Before AI Gives Advice in a Meeting, Make It Ask

The budget review is almost over. Everyone has seen the same forecast, the same customer pipeline, and the same recommendation: hire two more account managers before the next quarter.

Then the implementation lead mentions that three large customers have postponed onboarding. That fact was not in the forecast. It was not secret, either. It simply lived with one person, in one part of the business, and nobody had asked for it.

Many bad meeting decisions do not begin with false information. They begin with incomplete information that feels complete because the room has repeated what everyone already knows.

AI can make that feeling stronger. A copilot can organize the shared facts, restate the apparent consensus, and produce a persuasive recommendation in seconds. The answer may be coherent and well supported by the material it received. The missing evidence remains missing.

The better first move is not another answer. It is an inquiry round.

Good teams can still miss unique information

Decision researchers sometimes use a "hidden profile" task. Each team member receives some information that everyone shares and some that only they know. The team can reach the best answer only if members bring their unique information into the discussion.

The setup resembles ordinary work more than it may sound. Sales knows why a deal is fragile. Support has seen a new failure pattern. Finance understands a condition buried in the model. The project lead knows which dependency has quietly slipped. The meeting packet contains the common picture, while the decision-changing detail is scattered across people and tools.

A study published online in Human Factors on September 8, 2026 tested this problem in 98 human-AI teams and 65 human-only teams. The researchers found that information inquiry increased information sharing and improved decision performance in both team types. Their exploratory analysis suggested that the pattern of inquiry may differ when a teammate is AI, but the central result was refreshingly practical: asking for information helped teams use more of what they collectively knew.

This was a controlled experiment, not a field study of your next product review. It does not prove that one ritual will fix every meeting. It does support a habit that costs little and addresses a familiar failure: before choosing, deliberately ask for evidence that has not entered the room.

Run a 90-second inquiry round

Place the inquiry round after the team understands the options but before the facilitator asks for preferences. If people vote first, every later fact has to fight an emerging position.

The facilitator can open with one sentence:

Before we recommend an option, what do you know that the rest of us may not, and what should we ask before deciding?

Then move through five short steps.

1. State the choice precisely

"Should we hire?" is too broad. Try: "Should we approve two account-manager hires for October, based on the current onboarding forecast?"

A precise choice gives people a target for relevance. It also exposes assumptions hiding inside the question. In this case, the forecast and the October timing are both open to challenge.

2. Give everyone a silent inventory

Allow 20 or 30 seconds for participants to scan their own knowledge. Ask them to note one fact, exception, recent change, or source that has not appeared in the discussion.

Silence matters. Without it, the quickest speaker sets the frame and everyone else searches for a response to that frame. A brief inventory gives the implementation lead time to remember the postponed onboardings and gives a quieter participant a clean way into the conversation.

3. Invite evidence by role

Do not rely only on "Anything else?" It is easy to interpret that question as a request for objections, repetition, or permission to end.

Direct the inquiry toward different vantage points:

• "What has changed since this packet was prepared?"
• "Which customer behavior does the forecast not capture?"
• "What constraint would make this plan harder to reverse?"
• "Who will inherit the operational cost if we are wrong?"
• "What would someone absent from this meeting want checked?"

These prompts are not invitations to perform expertise. They are ways to search the room. The goal is to find information, not to require every function to manufacture a concern.

4. Ask AI for the gap, not the verdict

Now the copilot can help. Instead of asking, "Which option should we choose?" ask questions such as:

• "Which claims in our discussion rely only on shared sources?"
• "What relevant function or stakeholder has not contributed evidence?"
• "Which current statement conflicts with a connected project update?"
• "What question would distinguish option A from option B?"

This gives AI a narrower, more useful job. It can track what has been said, compare the conversation with approved context, and suggest a missing question. People remain responsible for judging the source, the social context, and the consequence.

The distinction matters because fluent advice can cause premature convergence. Inquiry keeps the copilot on the search side of the decision long enough for the team to improve its information.

Do not confuse inquiry with delay

An inquiry round can become a hiding place for indecision if every question opens another research project. Put a boundary around it.

Classify each new item as one of three things:

Decision-changing evidence: A fact that could reasonably alter the preferred option. Check it now if the cost is low, or pause the decision if the consequence is high.

Follow-up evidence: Useful information that affects execution but not today's choice. Assign an owner and checkpoint without reopening the whole discussion.

Interesting context: Relevant background with no clear bearing on the decision. Preserve it if useful, then move on.

The implementation lead's postponed onboardings may change the hiring decision, so they belong in the first category. A request to redesign the onboarding dashboard probably does not. The labels protect curiosity without rewarding endless scope.

Inquiry also needs psychological safety. A participant who introduces inconvenient evidence should not be treated as disloyal to the emerging plan. Facilitators can lower that risk by thanking people for changing the information set, even when the final decision stays the same.

Preserve the question beside the answer

Most meeting records keep the chosen option and discard the inquiry that improved it. That makes later review harder. Future readers see what the team decided, but not which uncertainty mattered or what evidence would have changed the call.

For consequential decisions, preserve four lines:

• Choice: what the team selected.
• Unique evidence: the new information that materially shaped the discussion.
• Open question: what remains unknown and who will check it.
• Revisit trigger: the condition that should bring the decision back.

This is also a better input for future AI assistance. Persistent memory becomes more trustworthy when it contains the boundaries of a decision, not only a polished conclusion.

Microsoft's 2026 Work Trend Index found that AI users most often named quality control and critical thinking among the human skills becoming more important as AI takes on more work. The report also says 86% of surveyed AI users treated AI output as a starting point rather than a final answer. An inquiry round turns that broad intention into meeting behavior.

The next time a recommendation looks obvious, resist the urge to make it sound even cleaner. Ask what the room knows but has not yet shared.

Caspi supports that moment with live recap, suggested questions, contextual chat, proactive flags from connected tools, post-call action items, and persistent meeting memory. Used well, those capabilities can help a team find the missing fact before it becomes a beautifully summarized mistake.