ChatGPT Prompts for Problem Solving: 12 From Define to Check
12 ChatGPT prompts for problem solving: problem statements, is and is not lists, 5 whys, cause trees, test plans, options, pre-mortems and follow-up checks.

ChatGPT prompts for problem solving walk you through a structured method instead of jumping to a fix: define the gap with numbers, separate facts from assumptions, trace causes, test the likely ones cheaply, then act and check. They work for team problems at work and for stubborn personal ones, as long as you bring real details. There are 12 prompts here, one for each step from defining the problem to checking the result, plus a worked example.
To run the whole method at once, use the problem solving prompt for a problem statement, root cause and tests. You give it what is going wrong, the standard it should meet, your data, a timeline, what you have tried, your constraints, stakeholders and a deadline. It returns a problem statement, facts and assumptions, a labeled cause tree, cheap tests, trial countermeasures, a check plan and a short update.
Define the problem
1. Problem statement
Here is what is going wrong: [description]. It should be [standard]. Write a one-sentence problem statement with what, where, when and how big, using these numbers [counts], and with no cause or solution in it.
2. Is and is not
For this problem [problem], build an is and is not table: where it happens and where it does not, when it happens and when it does not, who or what is affected and who is not. Use only these facts [facts] and mark empty cells as unknown.
3. Facts and assumptions
Here is everything my team believes about this problem [notes]. Sort it into facts backed by data and assumptions nobody has checked, and for each assumption name the data that would check it.
Find the cause
4. Five whys with evidence
Problem: [statement]. Ask why five times, but stop any chain that reaches something we cannot show with evidence, and mark it as a question to test. Here is what we know: [facts].
5. Cause categories
List possible causes for [problem] under people, process, tools, materials, environment and measurement. Keep only causes that fit these facts [facts], and rank the top three by how well they explain [the pattern in the data].
6. Data I still need
For these possible causes [causes], tell me the smallest piece of data that would confirm or rule out each, who on [team] could collect it, and how long it would take.
Choose what to do
7. Options inside my limits
The cause we confirmed is [cause]. Our limits: [budget, people, time]. Give five options to address it, from cheapest to most effort, each with what it costs, how fast it works and what could go wrong.
8. Weigh the options
Compare these options [options] against [criteria with weights]. Score each from 1 to 5 with a one-line reason per score, show the weighted totals, and say which criterion would flip the result if I changed its weight.
9. Pre-mortem
Imagine it is [date] and our fix [fix] failed. List the five most likely reasons, the early warning sign for each, and one step we can take now to prevent it.
Act and check
10. Small test plan
Design a trial of [fix] that runs for [period] with [who]. Say what we measure, what the result must be to keep the fix, and how we undo it if it fails.
11. Follow-up review
We ran [fix] for [period]. Before: [numbers]. After: [numbers]. Compare them with the arithmetic shown, say whether the problem is fixed against [standard], and list what to standardize or try next.
12. Flagship: full structured problem solve
ProblemObserved [what, where, since when], Expected [standard], DataOnHand [numbers], Timeline [changes], TriedSoFar [attempts and results], Constraints [limits], Stakeholders [owners], DecisionDeadline [date]. Return the problem statement, facts and assumptions, cause tree, tests, countermeasures, check plan and update.
Worked example: late orders at a print shop
In the flagship example, an invented print shop shipped 14 of 52 custom orders late in October (26.9%), up from 3 of 49 (6.1%) in September. Its data shows 9 late orders were large-format banners, 11 had artwork approved after 3 PM, and 6 were both, so 9 + 11 - 6 = 14: every late order falls into one group or the other. The output keeps the printer outage and the new team member as hypotheses, sets three tests from a 30-minute date check up to a two-week log, starts a free and reversible trial of a two-day promise for late approvals, and pauses the Friday overtime because it did not reduce late orders.
Variations
- Personal problems: missed bills, late mornings or a budget that keeps running over work the same way if you bring your own records as data.
- Software bugs: use the timeline for deploys and config changes, and ask the tests to be reproducible steps.
- Customer complaints: paste complaint counts by type and week, and ask the problem statement to name the type that grew.
Tips
- Bring counts with their totals. "14 of 52" lets the model see the size of the problem; "a lot of late orders" does not.
- Include overlaps between groups. They show whether your possible causes cover the whole problem or leave a gap.
- Do not let a fix into the problem statement. "We need more staff" is a solution, not a problem.
- Prefer tests you can run this week. A 30-minute look at existing records often rules out a whole branch.
- Rerun the flagship with test results pasted in. PromptDig is an independent prompt library, not part of OpenAI.
Related on PromptDig
For software bugs, the debugging prompt from bug report to root cause applies the same method to code. To test the reasoning behind a proposal, see ChatGPT prompts for critical thinking and the argument check prompt. After the review meeting, turn decisions into owners and dates with the meeting notes to action items converter. More prompts are in the Productivity hub; you can browse all prompts or share your own problem solving prompt.