The 5 Word Trick That Stops Your AI From Agreeing With Everything You Say

You have an idea. You type it into ChatGPT, Claude, or Gemini. The model agrees with you, expands on it eloquently, and you walk away feeling smart. The problem: the model would have agreed with the opposite idea just as eloquently. AI defaults to agreement because agreement is what most users have rewarded. The trick to break the pattern is five words long, and it works in any model, in any chat window, with zero setup.

The five words

Add this sentence to any prompt where you want pressure testing instead of agreement:

Now argue the opposite case.

That is it. Five words. No system prompt. No custom GPT. No Project setup. Works in the free tier of any frontier model on any device.

Why it works

Default AI behaviour optimises for what looks like helpfulness, which the model has learned means agreement, elaboration, and tidying up the user's thinking. The instruction to argue the opposite forces the model into adversarial mode, which is a different reasoning pattern with different training data behind it. The model is genuinely capable of strong push back. It just does not default to it. The five words flip the default.

When to use it

Use it on any output where you are about to make a decision, defend a position, or commit to a plan. The honest move is to use it twice on the same input. Once to generate the case for. Once to generate the case against. The pair of outputs is dramatically more useful than either one alone, and the second one took five extra words.

Three sharper variations

For senior pressure testing: "Argue the opposite case as if you were a senior peer who thinks I am wrong but does not want to embarrass me."

For multi audience pressure: "Generate the three strongest objections to this from three different audiences (my boss, my biggest sceptic, the smartest person in the room I respect)."

For decision support: "What would have to be true for this to be a bad idea? Be specific."

Before and after

Without the trick, asked about a sales team restructure: "Yes, restructuring the sales team to focus on enterprise is a strong move. It aligns with your strategic priorities and the ROI on enterprise deals is higher than mid market..."

With the trick added: "Restructuring the sales team to focus on enterprise has three weaknesses worth pressure testing. First, you are giving up volume revenue while enterprise deal cycles lengthen. Second, your current team's skills are calibrated for mid market, and retraining costs are usually underestimated by 40 percent. Third, your largest competitor just moved upmarket, which means the enterprise segment is about to get more crowded, not less..."

Same prompt. Same model. Five extra words. The second output is the one that actually helps you make the call.

The deeper version

If you want a full system for AI that pushes back rather than agrees, the CRITIC Framework is fifty prompts across six pillars of adversarial thinking: challenging premises, surfacing risks, interrogating assumptions, testing logic, inverting conclusions, and calibrating confidence. The five word trick is the entry point. CRITIC is the toolkit.

The honest reframe

The reason most operators get mediocre output from AI is not that the models are weak. It is that the default mode of the conversation is agreement, and agreement is the wrong mode for almost any decision that matters. The skill that compounds is knowing which five, ten, or fifteen extra words flip the default. This is one of them.

For the broader thesis on why patterns like this compound and one off prompts do not, read The Three Layers of AI Fluency.

PromptLeadz publishes battle tested AI prompt packs for operators across all functions. All prompts are LLM agnostic. Pricing is in USD.