A text can be correct and still miss the mark: it leaves out the information that should guide the decision, the tone does not fit the audience, or the conclusion makes a commitment nobody approved. When that keeps happening, look at the brief you gave the AI.
A brief has to make visible what an experienced person recognizes without explaining: why the deliverable exists, who will use it and what must be kept intact. That knowledge is part of the leader's job, and it guides the team as much as the technology.
Explain what changes the evaluation
Asking for "ten slides on the results" gives a format, not a purpose. The meeting might be about tracking an operation, deciding on an expansion or understanding a drop in performance, and the same data calls for a different structure in each case. For an expansion, the brief might say that the board needs to compare scenarios, see the assumptions and know what still needs validation.
Include the materials, the audience, the constraints and one example of a good deliverable, with a line on why it works. Without that, the tool copies the look and loses the judgment.
Tell a gap from a preference
Missing information about a deadline can stop a plan from being finished; a preference for a certain headline can be fixed in review. Ask the AI to flag only the gaps that could change the result. Sometimes writing the brief shows that the team has not yet made a decision, and better wording will not remove the need to make it.
When to let the AI ask the questions
If you know the problem but cannot yet explain everything the deliverable should cover, a short interview helps. A founder preparing a deck for partners who has not decided whether to present results, renegotiate terms or start a relationship has a key question still open; the headline can wait.
Steer the conversation so it builds on what you already said and digs only into the gaps that matter, one question at a time. Answer with examples: a deck that worked, a reaction you want to avoid, a commitment you need to protect. Then ask for a summary and read it looking for shifts in meaning: "improve the negotiation" may have turned into "explain the product better". Once there is enough understanding, move on. An interview that only opens new questions becomes another way to put things off.
A brief gets better with use
After the first deliverable, note which corrections came from criteria that existed only in your head and add them to the next version. Keep the approved summary with the project. In AI Thinking by Dani Hayoshi, this is the first move, Question: what does this work really need to achieve?
Dani Hayoshi is the author of AI Thinking. Her work connects artificial intelligence, education and entrepreneurship. About Dani.