Picture the owner of a corner bakery, three days before a Facebook ad launches for a new sourdough starter kit. She has written the headline, deleted it, written it again. She reads it for the eleventh time and thinks: this is obvious. Of course it's obvious. She has explained starter cultures to customers for six years. The problem is that her certainty is exactly what should worry her.
There is a name for what is happening in her head, and it isn't sloppiness. It's called the curse of knowledge, and the first demonstration of it wasn't about advertising at all. In 1990, a Stanford graduate student named Elizabeth Newton ran an experiment where one group of people tapped out the rhythm of a well-known song on a table, and another group tried to guess the song from the tapping alone. The tappers predicted that listeners would identify the song about half the time. The real number, as the research recounted in The Curse of Knowledge in Product Copy puts it, was stark: "The actual identification rate was two and a half percent." The tapper hears the melody in their head while they tap. The listener hears only knocking.
Why rereading your own ad doesn't work.
The instinct, when an ad feels slightly off, is to read it again, slower. That instinct assumes the problem is attention: if you just look carefully enough, you'll spot the confusing part. But the bakery owner's brain isn't failing to pay attention. It's doing something closer to the opposite — it already knows what a starter culture is, what "feeding" it means, why day three matters, and it fills every one of those gaps automatically, without telling her it did so.
That filling-in is not a bad habit she could break with discipline. The research points to a mechanism: once you know your product, your brain shortcuts the steps a first-time customer has to think through. You can't easily turn that off through willpower — the neural shortcuts that make you fast and fluent on your own product are the same shortcuts that make you blind to what a stranger doesn't have yet. fMRI studies cited in that piece go further, showing that experts and novices looking at the same material actually light up different parts of the brain: experts use the regions tied to retrieval and pattern-matching, novices use the regions tied to effortful, uncertain processing. Two different programs are running on the same screen, and only one of them is the customer's.
That's the uncomfortable part for a solo owner: self-awareness doesn't fix this. Knowing you're an expert doesn't let you turn the expertise off. You need a different person, or a structured substitute for one. Since you can't hire a second set of eyes for every ad you write, and it's the problem we're building Knowing to solve.
The five-second test, done properly.
The method for this has existed in product and startup circles for years under a plain name: the comprehension test, sometimes called the five-second test. The version laid out at Kromatic's Real Startup Book, a testing-methods guide used by startups and agencies, is simple enough to run with a phone and a few willing neighbors. Show someone your ad for about five seconds. Take it away. Ask them to explain, in their own words, what it said and what it's selling.
The test is not asking whether they liked it, and it is not asking whether they'd buy it. Those are different, later questions — about desire and price, not clarity. This test answers one narrower thing: did the words land. The guide is specific about why you don't just ask "did you understand it?": people say yes reflexively, to be polite or to end the conversation. "Half of them get it wrong, telling you the problem is your messaging, not your idea," the piece notes — and you only find that out by making them demonstrate the understanding, not report it.
The tapper hears the melody in their head while they tap. The listener hears only knocking.
There are real benchmarks here, not vibes. Eight or more out of ten: your message works, and any remaining problem lives elsewhere, maybe in the offer or the price. Five to seven: there's partial clarity, worth a targeted rewrite of the specific phrase that tripped people up. Below five: the clarity problem is fundamental, and no amount of testing the offer or the audience will help until the words themselves are fixed first.
Running it without turning it into a chore.
Methodology matters more than people expect. The test has to use the exact copy you intend to run, word for word, not a version you summarize out loud while gesturing at your phone. The guide is blunt about why: you're testing the wording itself, and paraphrasing it defeats the purpose before you've started.
The testers don't have to be your exact target customer — don't stress about finding ten people in your exact situation. A plumber testing an ad for emergency drain service doesn't need ten homeowners with a burst pipe right now; he just needs ten people who can read and understand at the level his real customers would. Ten is the floor, twenty is better, because a handful of answers can mislead you either direction.
The bakery owner texts the ad image to five customers and five people outside her circle — a cousin, a neighbor, someone from the gym. She asks each one: "Look at this for five seconds, then tell me what it's selling and who it's for." Three of ten mention "bread kit." Two say "some kind of subscription box." One says, flatly, "I have no idea." That's six accurate-or-close answers out of ten — partial clarity, not a disaster, but a specific phrase in the headline needs to go.
There's a shortcut worth knowing about and also worth not trusting too far. You can paste your ad copy into an AI tool and ask it to explain the ad back, as a cheap first pass before you bother a single human. It will catch the obvious stuff — a headline that's genuinely nonsensical, a claim with no noun attached to it. But the Kromatic guide flags the real limit plainly: "AI has far more reading comprehension than a real person scanning your page for 5 seconds." An AI reads carefully by default. Your customer, scrolling a feed at a stoplight, does not. The AI pass is a filter, not a verdict.
What a clarity problem actually costs.
It's tempting to treat this as a nice-to-have, something you'd get to after the budget and the targeting are locked in. The economics argue the opposite. The product-copy research runs a simple hypothetical: at a $50 cost per click, raising a landing page's conversion rate from 3% to 4% drops the cost of acquiring a customer from $1,600 to $1,250 — a 22% reduction, from one percentage point of clarity. Nothing about the offer changed. Nothing about the targeting changed. The only thing that moved was whether people understood what they were looking at.
The same four specific failures recur across badly-written ad and landing-page copy:
- Explaining a feature instead of what it does for the customer.
- Using an internal term the customer has never heard.
- Assuming context the reader doesn't have.
- Getting oddly specific too early, before the reader even knows why the specific thing matters.
Frequently asked
Common questions.
How many people do I need to test an ad on before I trust the result?
Ten is the minimum worth acting on; twenty gives a more reliable read. Fewer than ten and a couple of confused answers can make a clear ad look broken, or a couple of lucky guesses can make a confusing one look fine.
Should I ask testers if they understood my ad?
No. People tend to say yes to be agreeable, even when they didn't follow it. Show the ad, take it away, and ask them to explain back what it's selling and to whom — their explanation tells you the truth their yes-or-no answer won't.
Can I just use AI to check if my ad copy is clear?
It's a useful first filter for catching obviously broken copy, but not a substitute for real people. AI reads more carefully and patiently than someone scrolling a feed for five seconds, so it will pass copy that a distracted human would still find confusing.