Through 2025 our organic numbers started behaving strangely. Not dramatically, and that was the problem. A dramatic drop gets a meeting booked the same day. This was quieter: sessions drifting in ways seasonality did not explain, engagement patterns that did not match the campaigns running, pages earning conversions from visits that looked nothing like the visits we knew. Every month the movement was small enough to explain away, and every month the explanation felt a little thinner.
I remember the moment it stopped being deniable. I was segmenting landing page traffic for a routine monthly readout and kept finding a cohort that behaved like no channel we tracked. They arrived deep in the site, skipped the pages our journeys were built around, and either converted quickly or left immediately. Almost no middle. It looked less like browsing and more like someone arriving with the research already done.
Because it was. People had moved their questions to AI tools. The comparison, the shortlist, the "which one should I buy", all of that was happening somewhere we could not see, and only the end of the journey was reaching us. Our measurement framework, built carefully over years, had no name for this traffic. So it was being quietly filed under everything else, and slowly bending every baseline we trusted.
The tempting response was to wait. The industry did not agree on how to measure AI search, the tracking signals were messy, and any index we built would be imperfect. The argument I ended up making internally was that imperfect and consistent beats perfect and absent, because the drift was happening whether we measured it or not.
So we built an AI and Search impact index on top of the purchase intent framework we already used across our markets. Technically it was unglamorous: BigQuery, Python, classification rules, and a scoring layer. The genuinely hard part was definitional. What counts as AI-influenced? Which signals are strong enough to classify on? What do we do with the ambiguous middle? Those arguments took longer than the code, and they were worth every hour, because an index only works if you can hold its definition steady month after month while people challenge it.
Once it existed, the conversation changed shape. "Organic is down and we are not sure why" became "this much of the movement is the landscape shifting, this much is ours to act on". Content teams could see which pages were winning in AI-driven journeys and which were becoming invisible. And leadership could see the trend line early, while it was still a strategy question rather than an emergency.
The lesson I keep from it: your baselines have a shelf life. Every measurement framework quietly assumes the world that existed when it was designed, and the world owes it nothing. If your benchmarks were built before AI search, they are already drifting. You do not need to predict where this ends. You need to start measuring while it happens, because the alternative is standing in a meeting two years from now, explaining a trend with data you never collected.