I operate a small compliance software product. This spring, over three months, Google showed its pages 403,257 times. Those impressions produced 5,055 visits. I want to walk through what happened inside those two numbers, because I have since learned that some version of this is happening to almost every small site, and most operators are misreading it the way I nearly did.
Start with where the pages ranked, because that is what makes the numbers strange. The site's biggest page sat at average position 5.25, first page of Google, above the fold, 109,592 impressions in the quarter. For twenty years, a page at position five earned somewhere between five and seven percent click-through. Mine earned 0.93 percent. Its siblings at positions four through six earned 0.5 to 0.9. The rankings were fine. The clicks were gone.
The tell was in the query data. I pulled every long, natural-language query, the eight-words-and-up questions people type when they are really asking something: 110 of them, 11,223 impressions, average position six. Total clicks: 51. Under half a percent. Those conversational queries are exactly the ones that now trigger an AI-generated answer above the results, and an industry study across 2.43 billion impressions found the same collapse: click-through on AI-answered queries fell from 1.76 percent to 0.61 at the worst, roughly a two-thirds drop. My pages were being read, summarized, and answered on the results page itself. The impression happened. The visit did not.
Here is the mistake I nearly made, and it is the reason I am writing this down. The dashboard said the acquisition channel was dying, and the dashboard was measuring the wrong thing. In the same period the product kept receiving inbound leads, more than ten in a stretch, all organic, arriving by paths the search console cannot see: answers people got from assistants, journeys that started in an AI summary and ended at the contact form, engines other than the one I was staring at. The proxy said famine. The outcomes said otherwise. When your instruments contradict your outcomes, believe the outcomes and go find out which invisible path is feeding you.
So the playbook, learned the expensive way:
Change what you grade. Impressions are now inflated by machine reading that can never click, and sitewide click-through is polluted by informational pages that AI answers fully. The number that matters is clicks and conversions on the pages where buyers land. I stopped reporting the big vanity totals entirely; they measure how much the machines read me, which is a different fact from how many humans need me.
Instrument the invisible paths. We added one optional question to the contact form: how did you find us, with AI assistants as an explicit option alongside search. It is the cheapest analytics investment available in 2026, because it measures the only channel the analytics stack structurally cannot.
Stop resenting the machine readers and feed them deliberately. If the assistant is going to answer with your content anyway, the war is over who gets named in the answer. That means clean, quotable pages, claims stated plainly, and machine-facing surfaces kept accurate, because the summary is now the storefront.
And keep writing the informational content, but reprice it in your head. Those pages no longer buy visits. They buy presence in the answers, which is upstream of the leads you cannot trace. Mine cost the same to write as ever and pay in a currency the dashboard does not display.
The uncomfortable summary: my pages had their best-read quarter in history and their worst-visited one, and the business grew anyway. The meters were built for a city where reading and visiting were the same event. That city is gone. Measure accordingly.