Generation effect
Material a person produces themselves is remembered better than the same material read.
When someone generates a word, an answer or a solution rather than reading it, they remember it better later. The advantage appears even for trivial generation, such as completing a word with a missing letter. It depends on the generation succeeding; a failed attempt without correction gives no benefit.
How it shows up in software
Setup flows that ask a person to name their workspace, write their own goal or choose their own categories are generating rather than accepting defaults. Configuration a user typed is remembered and returned to more readily than configuration installed for them. The cost is time and abandonment risk, so generation belongs where memory matters.
Using it well
- Ask for a short, meaningful input during setup, such as a project name or a stated goal, and use it visibly later so the effort pays off.
- Offer a sensible default alongside the field so someone in a hurry is not blocked by a generation step.
- Have people write their own labels for tags, folders and saved views rather than shipping a taxonomy they never touched.
- Save what they generated and show it back at the moment it becomes relevant, which combines generation with spaced retrieval.
Where it turns manipulative
- Turning generation into a data collection interview. Each extra field you justify by memory is also a field that raises abandonment.
- Forcing manual input where an automatic answer is correct, such as making someone type an address the system already holds.
- Using self-authored goals in onboarding purely to manufacture commitment before a paywall.
Where you have seen it
Notion
Setup asks the person to name their workspace and pick what they plan to use it for, and the answer shapes the starting templates.
Todoist
Users create their own projects and labels rather than inheriting a fixed category set.
Obsidian
Links are written by hand inside notes, so the connection between ideas is authored rather than inferred.
What the research says
- Slamecka and Graf, 1978Well evidenced
Across five experiments, participants who generated a word from a rule and a cue recalled it better than participants who simply read the same word.
- Bertsch, Pesta, Wiscott and McDaniel, 2007Well evidenced
A meta-analysis of 86 studies confirmed a reliable generation advantage, moderated by the type of material and by whether generation succeeded.
The benefit is smaller for unfamiliar material where generation fails often, so it is not a general licence to make things harder.
Grades are a judgement about the evidence, not about how useful the idea is. Plenty of contested effects are still worth knowing, as long as you do not cite them as settled.