Chapter 2 — Where the Traffic Went

Before anything else, you need a clear-eyed account of what is actually happening to your numbers, because the wrong diagnosis leads to expensive wrong medicine.

Three things are happening at once, and they are frequently blamed on each other.

Thing one: Google is answering more questions itself

AI Overviews now appear on somewhere between one in five and two in five US searches, depending on who is measuring and what mix of queries they sampled. The spread is wide because query type drives everything.

A rigorous audit out of Washington University in St. Louis captures this well. Across 55,393 trending queries over forty days, AI Overviews appeared on 13.7 percent overall. But on queries phrased as questions, they appeared 64.7 percent of the time — nearly seven times as often. By category the range ran from 3.5 percent in beauty and fashion to 46 percent in hobbies and leisure, with politics and legal queries suppressed down near 8 percent.

So “how much of my category is affected” is a question with a real answer, and it is probably not the number in the headline you read. Run your own top hundred queries and count.

Thing two: your analytics is undercounting AI traffic

The AI referral traffic you can see is a floor, not a total.

Google Analytics added an AI Assistant channel group in 2026, which sets the medium to ai-assistant when it recognizes the referrer. Three problems: traffic from in-app browsers and mobile apps often arrives with no referrer at all and lands in Direct; Google has not published the list of assistants it recognizes; and it only captures referrals, not the far larger effect of answers where nobody clicks.

One nine-month first-party study tracking 51,200 AI Overview click events found 22 percent of that traffic misattributed to Direct rather than organic search. Whatever your dashboard says, the real number is higher.

Thing three: the assistants themselves are changing fast

The competitive picture is unstable in a way that should make you wary of platform-specific tactics.

Between mid-2025 and mid-2026, ChatGPT’s share of generative AI web traffic fell from roughly 76 percent to about 53 percent. Gemini rose from under 9 percent to around 27. Claude went from about 2 percent to 9. Perplexity and Copilot sit in low single digits.

A twenty-point swing in twelve months is not a market you tune a strategy to. It is a market you build platform-independent assets for.

The question that actually gets asked

Here is a finding that should change how content teams work, and it is not widely known.

Semrush, analyzing over a billion lines of clickstream data from a 200-million-user panel, found that between 65 and 85 percent of ChatGPT prompts do not match any keyword in a database of 27 billion keywords.

Read that again if your content plan starts with a keyword tool.

The majority of demand you are trying to serve is invisible to your primary instrument, because people phrase questions to an assistant the way they’d phrase them to a knowledgeable friend. Queries in Google’s AI Mode run about three times longer than classic search queries. People type “best CRM software.” They ask “what CRM should a twelve-person agency use if we bill on retainers and need QuickBooks to sync.”

Keyword research still tells you what people type into search boxes. It no longer tells you what they ask.

The gap you are being asked to manage

Put all this together and you arrive at the honest position, which most vendors will not lead with:

Google Search Console will show you AI impressions but not clicks, click-through rate, queries, or position. Bing’s tools will show you citations but not traffic value. Analytics will catch some referrals and lose the rest to Direct. And the dominant effect — being recommended in an answer nobody clicks — leaves no trace anywhere.

You are being asked to manage a channel where the cost is measurable and the benefit is not.

That is uncomfortable, and it is also the actual job. Chapter 11 lays out how to do it without lying to yourself or your board. But the discomfort is worth naming early, because every bad decision in this field starts with someone pretending the gap isn’t there.

What not to conclude

Two wrong turns are common at this point.

“AI search is hype, the traffic is a rounding error, ignore it.” The traffic is a rounding error. The influence is not. Sixty-eight percent of searches ending without a click means the research is happening somewhere; it just isn’t happening on your site anymore. Waiting for the referral numbers to justify the investment means waiting until the positions are taken.

“AI search changes everything, rip up the playbook.” It doesn’t and you shouldn’t. As Chapter 4 shows, most of what determines whether you appear in an AI answer is whether you were findable in the first place. The playbook gets a new chapter, not a new cover.

The right posture is in between and slightly boring: treat AI visibility as a consequence of being a well-defined, well-corroborated, technically accessible business — and fix the parts of that you have been neglecting.

Which requires knowing what the machine actually does. That’s next.