Separate fluent prose from supported claims

Generation and verification are different tasks

A language model can produce plausible text by learning patterns and using context. Its polished wording does not itself establish that its factual claims are correct. A useful draft can still contain a mistaken date, invented detail or unsupported inference. Inspect the claims rather than assigning truth from tone, length or confidence. These lessons use simulated outputs written for the classroom; no learner needs to call a model.