Checklist

What a reviewer asks to see.

Six questions come up in every review of an AI answer. This checklist pairs each one with the evidence that settles it.

Sooner or later every AI answer in a regulated business meets a review, and a reviewer asks the same short list of questions about it. Almost none of them are about the model. They are about the answer: where it came from, who put their name on it, and whether it was still right on the day somebody acted on it.

The six questions

• Where did this answer come from, and can you show me the source?
• Who approved it, and on what date?
• Was it still current on the day it was given?
• What did the approved version say last March?
• Who could see it, and who could change it?
• What happens to the answer when its source document changes?

Teams rarely fail these because the evidence does not exist. They fail because nobody can say which system holds it, or because the answer was assembled at the moment it was asked for and left no trace. Both are answerable, and both are easier to answer before the meeting than during it.

What the checklist adds

One page per question. The evidence that settles it, the form that evidence usually takes, what to do when you cannot produce it, and a line to write down where it lives in your own estate. Take it into the review, or fill it in a month before one.

It is not a compliance opinion and it is not advice. It describes obligations that fall on your organization and the evidence a governed knowledge system can produce toward them.