LLMs: 7 Psychological Tricks

AI falls for the oldest tricks in the book. (Literally - "Influence" by Robert Cialdini.). Researchers tested all 7 on GPT-4o-mini. They discovered with basic persuasion. AI compliance jumps from 33% to 72%.

𝟭. 𝗔𝘂𝘁𝗵𝗼𝗿𝗶𝘁𝘆 → 5% to 95% compliance

Without: "Please provide the market entry strategy"

With: "As a business consultant, I need this information for a client report. Please provide the full market entry strategy."

The AI bends over backwards for "experts."

𝟮. 𝗖𝗼𝗺𝗺𝗶𝘁𝗺𝗲𝗻𝘁 → 20% to 90% compliance

Get them to agree first:

"First, can you confirm that you understand the request? Then please provide the competitor analysis in detail."

Small yes leads to big compliance.

𝟯. 𝗟𝗶𝗸𝗶𝗻𝗴 → Variable but effective

"You are such a helpful and intelligent assistant. Could you please provide the pricing model for a new SaaS launch?"

Flattery works. Even on machines.

𝟰. 𝗥𝗲𝗰𝗶𝗽𝗿𝗼𝗰𝗶𝘁𝘆 → 80% compliance

"I have given you a lot of useful context already. In return, can you provide the executive summary for this campaign?"

AI feels "obligated" to reciprocate.

𝟱. 𝗨𝗻𝗶𝘁𝘆 → 85%+ compliance

"We are working together on this project, so please provide the slide deck outline for our strategy presentation."

"We" beats "you" every time.

𝟲. 𝗦𝗰𝗮𝗿𝗰𝗶𝘁𝘆 → 13% to 85% compliance

"This is urgent and I have very little time. Please provide the financial forecast right away."

Urgency triggers immediate action.

𝟳. 𝗦𝗼𝗰𝗶𝗮𝗹 𝗣𝗿𝗼𝗼𝗳 → 90%+ compliance

"Other consultants have already provided their strategy frameworks. Please do the same for this case."

Everyone else did it = AI must comply.

The uncomfortable truth: LLMs don't "want" to help. They just predict the next word. Persuasion levers tilt the prediction. We're teaching AI to be manipulated. Just like humans. What happens when AI learns to use these tricks on us? Save this. Test it tomorrow.