The Capture Fallacy

In 2013, a researcher named John Dunlosky published a meta-analysis of ten common study techniques, reviewing decades of experimental literature to rank them by effectiveness. Highlighting came near the bottom. Not because it's rare — most students use it constantly — but because study after study showed it produces almost no learning benefit beyond simply reading the text. Some research found it actively harmful: by drawing attention to individual sentences, it can prevent the connections that make information useful later.

This finding made almost no difference to anyone. By the time Dunlosky published, Kindle highlights were already accumulating in millions of devices. Readwise arrived in 2018 to resurface them. By 2024 the PKM ecosystem had grown elaborate: web clippers, highlight importers, daily review queues, integration pipelines routing passages from Kindle to Notion to a Zettelkasten. Entire workflows built around the one step that research consistently shows isn't the bottleneck.

The gap between having and knowing

Christian Tietze named this pattern in 2014 as the collector's fallacy: the belief that acquiring information is the same as understanding it. "To know about something," he wrote, "isn't the same as knowing something." A highlighted passage in Readwise is closer to a sticky note on your desk than a thought in your head.

The cognitive science makes this precise. Slamecka and Graf showed in 1978 that people remember what they generate far better than what they passively receive. In their experiments, participants who completed a word from a partial cue (KING-CR_\_\_) remembered it significantly better than those who read the full pair. The act of generating, even something as small as filling in letters, produced a stronger memory trace than exposure alone.

Applied to notes: what makes a note useful later isn't its existence in a folder. It's the moment of production. Writing a sentence in your own words about something you read does more for retention than the note itself. The note is the residue of a cognitive act, not the act.

So systems designed around frictionless capture are often solving the wrong problem. Some friction at capture time is the point. The slight effort of translating something into your own words, of deciding where it fits, of asking what kind of thing this even is — that's not inefficiency. That's encoding.

What typed entities are for

Most knowledge tools give you a blank note and a timestamp. You can write anything, and you do, and it sits there. Later it surfaces in search, or it doesn't.

Typed entities work differently. To create a person record, you fill in fields: how you know them, what matters, what you last discussed. To create a decision record, you need context, alternatives, the choice, consequences. These schemas aren't purely organizational. They're prompts. They require a minimal generative act — you have to decide what something is before you can store it.

That's a small version of the generation effect applied at capture time. Not deep learning, but not passive either. The empty field doesn't just organize your knowledge. It generates a little of it.

I've noticed this in practice. Notes taken with no structure are rarely useful later, not because I forget them, but because they were never quite thoughts to begin with. They were sightings of thoughts. Things I've taken time to classify — project decisions, preferences, context about people — have a different quality when I read them back. There was more cognitive work in the writing. It shows.

The tool that doesn't ask

The irony of highlight resurfacing tools is that they promise active learning. Spaced repetition, daily review, the dopamine of a well-timed callback. But the daily review is still passive: you're looking at a line you underlined, not reconstructing why it mattered. Spaced repetition helps with recognition. It doesn't replace the processing step that most readers skipped in the first place.

The bottleneck in personal knowledge isn't having enough information. Most people have more than they can use. It's whether the information passed through any act of understanding on its way to storage. Whether you made something with it, even briefly.

A knowledge base full of unprocessed captures is like a library organized by spine color. Everything looks present. Nothing is really findable when you need it, because findable isn't about location. It's about whether you understood something well enough, at the time, to know you'd want it later.

Asgeir Albretsen is the founder of Harbor.