One Idea, Three Hundred Pages

"99% were 1 idea stretched across 300 pages. They should have been blog posts."

— Nicolas Cole, after reading more than 500 business books

I have read a lot of business books, and most of them work in the same way. The author takes one simple idea and then turns it over from every side, on a long series of examples. Peter Sims' Little Bets is a clear example of the type. The whole book is really one idea: when something is cheaper to test than to plan and assess the risks, you should test it instead of planning. I had that on the first page, the first example made it concrete, and then the rest of the book was the same idea about thirty more times.

I liked Little Bets, I still use its idea regularly, and I recommend it to other people. That part never changed. What bothered me was the form. From the first read I was sure the book was artificially inflated, that the real content would fit in a few pages, and that everything else was there to turn it into something thick enough to sell.

For a long time my model of how an idea should be presented was the opposite of this. It was Martin Fowler's UML Distilled, a deliberately thin book about a very large subject. Fowler treats the slimness as the whole point rather than a limitation:

"My proudest thing about this book is that it's under half an inch thick — which was very unusual for OO modeling books at that time."

— Martin Fowler, UML Distilled

So my verdict on books like Little Bets was simple, and for years it did not move. They are padded. There is one paragraph of real content, and the rest exists to make the object heavy enough to sell. I was right about part of that, and wrong about the rest, and this is what changed my mind.

The economics: the padding is real

Let me start with the part I had right, because it is true and worth saying plainly.

A nonfiction book has a minimum practical size. Most of them land somewhere between 50,000 and 90,000 words, and the reason is physical rather than intellectual. A book too thin to have a visible spine does not survive on a shelf, and you cannot really sell a fifteen-page argument as a 25-dollar hardcover. The format needs a certain length, the idea itself rarely supplies it, and so the examples are what end up filling the gap.

The economics make this even clearer. Most business books sell fewer than 5,000 copies, at one or two dollars of royalty each, so the book itself is not actually the product. It works much more like a business card. The real income comes from talks and consulting, where a single paid workshop can bring in more than a thousand sold copies would. Seen from that side, "one idea stretched across 300 pages" is not a failure of the book at all. It is the book doing exactly the job it was written to do.

There is even a market that sells the short version directly. Blinkist, Headway and Shortform together form a several-hundred-million-dollar industry whose entire product is these same books with the repetition taken out, and Headway alone reports more than 50 million downloads. People are paying, specifically, to get the one sentence without the other 299 pages.

So the cynical view holds up well enough. The trouble is that it only explains why the books are built this way. It does not explain why some of them actually work on the reader.

What I saw when I explained things to people

The other half of the answer did not come from books at all. It came from years of explaining things to people.

The pattern is always the same. You explain a task, the person agrees and tells you it is clear, and a week later it turns out it was not clear at all. What took me far too long to accept is that they were not pretending. They were genuinely sure they had understood it. Either they did not ask the question that would have exposed the gap, or they could not see the gap themselves. At first I thought this was about particular people. Then I started seeing it in almost everyone I worked with, and eventually I saw it in myself.

This turns out to be well studied, and the experiments are more direct than you would expect.

In 2002, Rozenblit and Keil asked people how well they understood ordinary objects, such as a zipper, a flush toilet, or a helicopter. Then they asked them to explain, step by step, how each one actually works. The moment people tried to produce the explanation they were certain they already had, their own rating of their understanding dropped by almost a full point on a seven-point scale. The effect is strongest for how-and-why knowledge, which is exactly the kind of knowledge a nonfiction book is made of. The name for this is the illusion of explanatory depth.

Roediger and Karpicke ran a study like this in 2006. They split people into two groups. One group simply reread the material. The other group put the text away and practiced recalling it from memory instead of reading it again. The rereading group was the more confident that it would remember. A week later it actually remembered less, scoring 40 percent against 61 percent for the group that had practiced recalling. The group that felt more sure of itself was the group that did worse. Rereading feels easy the second time through, and it is that easy feeling that the mind mistakes for actually knowing the material.

So when someone says they understood something and they actually did not, it is not because they were careless, and it is not because they are not capable. It is a known and repeatable mistake in the way people judge their own understanding, and I make it just as easily as anyone else does.

The same mistake on my side

There is a matching error on my side of the table, and it is harder to catch because I cannot feel it from the inside either. It is called the curse of knowledge. In one well-known study, people tapped out the rhythm of a famous song and predicted that listeners would recognize it about half the time. The listeners recognized 3 out of 120, which is a little over two percent. Once the tune is already playing in your head, you simply cannot model the person who does not have it there.

When an idea is obvious to me after a single sentence, I am the person tapping out the tune, and I cannot honestly judge how many passes the other person needs. So "it is padding" and "they said they understood but they did not" turn out to be the same error seen from two ends. The reader believes more arrived than really did, and the writer believes one pass was enough. Both mistakes push in the same direction, and the repetition in a book is what corrects for both of them at the same time.

Where the science stops agreeing with me

Here I have to be careful, because the convenient version of this argument is actually wrong.

The science does not support "say it a hundred times." Rereading the same text is in fact one of the weakest ways to learn anything. In Dunlosky's 2013 review of study techniques, plain rereading and highlighting came out near the bottom, while self-testing and spaced practice came out on top. A book that repeats the same sentence thirty times in a row really is bad padding, and on that specific case the cynical view is entirely correct.

What actually works is not repetition but re-encounter: meeting the same idea again after some time has passed, in a different form, and ideally while you reconstruct it yourself rather than just reading quickly past it. A book that comes back to its central point through a different chapter, a different case and a different angle is doing exactly what the evidence supports. A book that comes back to it word for word is not. That, in the end, is the difference between a book that changes how you work and a book you only finish out of stubbornness. The thing that mattered was never the number of repetitions, it was their variety.

Why different examples, not more repetition

The clearest result in this whole area is the one about examples specifically.

Gick and Holyoak gave people a puzzle that almost no one could solve on their own. First they gave some of them one story that had nothing to do with the puzzle on the surface but was solved in the same way, and about 30 percent of those people then solved the puzzle. Then they gave another group two such stories, different on the surface but solved by the same trick, and about 52 percent of that group solved it. Seeing the same trick in two different disguises is what let people notice the trick itself, apart from the story it was wrapped in. Rohrer and Taylor found the same thing in ordinary math practice: students who practiced several problem types mixed together did far better on a later test, 63 percent, than students who practiced one type at a time in blocks, who scored 20 percent.

With a single example, you tend to remember only that example, tied to its own story. With several different examples, what stays with you is the general idea itself, and you can apply it to your own case, one the author never described. So "one idea, ten different stories" is not the author padding the book out. It is the only reliable way to make the idea usable in a situation other than the one on the page. The examples that felt repetitive were the part that was actually teaching you. They felt repetitive only because you cannot feel the learning happening while you read.

What I take from this

So my old verdict was only half right. Here is what I am actually taking from it.

The repetition in these books is doing real work, and I need to learn to do the same thing on purpose. If I want an idea to stay in a reader's head, saying it once is not enough. It has to come back, in different forms, until it is wired in and not just heard once.
This looks a lot like training a model. To fix a fact in a neural network you show it again and again, in different forms, until it settles into the weights. A book does the same thing, except the network being trained is the one in the reader's head. Different hardware, same need for repetition, and the neurons still have to do the work.
And this openly contradicts DRY and the principle of parsimony, which I have argued for for years, including in less documentation and more signal. Say a thing once, never repeat it: that is right for code and for reference docs. It is wrong for getting an idea to stay in a person. So the question I am left with is not "fewer words or more words." The information itself should still be stored parsimoniously: written once, tight, nothing repeated. The repetition belongs in how it is delivered, not in the source. You take that one tight version, give it to the person again in different forms, and check what actually stuck by making them recall it instead of rereading it. An LLM could sit right there, turning one parsimonious source into the varied repetition and the recall checks, different for each person. There is something worth thinking about here.

And this article will probably need to be repeated in other forms too. Right?