In this article
This works best when you want a more useful discussion, not a machine-generated argument that proves your preferred answer correct.
Give AI thinking partners a decision to examine
State the decision, options, evidence, constraints and unresolved questions. Remove confidential material unless its use is explicitly authorised.
Cassie Kozyrkov's WorkLab discussion raises issues including confirmation bias and decision-making. The exercise below is an original meeting-preparation method, not a claim that AI eliminates bias.
In a hypothetical operations meeting, the decision is whether to pilot a revised shift handover. The brief should include the current problem, proposed change, known burden and information that remains unavailable.
Do not ask only, “Why is this a good idea?” That invites support for the conclusion you already favour.
Ask for the strongest reasonable objection
Request an objection that a thoughtful colleague could raise, grounded in the information provided. Avoid caricatures of disagreement.
Try:
Review this fictional proposal for a handover pilot. Identify the strongest reasonable objection, the assumption it challenges and the evidence needed to address it. Do not invent facts or claim to represent actual employees.
The distinction matters. A simulated concern is preparation for listening; it is not evidence of what staff believe.
Read the objection before asking for a rebuttal. Sometimes the useful response is to revise the proposal or obtain information, rather than defend it more persuasively.
Separate uncertainty from disagreement
An unresolved factual question needs evidence. A difference in priorities needs discussion. Treating both as persuasion problems can make the meeting less useful.
| Issue | Appropriate next step |
|---|---|
| Unknown preparation time | Observe or estimate through a bounded test |
| Unclear approval authority | Confirm the responsible owner |
| Different service priorities | Discuss the trade-off explicitly |
| Unsupported causal claim | Investigate alternatives |
For the hypothetical handover, staff may agree that clarity matters but disagree about how much preparation the new format should require. More polished language will not resolve that trade-off.
Ask the tool to classify its objections so you can prepare the right kind of response.
Test what would change your mind
Write down conditions under which you would revise or abandon the proposal. Then ask the tool whether your evidence addresses those conditions.
A useful question is: “What finding would make this pilot inappropriate, even if the problem is real?” This can reveal hidden commitments such as inadequate staffing, incompatible systems or an unmanageable review burden.
Be specific. “If it does not work” is too vague. Define what you would observe and who should judge it.
Do not let the model manufacture a forecast to close the gap. Where the information is absent, leave the uncertainty visible in the meeting brief.
Prepare questions, not scripted certainty
Turn the analysis into questions for the actual participants. They may know more about the process than either you or the tool.
For example:
- Which part of the current handover causes the most avoidable confusion?
- What would make the proposed format too burdensome?
- Which exception must the pilot preserve?
- Who needs to approve the test and review its results?
Use these as a starting point, then adapt them to the real discussion. Listening matters more than following a generated script.
Keep a challenge log for your preferred option
Record the objection, the assumption it targets and your response. Distinguish a response supported by evidence from one that simply sounds persuasive.
For the hypothetical handover proposal, an objection might be that the new format creates extra preparation work. A response such as “staff will quickly adapt” is an assumption. A small observation of preparation effort would provide a more useful basis for discussion.
You do not need to resolve every objection before the meeting. Some belong in the meeting precisely because the participants hold the missing information or authority.
Mark the issues accordingly: resolved from evidence, open for discussion or outside the current scope. This makes the preparation useful without pretending to have certainty you lack.
Watch for objections the tool fails to raise. It may not understand your local constraints, informal responsibilities or history. Ask a trusted colleague to identify a concern absent from the AI exercise. If their point materially changes the proposal, add it to the log and acknowledge its source accurately.
A thinking partner should widen your consideration. The challenge log lets you see whether the exercise actually did that.
Check the preparation against reliable sources
Verify any factual statement the tool introduces. OpenAI's prompting guidance supports explicit instructions; your review remains necessary when the result informs a decision.
Keep a record of your initial assumption and what the challenge changed. If nothing changes across several exercises, consider whether you are asking for agreement in disguise.
Before your next meeting, prepare a short decision brief and request one strong objection. Bring the unresolved question into the room rather than hiding it behind confident slides.
The PiOfAI hub, guide to clear manager communication and earning trust through leadership offer related ways to make your reasoning visible.
AI-assisted researched guidance from PeakBelt. Hypothetical examples are labelled in the text. See our editorial approach.