Structure beats freeform text for working with AI

Ask an AI for "a summary of this contract" and you get fluent prose you have to read closely to trust. Ask it to fill in named fields — parties, term, payment, termination, governing law — and you get an answer you can check one box at a time. The freeform version reads well and hides its gaps; the structured one tells you instantly when a field came back empty or wrong.

A defined shape — fields, a schema, a template — is a cognitive artifact: it carries part of the thinking outside your head and pins down what a good answer must contain before the model ever runs. That makes the request precise and the output verifiable line by line. It's the same reason Knowledge compounds when it's linked, not just stored rather than piling up as prose, and the same discipline that produces A good explanation is hard to vary.

Structure isn't bureaucracy. It relocates the checking to once, up front, in a form you can reuse — so you're not re-reading every output from scratch. See the frameworks for cognitive artifacts.

Next: judge explanations by what they commit to, since A good explanation is hard to vary.