What should accompany a recommendation prepared with AI?
Keep a short decision record whenever AI helps prepare a recommendation. Write down the question, the sources, the date you checked them and any assumptions that affect the choice, and separate the facts you were given from the conclusions your team drew. Then someone accountable can question the decision later.
A recommendation is hard to review when the explanation reads better than the evidence behind it. The trust questions raised at the summit apply to that supporting material as much as to the final answer.
List the alternatives you seriously considered and say why you turned each one down. A reviewer shouldn't have to accept a ranking with no explanation. If an important source couldn't be checked, leave that limitation right next to the recommendation.
Save the version you relied on when the decision is made. A link to a page that keeps changing may not show what the team saw at the time. Follow your organization's rules for keeping evidence and sensitive information.
You don't need every conversation with the tool. Keep enough that an accountable person can see the basis, question it and come back to it when a load bearing fact changes.
What to try this week
Take your last AI assisted recommendation and write its decision record: question, sources, check dates, assumptions and the alternatives you rejected. Flag any source you couldn't check, then file it under your organization's retention rules.