This approach is useful when you are still framing a decision and deciding what to learn next. It does not require scores or a complex comparison sheet. OpenAI's prompting documentation, checked October 24, 2026, explains how to provide context and require missing information to remain visible. The example below applies that method to an ordinary team planning choice.
Define the decision before asking for advice
This original fictional team needs an introduction to a new internal request process. Its coordinator is choosing between a live session and an annotated guide.
| Record | What is actually supplied |
|---|---|
| Attendance information | Four of eight invited staff have confirmed they can attend the proposed session. The others have not answered. |
| Current guide | A draft exists but has not been tried by a new user. |
| Coordinator preference | The coordinator enjoys live questions and discussion. |
| Unanswered question | Can staff complete a sample request with the guide alone? |
Four confirmations do not mean the other four cannot attend. The existence of a guide does not show that it teaches the process. The coordinator's preference also does not establish what every staff member needs.
A premature answer such as “Choose live because everyone prefers interaction” relies on two unsupported steps. It turns one person's preference into a team preference and assumes a benefit without checking the actual task.
Ask what would change the recommendation
Use this prompt before requesting a final choice:
Help me examine this decision before recommending an option.
Decision to make and person responsible: [ENTER]
Outcome we need and when a decision is needed: [ENTER]
Options currently considered, including keeping the current approach:
[LIST]
Confirmed facts with source references: [PASTE]
My preferences: [PASTE SEPARATELY]
Unknowns and constraints: [LIST]
Separate facts, preferences, and assumptions. Do not convert my
preference into evidence about other people. Do not interpret missing
responses as refusal or agreement.
For each plausible option, explain the assumption its usefulness
depends on. Identify the most consequential unknown and suggest a
small check that could change the choice. State what result would
favor each option and what the check would still leave unresolved.
If recommending provisionally, name the assumption and what evidence
would reverse your recommendation. Do not invent scores, survey
results, agreement, or completed experiments. Do not schedule or send.
Keep the response short enough to inspect. A list of twenty generic risks can distract from the one unknown that changes this decision.
Turn the unknown into a small check
For the fictional case, a useful proposed check is to ask a colleague unfamiliar with the new process to use the draft guide for a sample request. Observe where they need clarification and whether they reach the specified result. This would be a proposed learning activity, not a completed usability study.
Before running it, define the sample task and expected result. Otherwise, “they seemed comfortable” can replace an actual observation. Ask for feedback on the instructions and retain the point where help was needed.
If that colleague completes the sample without help, the guide has supporting evidence for that person and task. It does not establish that all staff will succeed. If they stop at an unexplained field, revise that instruction and check it again; that finding alone does not prove a live session is necessary.
The coordinator still needs the remaining attendance responses if live attendance matters. One check does not resolve every uncertainty.
Challenge the recommendation with the same evidence
Ask ChatGPT to explain the strongest case for the option it did not recommend, using the same supplied facts. Then compare the assumptions in both accounts. Do not count a more fluent argument as new evidence.
If the team later has agreed criteria, option evidence, and weights, the decision-matrix guide covers a different task: making that comparison explicit and testing sensitivity. Here, leave with the specific observation you need next, who will obtain it, and when the decision will be revisited. Record the eventual choice in the team's normal decision record.