Why I named the AI Selection Problem

The AI Selection Problem is the loss of control over the path between being discovered and being chosen. Brands are losing control over the path between being discovered and being chosen. I named it because the industry had a word for the symptom, visibility, and no word for the thing that actually changed.

Here is what changed. Search was measured in clicks because the click was where discovery ended and a visit began, and for most of the web's life that was a fair unit. It is no longer the unit. In January through April 2026, 68 percent of US Google searches ended without a click, and only 276 of every 1,000 searches sent a click to the open web (SparkToro with Similarweb, 2026-06-09). When an AI Overview appears, the number one organic result loses 58 percent of its clicks (Ahrefs, data through December 2025, published February 2026). Inside the summaries themselves, users click a link on about 1 percent of visits; Pew's panel found 8 percent of visits produced a link click when a summary was present versus 15 percent without one (Pew Research, 2025-07-22). Meanwhile Google reports 2.5 billion monthly users of AI Overviews and 1 billion of AI Mode (Google I/O, 2026-05-19), and AI referrals still account for about 1 percent of site traffic (Conductor, updated 2026-07-06). Attention moved. The clicks did not follow it. Every instrument marketers relied on kept counting the thing it could count, and the reports kept looking fine while the buying decision moved somewhere the reports could not see.

Calling that a visibility problem feels right, and it is the feeling I had to argue myself out of. Visibility is one state in a chain: discovered, understood, relevant, trusted, considered, recommended, selected. A brand can hold the first state and lose every one after it, and a single visibility score will not tell you which. It will tell you that you appeared. It will not tell you which decisions were worth appearing in, what the appearance amounted to, or what to change first. The problem was never that brands could not be seen. It was that they stopped being chosen, and nobody was measuring the choosing.

If you can name the ten buyer questions that decide your category, say on which surfaces and under which conditions your brand enters the shortlist for each, and tie one of those shortlists to a number in your pipeline, you do not have the AI Selection Problem. If you can only say how often you appeared, you do, and the name is the first step toward the reading.