The problem
The referral report had a row that shouldn’t have been there. Traffic from AI chat assistants had climbed far enough that, for a stretch, it was the third-largest source on the site - and its revenue-per-session was the highest of any channel we ran, paid or organic.
The reasonable reading was: measurement artifact, ignore it. The uncomfortable reading was: customers are researching gear inside chat assistants, arriving already convinced, and we have no idea whether those assistants can read our catalog correctly.
What I did
- 01Built the internal case first.Segmented AI-chat sessions in GA4, held them up against every other channel on revenue-per-session, and argued this was a channel to staff rather than a spike to explain away.
- 02Rewrote PDPs for machine comprehension.Clarified what each product is actually for, then expanded specifications and FAQ coverage until an assistant could answer a shopper’s question without guessing.
- 03Fixed the structured data.Product, offer, and FAQ schema across the catalog so entity relationships were explicit instead of inferred. Unglamorous, and the highest-leverage work of the year.
- 04Kept running the rest of the business.Alongside core KPI ownership: a $65K monthly digital budget, ROAS improvement to 3.1, and conversion, AOV, and retention work across the funnel.
What happened
AI chat is now a consistent top-5 traffic source that outperforms social on revenue, at roughly double the revenue-per-session of the next-highest channel. Annual online revenue grew 17% over the same period.
The part I care about more: the reporting now treats AI discovery as a first-class channel, so the next person in this seat inherits a measured asset instead of an anomaly.
What I’d tell you in the interview
Nobody handed me this project. It came from reading the reports closely enough to be bothered by a number that didn’t fit, then doing the political work of getting a weird idea funded.