KEY TAKEAWAYS
-
AI Search Traffic is Undercounted:
Much of AI-driven web traffic gets categorized as standard Google Search or "unassigned" due to stripped referrers (e.g., paid ChatGPT, mobile app web views, and AI Overviews). -
Long-Term ROI vs. Short-Term Attribution:
AI authority content functions like brand building; forcing strict quarterly attribution models onto it creates an inaccurate representation of performance. -
Higher Conversion Quality:
Users originating from AI-guided paths demonstrate stronger conversion rates, even if the multi-touch attribution path cannot be fully traced. -
Transparency with Leadership:
Be honest with boards about measurement gaps. Communicating current tracking limitations clearly builds far more credibility than forced, precision-based guesswork.
There’s a statistic doing the rounds that AI sends about 1% of traffic to websites while accounting for something closer to 17% of search sessions. John Wilkes put it to Stephan Marais, head of SEO at Somebody Digital, to check whether the shape of it holds up.
Marais thinks it roughly does, with a significant caveat about what the 1% is actually measuring. “There are huge amounts of gaps in that visibility,” he says.
Why AI search data goes missing
Google’s own AI clicks don’t get counted as AI. A click from an AI overview lands in your reporting as an ordinary Google search. The AI visibility report in Search Console gives a little more, though not much that’s usable. “All it shows you is impressions,” Marais says. “No clicks, no position, no click-through rate, and that’s by design.”
Elsewhere the trail disappears completely. Mobile app web views strip the referrer. Paid ChatGPT passes no referrer at all. “All of that data is being bucketed into unassigned,” he says, which means a growing share of genuinely useful traffic arrives labelled as nothing in particular.
There’s a tempting hypothesis that the research happens in AI and the visit arrives later as a brand or direct search. Marais won’t sign it off. “Based on the data that we are seeing, that is not something that we can corroborate at all.” What he will say is that conversion rate looks stronger for users who come through the AI route.
The ROI and attribution problem this creates
Wilkes named the awkward implication for anyone asking a board for budget: If the content is doing its work inside AI answers people never click, does every penny still need to be attributable?
Marais’s answer is to be honest about the timescale rather than invent a number. Brand building has always had this shape. “To put an ROI next to brand building activity for the period that that activity runs is near impossible,” he says. “The results come later, after the fact.” Content built to establish authority sits in the same category, and forcing a quarterly attribution model onto it produces a false picture in both directions.
Briefing your board when attribution doesn’t exist yet
The position, as both of them describe it, is that measurement hasn’t caught up with behaviour.
“We’re almost building the plane as we’re going down the runway,” Wilkes says. “It’s changing so much, the measurement hasn’t caught up. So it’s communicating as well as we can given the state at the moment, but also just being comfortable saying that we don’t know all the answers right now, but we’re still doing all the work.”
Marais agrees, from the client side of the same conversation. In his experience clients would far rather hear “I don’t know yet” than a tidy story told to cover a gap.
For leaders briefing upward, that’s the level to pitch at: clear on the mechanics, honest about the measurement gaps, and specific about the work happening regardless. Boards can work with that. Confident precision that falls apart under a follow-up question loses the room much faster than an admitted gap.
Much of AI-driven traffic gets categorized as standard Google Search or “unassigned” because Google AI Overviews show impressions without detailed click or position metrics, while platforms like paid ChatGPT and mobile app web views strip referrer data entirely.
AI authority content functions like brand building; its value accumulates over time. Forcing a short-term quarterly attribution model onto it creates a false picture, so performance should be evaluated on longer-term strategic impact rather than immediate per-click tracking.
Leaders should be completely honest with boards about measurement gaps, explaining the underlying mechanics and ongoing strategy clearly. Openly acknowledging tracking limitations builds significantly more credibility than relying on forced, precision-based guesswork.
Yes. Users originating from AI-guided paths demonstrate stronger conversion rates, even if the multi-touch attribution path cannot be fully traced.
