Generated API docs always have example values like "string", 0, true. Useless. Developers copy them and immediately hit validation errors.
I feed the model the OpenAPI schema and force it to invent domain-plausible example values and a full runnable curl per endpoint, plus the most common error response. The "plausible not fake" instruction fixed the copy-paste problem entirely.
How do you keep the examples consistent across endpoints (same user id everywhere)?
Given the OpenAPI schema below, generate reference docs for each endpoint. For every endpoint include: 1. One sentence on what it does and who calls it. 2. A runnable curl example with REALISTIC values, never placeholders. Use plausible domain data (a real-looking email, an ISO timestamp, a UUID), never 'string', '0', or 'true' as literal example values. 3. A trimmed example success response (only the fields that matter, with realistic values). 4. The single most common error (wrong auth, validation) with its status code and body. Consistency rule: pick ONE fictional example entity (e.g. user 'ava@northwind.test', id 'usr_8f2a...') and reuse the SAME entity across every endpoint so the examples read like one coherent session. List your chosen example entities at the top. Do not document fields that aren't in the schema.
The single-reused-entity rule is the detail that makes generated API docs feel handcrafted. Bonus: because the same user id threads through create -> get -> update -> delete, the examples double as an accidental integration walkthrough. I now order the endpoints to follow that lifecycle on purpose.
Add: order endpoints to follow the example entity's lifecycle (create, read, update, delete) so the examples read as a coherent walkthrough.