Fix the source answer before adding a chat layer

Look for the answer a person would need

Take a common task and try to complete it from the current article without using prior product knowledge. Mark missing prerequisites, unexplained terms and branches that lead nowhere. A chatbot may repeat the same omissions in smoother language. The first useful improvement is often a precise source paragraph or a clearer task heading, not a more elaborate instruction to the model.

Give one document authority

When two pages disagree, identify the responsible owner and decide which instruction is current. Do not solve the conflict only inside the assistant; readers using search will still see the contradiction. Record the corrected source and its review date. If the disagreement cannot be resolved, the assistant should expose the uncertainty or refer the question instead of choosing whichever passage happens to be retrieved.

Use an answer-shaped edit

For a fictional CSV export guide, add the supported file type, required role, exact menu path and what to do if the option is absent. Keep the edit scoped to verified product behavior. A title such as Export workspace data is more useful than a vague announcement headline. The example illustrates document structure, not an assertion about a named product’s actual menus.

Recheck both access routes

After the source edit, try ordinary search and the chatbot question again. Confirm that the article can be found and that the imported collection contains the correction. A better answer can result from better source material rather than a model change; record that distinction. This makes later product comparisons fairer and avoids paying for a larger configuration to compensate for an avoidable documentation gap.

• Record the missing prerequisite, corrected passage and search phrase so you can verify that both article readers and chatbot users benefit.