the click. A BCG survey of 3,000 senior measurement professionals found 46% already combine MMM, incrementality testing and multi-touch attribution, and those integrated users report up to 70% stronger revenue growth.
Reported by Growth Memo ↗
Topics: Measurement, AI Overviews, ChatGPT