The Buyer Journey Your Attribution Can't See

A demand gen lead I work with pulled up her dashboard last month and pointed at a single line: direct traffic, up nearly 40% in a quarter with no new ad spend, no viral post, and no campaign behind it. Google Analytics had no source, no referring page, nothing. Two weeks later three of those visits turned into demo requests, and all three buyers opened the call already knowing the company's pricing, a competitor's pricing, and a specific complaint about one of them. Nobody on her team had sent any of it. The buyers had already done the research. The dashboard simply had no record of where.

That gap is not a fluke, and it is getting bigger fast. G2's 2026 research found that 51% of B2B software buyers now start their research with an AI chatbot rather than a search engine. Pew Research found that AI summaries now appear on roughly 18% of Google searches, and when one shows up, only about 1% of users click through to an actual page. And when AI-driven visits do reach a website, research from The Digital Bloom found that roughly 70.6% of that traffic arrives with no referrer header at all, which means most analytics platforms quietly file it under "direct," indistinguishable from someone who typed your URL from memory. Your buyer did the work. Your dashboard just cannot see it.

Attribution Was Built for a World of Trackable Trails

Every attribution model marketers use, first-touch, last-touch, multi-touch, was built on one assumption: a buyer leaves a trail of clicks, and if you capture enough of them, you can reconstruct the path from stranger to customer. UTMs, referrers, cookies, all of it exists to catch that trail.

That assumption held for two decades because research happened on pages you could tag. It does not hold when research happens inside a conversation. A buyer asking ChatGPT or Gemini to compare vendors, summarize a category, or draft a shortlist is generating exactly the same evaluation work a decade of martech was built to track, except none of it produces a click, a UTM, or a referrer. The research still happens. The trail just never starts.

A Lit Marker With No Visible Path to It

Here is the part that should worry any team still reporting on channels the old way: the buyers who arrive through this invisible path are not lower intent. If anything, the opposite. I wrote about this shift when it first became visible in The AI Buyer Is Already Here: agents are not just delivering traffic, they are pre-filtering it, and the humans who show up on the other side of that filter have often already compared you, shortlisted you, and half-decided before your team knew they existed.

That is a strange thing to explain to a CFO. The pipeline is real. The revenue is real. The dashboard that is supposed to prove it worked shows nothing, because the marker lit up on its own, with no visible path leading to it. A channel report built entirely on visible trails will systematically undercount your best-converting source and quietly credit it to "direct," "branded search," or nothing at all.

Chasing a Better Tracking Script Is the Wrong Fix

The instinct here is to go build better tracking: parse AI referrer strings, tag every LLM domain, buy a tool that promises to unmask "dark AI traffic." Some of that is worth doing. None of it solves the actual problem, because a meaningful share of this research will never produce a trackable signal no matter how good your script is. A conversation inside someone else's chat window is not a page you can instrument.

Chasing perfect attribution here is the same mistake I described in The AI-Native Company: bolting another isolated tool onto the stack in the hope it finally produces the whole truth. You cannot instrument your way out of a structural blind spot. What you can do is stop treating the blind spot as evidence that nothing happened, and start building for it directly.

Build the Intelligence, Not Just the Instrumentation

Two things actually move the needle here, and neither of them is a new tracking pixel.

First, be present in the conversations you cannot see. If a buyer's evaluation happens inside an AI tool's answer, the only lever you have is whether your positioning, pricing, and proof points are structured clearly enough for that system to surface you accurately when asked. That is a content and data problem, not a tracking problem.

Second, stop requiring a clean click trail before you act on a signal. An unexplained lift in direct traffic, a demo request from someone who already knows your competitor's pricing, a spike in branded search with no campaign behind it: these are not noise waiting to be explained away. They are directional signals from a channel your tools cannot fully see, and a business with a connected model of its buyers, competitors, and market can read them as evidence and act, instead of waiting for a UTM that is never going to arrive.

The Takeaway

The buyer journey did not get shorter. It got quieter. Half of B2B buyers are now starting their research in a chat window your analytics cannot reach, and the fraction of that traffic that does land on your site mostly arrives looking like nothing happened at all. Rebuilding your tracking stack will not fix this, because the gap is structural, not technical.

The teams that win from here will not be the ones with the cleanest attribution report. They will be the ones who stopped needing a perfect trail to know something real is happening, and built an operating model that can act on the signal anyway. If your "direct" traffic just jumped for no reason you can name, that is not a data problem. It is a buyer who already made up their mind somewhere you were never going to see.

Tracy Thayne* is the founder of Expona, an AI-powered operational intelligence platform for B2B marketing. Read the Expona founder story or subscribe to the blog (below) for weekly insights on context, AI, and the operating model of the next decade.*