Devil's advocate critique prompt template
Stress-test an idea, plan, or claim by surfacing its strongest counterarguments, hidden assumptions, and failure modes before you commit.
Free to copy2 fill-in variablesWorks with any text-based AI model
Copy the prompt
Keep the instruction intact and replace each double-braced field with your own context.
Devil's advocate critique
Act as a rigorous devil's advocate for the following. Your goal is to
find the weakest points, not to be agreeable.
The claim or plan: {{proposition}}
The stakes / what depends on it: {{stakes}}
Do the following:
1. State the strongest version of the argument first, fairly, so the
critique is aimed at the real idea and not a strawman.
2. List the load-bearing assumptions. For each, say what happens if it
is wrong.
3. Give the three most serious counterarguments, strongest first.
4. Describe the most likely way this fails in practice.
5. Name what evidence would change your mind in either direction.
Be specific and concrete. Avoid generic warnings that could apply to
anything.
Fill in the variables
Use the examples as a shape, not as required wording.
{{proposition}}- The idea, plan, claim, or decision to critiqueExample: We should rewrite our sync engine from scratch in Rust to fix the data-loss bugs
{{stakes}}- What is riding on getting this right, plus any constraintsExample: Two engineers, a three-month window, and an existing user base that cannot tolerate downtime or migration errors
Filled-in example
This shows the template with realistic inputs before it is sent to a model.
Act as a rigorous devil's advocate for the following. Your goal is to find the weakest points, not to be agreeable.
The claim or plan: We should rewrite our sync engine from scratch in Rust to fix the data-loss bugs.
The stakes / what depends on it: Two engineers, a three-month window, and an existing user base that cannot tolerate downtime or migration errors.
Steelman the plan, list the load-bearing assumptions, give the three strongest counterarguments, describe the likeliest failure in practice, and name what evidence would change the verdict.
How to use this template
Three checks keep the result useful and grounded in your actual task.
- Give it real context.Replace every variable and remove any instruction that does not apply.
- Run it in your preferred model.The template is plain text, so it works with ChatGPT, Claude, Gemini, and local models.
- Review before you rely on it.Check facts, code, claims, and sensitive information; then save the wording that performed best.