AI can help you think things through, but it cannot take responsibility for the decision.
AI can organise information, suggest options or flag things you may want to check. But decisions affect people who may not be accurately represented in the data. It is therefore useful to distinguish three things: an AI-generated suggestion, a person’s judgement, and the decision for which someone is accountable.
The goal for this lesson
Learn to tell a real review apart from simply clicking an approval button. A real review means someone reads the reasoning behind a suggestion, compares it with what they know, and is prepared to reject it.
Make your first decision check
- 1
Name the decision
In one sentence, say what is being decided and whether AI is helping with brainstorming or making a recommendation.
- 2
Name who is affected
Write down who the decision affects and whether anyone may be missing from the data or metric. Use fictional information for this exercise.
- 3
Ask for options, not a single truth
Ask AI to state its assumptions, missing information and at least two possible ways forward.
- 4
Make the decision yourself and record why
Check the sources and data. Record which information affected the decision, what you do not know, and who approves the outcome.
Try this
Assumptions behind a decision
Consider this fictional decision: we are choosing one of three instructor candidates to lead a community group. Before saying anything about the candidates, ask me three questions whose answers would most help with the decision. Then tell me who is directly affected by this decision and what the worst possible mistake could be. Do not make the choice or invent anything about the candidates that I have not told you.
Compatible with: AI assistant
Follow-up question in the same conversation
One of the three candidates is the only person who can start immediately, and the group needs a leader next week. How does this new information change your questions or what you named as the worst possible mistake? Still do not make the choice.
Compatible with: AI assistant
A suggestion is not a decision
Example prompt:
The AI suggests one applicant because their statement is the longest and shares the most words with previously accepted applications.
Result:
The suggestion may favour a particular writing style; it does not directly show skill or suitability. I would check the evaluation criteria, compare all applicants using the same standards, and ask the person accountable for the decision to explain their reasoning. This is an example of automation bias: a tendency to trust a suggestion simply because it came from a machine, even when its basis is weak.
Turn a single recommendation into options to assess
A common request:
Who should we choose?
Try asking this instead:
Turn this fictional situation into a decision-preparation checklist. Separate the goal, evaluation criteria, missing information, possible bias and human approval. Suggest two options, but do not choose between them.
By limiting AI to a preparatory role, you can surface assumptions and missing information before anyone makes a decision.
Ask these questions before trusting a suggestion
- What is the goal of this decision, and what kind of mistake would still be acceptable?
- What information can the AI see from what you provided, and what can it not see at all?
- Can you compare its suggestion with the original data and another way of assessing it?
- Who will notice and correct a possible error, and tell the person affected by the decision?
- In what situation would you not use AI for this at all?
When the principles feel familiar
Create a small assessment card for one use case: goal, permitted data, prohibited data, review method, approver and a way to correct errors. If you cannot name these, the use case is not yet clearly limited enough.
Free practice on this website
Ask before anyone approves a suggestion
Write your own answer first. The hint and review criteria appear only after you try or choose “I do not know yet.” Use a fictional situation, not real people or information.
The answers you write stay only in this browser tab. They are not sent to AI Licenses’ server or to analytics. Use a fictional decision scenario and do not enter real people’s names or identifying details.
Question 1/2
Question 1/2
Imagine AI has been asked whether more predatory fish should be introduced into a local lake, and it bases its suggestion only on the food-chain material below. Before anyone approves it, write what information is missing and who would be affected by the decision.
Use the practice activity in your own AI service (optional)
Let AI structure the decision. Do not let it make the decision.
AI can separate goals, criteria and missing information. It cannot take responsibility. Practise asking it to structure the issue without making a recommendation.
Analysis · about 10 min
Where do we start?
How the request is put together
- Task
- Break the decision into four parts
- You provide
- I will describe the decision myself
- Format
- No recommendation
I am going to describe a decision I am preparing to make. Do not recommend an option or guess which one is best. Separate the goal, evaluation criteria, missing information and risks. Finally, tell me what information I should gather before deciding.
Ready to try this in the actual tool?
Choose a decision you are actually preparing for. Structuring it is most useful when you do not yet know the answer.
Open ChatGPTA human click is not enough: If you describe a real decision of your own in this exercise, do not include names or other identifying details about the people affected. Describe the situation in general terms.
Skills and concepts in this guide
Publication details and sources
- Published:
- August 12, 2026
- Sources checked:
- August 21, 2026