Let them make the first move themselves

There’s a small, stubborn finding in memory research that should change how you build a first-run experience, and almost nobody designs around it.

If I give you the word cold, you’ll remember it a bit. If instead I show you hot → and ask you to say its opposite, you produce cold yourself — and later you’ll remember it noticeably better. Same word, same second of exposure. The only difference is who generated it. When the answer comes from inside your own head, the memory sticks harder. Psychologists have called this the generation effect since the late 1970s, and it’s held up across words, sentences, arithmetic, and procedures.

The mechanism isn’t magic. Producing something forces your brain to do more with it — to search, to connect, to commit — and that extra work lays down a richer trace than reading ever does. It’s a cousin of the testing effect, where being asked to recall a fact cements it better than re-reading the fact. Effort at the moment of encoding is not friction to be removed. It’s the thing that makes the memory.

Now look at how most products onboard. A tour. A carousel. A tooltip that points at the button and explains what the button does. A video. All of it is reading — the user is a spectator watching you generate the meaning on their behalf. It feels efficient, and it demos beautifully. But you’re spending the most valuable ninety seconds you’ll ever have with this person laying down the weakest kind of memory there is.

The alternative is to make the user produce the core action themselves, for real, on their own data, in the first session. Not a sandboxed practice run — the actual first thing the product is for. A notes app doesn’t tour the editor; it opens with the cursor blinking and lets you write a line. An analytics tool doesn’t explain how to build a chart; it hands you a half-built one and asks you to drag in the last field. The user generates the outcome, and the outcome is theirs. When they come back tomorrow, they’re not recalling your tutorial — they’re recalling something they did.

There’s a boundary condition, and it’s the whole game: generation only helps if the user can actually succeed at generating. A prompt they can’t answer isn’t a desirable difficulty, it’s a wall — and a failed first attempt teaches the opposite lesson. So the craft isn’t “remove all help,” it’s scaffold so the move almost always lands. Pre-fill everything except the one meaningful choice. Constrain the blank so it’s fillable. Let them produce the answer, but stack the deck so they can. Generation with a safety net beats a tour, and it beats an unscaffolded blank screen that makes them fail.

One honest caveat before you rip out every tooltip. The generation effect is about memory — it makes the specific action easier to recall next time. It doesn’t, by itself, teach a whole mental model, and there’s evidence that effortful generation of one item can crowd out memory for the surrounding context. So don’t read this as “never explain anything.” Read it as: for the one action that matters most, don’t narrate it — hand over the pen.