In 1968, a psychologist named Robert Zajonc showed people random Chinese characters — some once, some twenty-five times — and asked which ones felt more "good." People consistently rated the repeated ones higher. Not because they understood them. Because they'd seen them before. He called it the mere-exposure effect: familiarity reads as trust, even when nothing about the thing itself changed. Fifty-eight years later, that finding quietly explains why a company can rank #1 on Google and still not exist to ChatGPT. 🍂
Ranking and being trusted turned out to be two different games. Here's the data behind that split, why it works the way it does, and what's actually still open.
The split, measured
Ahrefs tracked which pages get cited inside Google's AI Overviews against which pages rank in the ordinary top 10 for the same query — 863,000 keywords, 4 million AI Overview URLs. Seven months earlier, 76% of citations came from a top-10 page. By March 2026, that had fallen to 38%.
Share of AI citations pulled from a top-10 page
Independent assistants read even less of the ranked web: a separate 15,000-prompt study across ChatGPT, Gemini and Copilot found only 12% of what they cite ranks top 10 on Google at all.
If your whole plan was "rank well, get cited," that plan is losing ground every month. Something else is doing the deciding.
Why repetition wins
A language model doesn't retrieve your page and check its position. It predicts what's likely given everything it has read — and a name repeated consistently, in many independent places, by people who aren't you, looks a lot more like the "familiar" signal Zajonc found in 1968 than like a ranking score. It's the same psychology as a jingle you can't get out of your head: repetition doesn't need to be persuasive to work. It just needs to happen enough, from enough angles, that the answer feels obvious.
Semrush tested this at scale — 50,000 brands, 1,094 categories, five buyer-style prompts per category. Traditional SEO strength barely predicted who ChatGPT recommended: branded search volume correlated with topic ownership at just 55.7%. Weak enough that ranking clearly isn't running the show. What separated the winners was coverage — showing up, described the same way, across the reviews, forums and comparison pages the model actually pulls from.
The empty room 🕯️
Here's the part that should change your mood about all this: most of the room is still empty. Of those same 1,094 categories, 53.7% had no brand appearing in even 3 of 5 prompts — no established answer at all. Not a competitive field with a leader to dethrone. A blank page waiting for whoever gets repeated first.
The three questions people actually ask me
- "I rank #1 — why doesn't ChatGPT know me?" Ranking indexes your page. AI trust indexes your name, repeated on pages you don't own.
- "Do I need PR deals like the big publishers?" No — that's a different game with different economics. Unlinked mentions (reviews, directories, forums) work on the same mere-exposure logic and cost nothing but asking.
- "Which AI tool gets me cited?" None of them, directly. Tools draft faster. Only being talked about, repeatedly, by people who aren't you, earns the mention.
Google's Universal Commerce Protocol and Gemini's Gmail-connected search both point the same direction: agents will soon compare and buy on a person's behalf, not just answer their questions. If that lands the way it's being built, the mere-exposure effect stops being a slow brand-building idea and starts compounding at machine speed — the businesses repeated early become the businesses every future agent defaults to, the same way early reviews snowball into permanent bestseller status. Nobody has measured that yet. But the room won't stay empty forever, and the cost of finding out later is higher than the cost of showing up now.
The one-sentence version: stop optimizing for a machine that checks your position, and start showing up — the same way, everywhere — for a machine that's building a memory.