Attribution that survives the AI-direct era

Last-click attribution has always been a simplification, but it used to be a tolerable one. A buyer clicked an ad, landed on your site, and booked — the ad got the credit, and that was roughly true often enough to act on. AI-direct traffic breaks that simplification badly. A buyer now asks an AI tool a question, gets a confident answer naming your business, and arrives at your site or your booking flow already decided. If your last-click tool only sees "direct traffic" or "referral: unknown," it has no idea that an AI recommendation is what actually won the sale.

This isn't a hypothetical edge case anymore — it's a growing share of how buyers reach a decision before they ever touch a search engine results page. And it creates a dangerous blind spot: channels that are actually working (being cited well by AI tools) look, in your dashboard, identical to random unattributed traffic. Owners cut budget from what looks unproductive and pour more into what's easy to measure, which is exactly backwards when the easy-to-measure channels aren't the ones doing the work anymore.

Fixing this doesn't mean throwing out measurement — it means adding the layers that catch what last-click misses. Server-side conversion tracking, not just pixel-based, so ad blockers and privacy settings don't create silent gaps. UTM discipline on every surface you control, so at minimum you're not throwing away your own signal. And directly asking new customers a plain question — "how did you hear about us?" — at the point of conversion, because self-reported data is noisy but it's the one source that can actually detect an AI recommendation that no analytics tool saw.

The goal isn't a perfect model. It's avoiding the specific failure mode where you defund the channel that's quietly winning you customers because your tooling can't see it. Triangulating across server-side tracking, UTM-tagged surfaces you control, and self-reported source at conversion gets you close enough to make real budget decisions instead of superstitious ones.

This matters most for businesses just starting to invest in AI search visibility, because the payoff of that work is specifically the kind of traffic last-click attribution is worst at seeing. If you can't measure it, you'll conclude it isn't working — right as it starts to.

Treat attribution as an ongoing discipline, not a one-time dashboard setup. The channels buyers use to decide keep shifting, and your measurement has to keep pace or it starts lying to you quietly, which is worse than not measuring at all.