Main idea

Asking an AI assistant to "find me thirty good sources on this subject" means deciding the size of the research before knowing what is missing. The number then becomes the implicit measure of quality: thirty sources give an impression of coverage.

Yet thirty documents may carry only a few repeated ideas. They may come from the same kind of player, cite the same work, share the same assumptions and miss the same contradiction. Volume measures the effort of collecting, not the extent of what has been understood — and the angles poorly represented in the first results stay invisible, however many documents you gather.

A large number of sources isn't the problem: some research needs hundreds. The problem is fixing it in advance, along the sequence "scope → thirty sources → reading → synthesis", instead of letting it follow from what the research uncovers.

Why it matters

It corrects a very widespread reflex in how people brief a research tool: the relevant question isn't "how many sources do I have?" but "what does my corpus already know, and what doesn't it know?".

And it explains why a large corpus can be poorer than a short one: redundancy produces confirmation, not knowledge.

Nuances and limits

A number fixed in advance is still useful as a budget — a cap on cost or time. It becomes a problem when you read it as evidence of coverage.

And a redundant corpus isn't useless in itself: independent evidence that converges strengthens a claim. What misleads is unintended redundancy, mistaken for diversity.

Open questions

• How do you detect, without rereading everything, that a set of sources merely repeats the same work under different bylines?