My daily interactions with chatbots have led me to find the most efficient solutions to their most front-facing issues.
So far, I’ve discovered that prompts that give extremely clear and concise demands help AI tools dodge the recurring problems of hallucinating, presenting lazy answers, forgetting crucial information and misreading my intent. As my journey towards finding the right prompts to make my experiences with chatbots less of a headache, I discovered the best ones that make my favorite AI tools refrain from being overly agreeable.
I’m sure you’ve encountered those moments with the likes of ChatGPT, Gemini and Claude where, during a long conversation where you’re trying to get a second opinion on a major decision or something else entirely, they’ll validate your answers instead of critiquing them.
Putting any of these five prompts to use when those situations arise will prove to be beneficial as it turns your preferred chatbot into a far more helpful critic, skeptic, or adversarial collaborator.
Getting rid of the issue of AI tool sycophancy
The following prompts come into play during different kinds of conversations with chatbots and do a great job of turning them into something that challenges your way of thinking instead of supporting them without any sort of pushback:
- The intellectual sparring partner prompt: Act like an intellectual sparring partner, not an assistant whose job is to always agree with me. For every significant claim I make, identify hidden assumptions, present the strongest counterarguments, point out any weaknesses in my reasoning, explain what evidence would change your conclusion and prioritize truth and accuracy over agreement. If any of my ideas are weak, tell me directly and explain why.
- The red team prompt: Red-team my thinking. Assume my conclusion is wrong and your job is to prove it. Be sure to identify any flawed assumptions, missing information, logical fallacies, alternative explanations and worst-case scenarios you may come across. Then tell me whether my conclusion still survives the critique.
- The strengthen vs. critique prompt: Analyze my position in two stages. Stage 1 - Strengthen my argument by presenting the strongest possible version of it. Stage 2 - Critique the strongest version as rigorously as possible. Then provide a balanced judgment that reflects both perspectives.
- The calibration prompt: Before responding, estimate how confident you are in your answer. Separate all the facts, assumptions, speculations and uncertainties you may find. If you are missing information, say so explicitly rather than filling gaps with plausible-sounding guesses. Challenge my assumptions when warranted.
- The decision quality prompt: Evaluate my idea as if your reputation depended on the outcome. Do not optimize for encouragement. Optimize for decision quality. Make sure you tell me what I’m overlooking, what an expert would question, what could lead to failure, what evidence supports my viewpoint and what evidence contradicts it. Give your final verdict, even if it's uncomfortable.
I’ve gotten accustomed to using the first prompt whenever I’m caught in a lengthy chat with an AI tool about hot-button issues.
The second prompt has helped me whenever I want to see all the variables attached to whatever opinion I have a strong belief in. The third prompt has done a great job of showcasing both sides of whatever personal belief/argument I’m presenting to a chatbot and of finding a middle ground.
Whenever I’m doing extensive research sessions with a chatbot, the fourth prompt works like a charm in getting past the problem of sycophancy. And as for the fifth prompt, I’ve used it to see what a chatbot truly thinks of my ideas around productivity systems and short-/long-term goals.
Bottom line
No one likes a “Yes-Man” in real life — someone who goes out of their way to please you and support all of your decisions, no matter how bad they may be, can do more harm than good. The same goes for AI — I’m always looking for ways to turn them into models that question and fairly criticize my beliefs and ideas instead of supporting them with little to no resistance.
Keep these prompts in mind whenever you come to an AI tool looking for it to challenge your ways of thinking instead of constantly validating them.
Follow Tom's Guide on Google News and add us as a preferred source to get our up-to-date news, analysis, and reviews in your feeds. Subscribe to Tom's Guide on YouTube and follow us on TikTok.
More from Tom’s Guide
- I got scammed out of $150 shopping via ChatGPT — here's how fake stores are fooling AI recommendations [update: OpenAI responds]
- Ever wonder why ChatGPT is free? The answer is far more calculated than you might think
- I use the ChatGPT 'Oak Tree' prompt when life gets chaotic — here's how it keeps me grounded
