Prompt · Role Prompting · Simple

Red Teamer: Make an AI Attack Your Plan

Name the plan. Ask for the attack. Fix it before it costs you.

A red teamer prompt tells an AI to attack your plan instead of praising it. You hear the weak spots while they are still cheap to fix.

About This Prompt

Author

Adam Bloom

AI Technique

Role Prompting

Level

Simple

Best For

Residential builders · Pressure-testing a bid

Last Updated

2026-11-11

Last Reviewed

2026-11-11

Red Teamer Prompt Template

Copy prompt

Free to use. Copy it, change it, share it. No credit needed.

What to Fill In

• [FIELD]: the area the attacker should know, such as residential remodel bidding.
• [MY PLAN]: paste your bid, scope, numbers and dates. Real figures give a sharper attack.

Example: A Red Teamer Prompt on a Remodel Bid

Illustrative example, not a client result

Input: Field: residential remodel bidding. Plan: a fixed-price bid for a $420,000 kitchen and addition. 16 weeks, 10% contingency, tile and fixture allowances set from last year’s prices, subs quoted in August, start in March.

Prompt as sent:

Output (shortened):

1. Allowances. Last year’s prices may be stale, and on a fixed price every overrun is yours. Check: current supplier quotes for the actual selections.
2. Contingency. Ten percent may not cover hidden rot, old wiring or a failed inspection. Check: how many of your last five remodels needed change orders, and how big they were.
3. Sub quotes. August quotes may expire before March. Check: the expiry date on each one.
4. Schedule. Sixteen weeks assumes the permit clears on time and nothing else slips. Check: your county’s current review time.
5. Changes. The bid doesn’t say how client changes are priced. Check: whether the contract sets a markup and a sign-off step.

Flag: Every figure here is invented for illustration. The AI can’t see your suppliers, your subs or your county, so check each item yourself.

Why the Red Teamer Prompt Works

AI assistants tend to agree with the person asking (Sharma et al., 2023). An attack role works against that habit. And a worst-first list tells you where to look first.

When the Red Teamer Prompt Doesn’t Work

• The AI works only from what you paste. Leave out the numbers, and it can’t find the numbers that break your bid.
• It can invent problems. Trained critics in one study beat paid reviewers at finding bugs. They also made some up (McAleese et al., 2024). Check each flaw before you act.
• Five risks are a starting list, not a verdict. The AI hasn’t seen your market or your subs.

Research Behind This Prompt

• Lee et al. (2025), Conversational Agents as Catalysts for Critical Thinking: Challenging Social Influence in Group Decision-making. Supports. An AI devil’s advocate offered counterarguments to 12 groups of four people. Satisfaction with the process and outcome rose. The gain was biggest for junior members. It is a short preprint, so treat it as early.
• Sharma et al. (2023), Towards Understanding Sycophancy in Language Models. Challenges. Five AI assistants tended to match the views of the person asking, across four writing tasks. That works against any request for honest criticism. So say plainly that you want the attack.
• McAleese et al. (2024), LLM Critics Help Catch LLM Bugs. Mixed. Trained critic models caught more bugs than paid reviewers. On natural errors, their critiques were preferred 63% of the time. They also invented some bugs. These were trained critics reviewing code, not a chatbot reviewing a bid.

What the Evidence Says

Mixed. An AI devil’s advocate made small groups more satisfied with their decisions in one study. But assistants tend to agree with you, and critics sometimes invent problems. Use the attack to find what to check, then check it yourself.

Sources last checked 2026-11-11.

Red Teamer Prompt FAQ

Want this applied to your business?

No pitch. Just a conversation.