In October 1994 a reader named Terry Mullen sent a short letter to the St. Paul Pioneer Press describing something small and slightly ridiculous. He had encountered the name Baader-Meinhof once, the German militant group, and afterwards kept running into it: a radio segment, a magazine caption, a passing remark at a table. Nothing had changed in the world. What had changed was that the name was now loaded in his head, and once loaded it seemed to fire everywhere. He proposed a name for the experience and the paper printed it, and other readers wrote in to report the same thing happening with a word, a car model, a medical diagnosis, a surname. Stanford linguist Arnold Zwicky gave it a cleaner name in 2005 on Language Log: the frequency illusion. The term is duller than the experience, which feels less like an illusion than like the world quietly turning up the volume on one channel.
🧠 The Man Who Kept Meeting the Red Army Faction
What Mullen described has a specific and unglamorous cause, and the interesting part is how little of it involves being wrong.
Say you learn a new word on a Tuesday. On Wednesday it appears twice: once in a headline, once in something a colleague says. On Thursday, three times. By the weekend you are telling someone that this word is suddenly everywhere, that you have been hearing it constantly, that something has shifted. And something has, just not in the world.
The explanation Zwicky landed on in a 2006 paper has two moving parts, selective attention and confirmation bias, and the order in which they act matters more than either one alone.
⚙️ How It Works
First, the gate gets recalibrated. Your attention is not a window, it is a bouncer. Most of what reaches your eyes and ears is filtered before it ever becomes conscious experience, because the alternative is drowning. A concept you have just learned, or one that carries emotional charge, gets a higher weight in that filter. It moves from the category of background noise to the category of things I can see. This is not a decision you make. It happens to you, the way your name cuts through a crowded room.
Second, the hits leave marks and the misses leave nothing. Each time the word appears, you register it as an event. You remember noticing. But the hundreds of minutes in which the word did not appear leave no trace at all, because absence has no texture. There is nothing to count. So your evidence accumulates on one side only, and the denominator quietly disappears.
Third, confirmation bias reads that lopsided ledger as proof. Once the hypothesis “this word is everywhere” is live, every sighting becomes support for it. The places where it should have shown up and didn’t — a whole article about language, a whole conversation, a whole day — never get entered into the account. Zwicky’s phrasing is blunt: we notice what is salient to us and disregard the rest; we look for what supports our hypothesis and disregard possible counterevidence.
Fourth, the loop closes. Perceived frequency raises salience, higher salience produces more hits, more hits raise perceived frequency. The output of this machine has a distinctive shape: a step change. Not more often than before, but all at once, out of nowhere. Real change in the world usually arrives as a slope. This arrives as a cliff, which is exactly why it feels like news.
There is one modern addition Zwicky did not have to account for. Search for a term once and the recommendation systems take the hint. They feed you more of it. At that point the illusion and the reality converge, and the true frequency really does rise, which makes the whole thing much harder to see through.
🤔 The Uncomfortable Part
The usual reading of this is that you were fooled. I don’t think that’s the right frame. Both processes are doing their jobs. Selective attention is what keeps you functional in a world that produces more signal per second than you could ever process. Confirmation bias is what lets you form a hypothesis quickly instead of standing in permanent suspension of judgment. The failure is not in the machinery; it is in the confidence you place in its output. Your felt sense of frequency is a reconstruction, not a measurement.
Two corollaries are worth keeping.
The filter does not distinguish between loving a thing and hating it. It only distinguishes between caring and not caring. Start finding a phrase irritating and it will also start appearing everywhere, and the irritation will feel like evidence about the world rather than a property of your own attention.
And the illusion has a twin pointing the other way. Zwicky called it the recency illusion: if you have only just noticed something, you conclude it must be new. Generations of readers have been certain that a particular usage is a recent corruption, when the same complaint can be found in print a century earlier. The same machinery, running backwards.
🔗 Where This Bites
The place this costs the most is the moment right after you form a hypothesis.
You pick a sector. You decide that a certain kind of company is undervalued, or that a technology is about to break, and then for two weeks you cannot open a feed without seeing confirmation. It feels like the market is agreeing with you. It is not. You have simply installed the filter that lets the signal through, and you have stopped keeping the denominator. The fix is boring and it works: write down the count in a fixed window before you have a theory, then count again after. If the number went from two to six because you started looking, that is a fact about your looking, not about the market.
It happens with names, too. Choose a name for a character and the name will turn up on a book spine, in a byline, in a street sign, within the week. Nothing has changed in the culture. Your retriever came online, and a retriever that is online finds things.
For anyone building a product in a category that doesn’t have a name yet, the same effect is an asset and a trap at once. The first time a user hears the category described, they will start seeing it everywhere, which is a real window for adoption. It also means the founder’s sense of “this is exploding” can be manufactured entirely by their own attention, and a market-size estimate built on that feeling will be wrong in a specific and predictable direction.
🎲 The Five Illusions
Zwicky eventually noticed that the frequency illusion belongs to a family, and the family is a better lesson than the single member. He listed five: the recency illusion (it must be new, I only just noticed it); the antiquity illusion (I have always done this, so everyone always has); the out-group illusion (that novelty is characteristic of them); the in-group illusion (this quirk is ours alone); and the frequency illusion (now that I see it, it is everywhere).
One machine, five exits. All of them run on the same handicap he named in the paper’s opening: hardly anyone has a panoramic view of how a language, or a market, or a culture actually varies. We are all working from a personal sample and mistaking it for the population.
So here is a small exercise. Pick the word or phrase you are currently convinced has taken over. For one week, keep a tally of every time you encounter it, and also note, roughly, how many hours you spent in situations where it plausibly could have appeared. You will end up with the number you never had before: a denominator.
Most of the time it will turn out to be smaller than it felt. Occasionally it will be larger, and then you will have learned something the illusion could never have told you.