Before you pitch an idea or present a proposal, deliberately ask an AI to punch holes in it—to find the weakest arguments, the missing assumptions, the ways it could fail. Catching your own vulnerabilities before a skeptical audience does forces you to either strengthen your case or abandon a flawed direction early.
Adversarial prompting is the practice of intentionally trying to break or expose weaknesses in AI output by challenging it, asking it to argue against itself, or feeding it confusing or contradictory instructions. Instead of accepting AI output at face value, you stress-test it. For freelancers, this means asking an AI to poke holes in a proposal before the client does.
It's a form of internal quality assurance using the AI itself as a critical reviewer.
When you ask an AI to generate a proposal, it optimizes for coherence and persuasion within the constraints you've given. But it rarely identifies logical gaps, unsupported claims, or weak points in positioning. A human review catches some issues, but a different AI perspective—especially one instructed to be critical—catches others.
Adversarial prompting surfaces weaknesses before clients see them, giving you a chance to strengthen the proposal or prepare defensible responses.
The skeptic test: After generating a proposal, use the same AI (or a different model) with this prompt: "A skeptical prospect reads this proposal. What are the three weakest points? Where would they push back most?" The AI identifies vulnerabilities that aren't obvious to you.
The competitor angle: "A competitor offering similar services reads this proposal. What would they criticize about our positioning or value claims?" The AI plays competitor and highlights where you're vulnerable to competitive undermining.
The budget question: "If this prospect has limited budget, which claims in this proposal are hardest to justify? Which feel like scope creep?" This surfaces arguments the prospect might use to negotiate down your price or trim scope.
The assumption audit: "What underlying assumptions does this proposal make? Which assumptions might this prospect reject?" Proposals always rest on assumptions (e.g., that budget exists, that timeline is realistic, that the prospect's stated problem is the real problem). Naming them lets you address them proactively.
The timeline stress test: "If the prospect wants to accelerate this timeline by 30%, what breaks? What becomes impossible?" This preps you for scope-creep negotiations and helps you articulate realistic constraints before the client pressures you.
Step 1: Generate your proposal normally. Step 2: Copy the proposal. Step 3: Create a new conversation. Step 4: Paste the proposal and use an adversarial prompt. Step 5: Review the AI's critiques. Step 6: Revise the proposal based on legitimate vulnerabilities. Step 7: Send revised version.
This adds 20 minutes to proposal writing but prevents sending weak work and prepares you for client objections.
One risk: adversarial testing can lead to bloated, overly defensive proposals where you address every possible objection. This backfires—it reads as uncertain. The goal isn't to defend against every critique but to identify which vulnerabilities are real enough to address and which are acceptable (every proposal has some weakness).
Use adversarial feedback to prioritize: fix vulnerabilities in core value positioning and timeline feasibility; accept minor vulnerabilities in approach or methodology that don't undermine the core ask.
An advanced technique: generate two competing proposal approaches (e.g., one emphasizing efficiency, one emphasizing transformation). Ask the AI: "Which approach is more persuasive to a client who is risk-averse? Which to an innovation-focused client?" The AI comparison reveals which proposal works for which prospect profile, helping you personalize.
Try this: Generate a proposal for a prospect you're actually pursuing. Copy it into a new AI conversation. Use this prompt: "I just sent this proposal to a prospect. Imagine you're the prospect's skeptical CFO. What are your three biggest objections to this proposal? Where do you see overpricing, scope ambiguity, or unsupported claims?" Read the AI's objections. Ask yourself: "Are these fair?" If yes, revise the proposal to address them. If no, prepare your response for when the actual CFO raises the same point. Send either the revised proposal or the original, armed with counter-arguments.
Peri can explain this concept, give practical examples, help you decide whether it applies to your situation, or recommend a journey if appropriate.
Explore related journeys or tell Peri what you're working through.