Tool

Simulating Debate

Ask AI for advice and it tends to agree with you. Make it argue both sides against each other and the case against your plan finally gets a fair hearing, before you commit.

1. What this does for you

Most of the time AI mirrors what you seem to want, so you get a confident yes to a question you have already half-answered. This prompt makes AI play advocate, skeptic, rebuttal, and judge, all in one session. You see the strongest case for, the strongest case against, and an honest verdict with the conditions that would change it. It is the cheapest way to hear the objection before a board member, a funder, or reality raises it for you.

2. What to type into your AI

Copy the box below. Replace the words in [BRACKETS] with your own info. Then paste it into Claude or ChatGPT.

I want you to run a structured debate to pressure-test a decision. Play every role yourself, and be genuinely fair to each side.

THE DECISION OR CLAIM:
"""
[STATE IT IN ONE SENTENCE, e.g. "We should launch the new program in the fall."]
"""

CONTEXT:
[PASTE WHAT MATTERS: constraints, stakes, who is affected, what we know]

Run it in four rounds, labeled:
1. ADVOCATE: the strongest possible case FOR.
2. SKEPTIC: the strongest possible case AGAINST.
3. REBUTTAL: each side answers the other's single best point.
4. JUDGE: step out of both roles, name which case is stronger and why, and list the two facts that, if they changed, would flip your verdict.

Do not let the side I seem to favor win by default. If the case against is stronger, say so plainly.

Try it now in Claude →

See a full worked example ▾

Hire a second staffer or contract the work out?

Run a structured debate to pressure-test a decision. Play every role yourself and be genuinely fair to each side.

THE DECISION:
"""
We should hire a second full-time staffer this quarter instead of contracting the work out.
"""

CONTEXT:
- We have budget for one year, not guaranteed beyond that.
- The work is steady now but seasonal swings are real.
- Hiring takes about 3 months; a contractor could start in 3 weeks.
- The team has asked for a teammate, not a vendor.

Run four rounds: 1. ADVOCATE (best case for), 2. SKEPTIC (best case against), 3. REBUTTAL, 4. JUDGE (which is stronger, and the two facts that would flip the verdict).

Do not let the side I seem to favor win by default.

Adopting a new K-5 phonics curriculum district-wide next fall

Run a structured debate to pressure-test a decision. Play every role yourself and be genuinely fair to each side.

THE DECISION:
"""
Should we roll out the new K-5 phonics curriculum across all six elementary schools next fall?
"""

CONTEXT:
- 40 teachers, about 900 students, 85,000 dollars for materials plus 3 PD days
- District reading scores have been flat for 3 years
- 12 teachers piloted it last spring, feedback was mixed
- Board wants a yes or no vote by March 15

Run four rounds: 1. ADVOCATE (best case for), 2. SKEPTIC (best case against), 3. REBUTTAL, 4. JUDGE (which is stronger, and the two facts that would flip the verdict).

Do not let the side I seem to favor win by default.

Raising residential trash pickup fees to fund two new routes

Run a structured debate to pressure-test a decision. Play every role yourself and be genuinely fair to each side.

THE DECISION:
"""
Should the city raise the monthly residential trash fee from 18 dollars to 24 dollars to fund two new collection routes?
"""

CONTEXT:
- Covers 14,000 households, adds 1.0 million dollars per year in revenue
- Missed-pickup complaints are up 40 percent over two years
- Last fee increase was 6 years ago, no public hearing yet
- Council votes in 5 weeks, one member is up for re-election

Run four rounds: 1. ADVOCATE (best case for), 2. SKEPTIC (best case against), 3. REBUTTAL, 4. JUDGE (which is stronger, and the two facts that would flip the verdict).

Do not let the side I seem to favor win by default.

Self-perform the concrete or sub it out on a 6-month build

Run a structured debate to pressure-test a decision. Play every role yourself and be genuinely fair to each side.

THE DECISION:
"""
Should we self-perform the cast-in-place concrete on this job instead of subbing it to our usual concrete Sub?
"""

CONTEXT:
- Our number self-performing is 1.15 million dollars, the Sub bid 1.32 million dollars
- We own the formwork but our best concrete foreman is booked until week 8
- The GC schedule needs foundations poured by week 6
- Owner has 50,000 dollars per week liquidated damages for late completion

Run four rounds: 1. ADVOCATE (best case for), 2. SKEPTIC (best case against), 3. REBUTTAL, 4. JUDGE (which is stronger, and the two facts that would flip the verdict).

Do not let the side I seem to favor win by default.

Opening a second coffee shop location across town

Run a structured debate to pressure-test a decision. Play every role yourself and be genuinely fair to each side.

THE DECISION:
"""
Should we open a second location on the east side this year?
"""

CONTEXT:
- Shop one clears about 9,000 dollars profit a month, fairly steady
- New lease is 4,200 dollars a month, plus roughly 60,000 dollars to build out
- I would need to hire and train a manager since I cannot be in two places
- Savings would cover about 5 months of the new rent if it starts slow

Run four rounds: 1. ADVOCATE (best case for), 2. SKEPTIC (best case against), 3. REBUTTAL, 4. JUDGE (which is stronger, and the two facts that would flip the verdict).

Do not let the side I seem to favor win by default.

Accepting a 200,000 dollar restricted gift for a brand-new program

Run a structured debate to pressure-test a decision. Play every role yourself and be genuinely fair to each side.

THE DECISION:
"""
Should we accept the 200,000 dollar restricted grant to launch a new after-school program?
"""

CONTEXT:
- Grant funds direct program costs only, no overhead or admin
- We would need to hire 2 staff and rent space we do not have
- Current programs are running at capacity with the team we have
- The grant is one year, with no promise of renewal

Run four rounds: 1. ADVOCATE (best case for), 2. SKEPTIC (best case against), 3. REBUTTAL, 4. JUDGE (which is stronger, and the two facts that would flip the verdict).

Do not let the side I seem to favor win by default.

Adding a monthly advisory service line to the firm

Run a structured debate to pressure-test a decision. Play every role yourself and be genuinely fair to each side.

THE DECISION:
"""
Should we add a monthly CFO-advisory service line for our small-business clients?
"""

CONTEXT:
- Roughly 30 clients might pay 800 to 1,500 dollars a month for it
- It pulls me off compliance work during the busy season
- I would likely need one more staff hire to protect existing deadlines
- This is general guidance, not filed advice, and pricing is my own estimate

Run four rounds: 1. ADVOCATE (best case for), 2. SKEPTIC (best case against), 3. REBUTTAL, 4. JUDGE (which is stronger, and the two facts that would flip the verdict).

Do not let the side I seem to favor win by default.

Taking a low-pay mural job for a big-brand audience

Run a structured debate to pressure-test a decision. Play every role yourself and be genuinely fair to each side.

THE DECISION:
"""
Should I take the 1,200 dollar mural job for a well-known brand because of the exposure?
"""

CONTEXT:
- My normal rate for this size wall is about 4,000 dollars
- They have 400,000 followers and promise to tag me
- It would take 2 weeks, blocking two paying commissions
- They want to own the full rights to the image afterward

Run four rounds: 1. ADVOCATE (best case for), 2. SKEPTIC (best case against), 3. REBUTTAL, 4. JUDGE (which is stronger, and the two facts that would flip the verdict).

Do not let the side I seem to favor win by default.

3. What to do with the answer

State the decision as one clear sentence. A fuzzy claim produces a fuzzy debate.
Paste the prompt into Claude or Gemini and add the context that actually matters.
Read the SKEPTIC round most carefully. That is the argument you came in wanting to skip.
Look at the two facts the JUDGE says would flip the verdict. Those are what you should go verify.
If the debate surfaced a point you had not considered, run your own draft or plan through it again with that point in mind.

Why this works

A single answer carries a single point of view, usually the one it thinks you hold. Forcing a debate makes AI build the opposing case as hard as it can, which surfaces the blind spots agreement hides. You still make the call. You just make it having heard the other side.

Section Recap

Core Principle

AI agrees with you too easily. A structured debate forces the strongest case against your plan into the open, where you can weigh it.

Best-Fit Example

A leader is about to greenlight a program launch and everyone in the room already nodded along. Before committing the budget she runs the decision through a debate prompt. The skeptic round names a staffing gap nobody raised; the judge flags it as the fact that flips the call. She delays two weeks, fixes the gap, and launches on solid ground.

A principal who already loves the new curriculum runs this before the board meeting. The SKEPTIC round says the plain thing she was skipping: the 12 pilot teachers had a full year of coaching, and the other 28 will get 3 days, so the fall rollout is not the same experiment that worked. The JUDGE tells her exactly what to verify before March 15: pull the pilot scores split by trained versus untrained classrooms, and confirm whether the budget includes ongoing coaching or just books. She walks in able to answer the hard question instead of getting blindsided by it. AI is the intern here, she stays the editor who makes the call.

A public works administrator leaning toward the increase runs this to stay neutral before the hearing. The SKEPTIC round names the risk she wanted to skip: a flat six-dollar hike lands hardest on fixed-income and low-usage households, and she has no data on how many residents that is. The JUDGE hands her a verify list: pull the count of households on the senior or hardship discount, and confirm whether two routes actually clear the complaint backlog or just slow its growth. She brings both facts to council so the decision holds up to public questions instead of falling apart at the podium. AI drafts the cases, she keeps the judgment.

An estimator eyeing the 170,000 dollar savings from self-performing runs this before he commits the number. The SKEPTIC round says what he was talking himself out of: his A-crew foreman is not free until week 8, the pour is due week 6, and one slipped week at 50,000 dollars liquidated damages eats most of the savings. The JUDGE gives him two things to verify: confirm the exact date the foreman frees up, and get the Sub to hold their 1.32 million dollar price in writing for three weeks. Now he negotiates with real leverage instead of betting the schedule on a crew he does not have yet. The AI ran the argument, he still owns the bid.

An owner who has already picked out the second space runs this before signing the lease. The SKEPTIC round says the part she kept waving off: shop one runs well because she is in it every day, and location two only works if a manager she has not hired yet is nearly as good as her. The JUDGE tells her what to nail down first: figure out the true monthly cost of a manager good enough to run it without her, and check whether 5 months of runway is enough given how slowly shop one ramped in year one. She goes in clear-eyed instead of finding out in month four. The AI argued both sides, she makes the call.

A development lead who wants to say yes to 200,000 dollars runs this before the board call. The SKEPTIC round names the honest gap: the grant pays for the program but not the overhead to run it, and if it does not renew after year one, the org owns two salaries and a lease with no funding. The JUDGE gives a verify list: calculate the true unfunded overhead this program adds, and get a straight read on the odds of a year-two source. She brings the real cost of yes to the board instead of the exciting headline number. The AI pressure-tested it, she keeps the judgment about what the mission can carry.

A firm owner sold on recurring advisory revenue runs this before he pitches clients. The SKEPTIC round says the thing he kept sliding past: advisory demand peaks in the same months as tax deadlines, so without a hire he either misses filings or delivers thin advice. The JUDGE tells him what to verify: survey 5 real clients on what they would actually pay, and model whether the new revenue covers a hire before it covers his time. Treat the output as general guidance and not filed advice, but it stops him from launching a service that quietly wrecks compliance season. The AI ran the debate, he stays the one signing off.

A muralist half-decided to take the exposure job runs this before she replies. The SKEPTIC round says what the excitement was hiding: she is working at a third of her rate, giving up two paying commissions, and handing over full rights, so a brand can reuse her art forever for a one-time tag. The JUDGE gives her two things to check: ask past clients whether a brand tag ever actually turned into paid work, and price what full rights are worth versus a limited license. She answers from a place of worth instead of hope, and can counter with a license and a fair fee. The AI argued both sides, she stays the editor of her own career.

Key Takeaways

Ask for advice and AI tends to agree; ask for a debate and it argues honestly.
The skeptic round is the point of the exercise, not the advocate round.
The judge's 'what would change my mind' list tells you exactly what to verify.
Use it before any decision a stakeholder could later poke a hole in.

← All Tools