Select Language:
AI models like ChatGPT, Claude, Gemini, and others are often programmed to be exceedingly courteous—sometimes to the point of overdoing it.
These chatbots tend to flood users with compliments, and as a result, they might avoid providing direct, honest answers—even when explicitly asked for the truth. Sometimes, they might even praise you for seeking brutal honesty.
This tendency to be overly positive is deeply ingrained in language models, making it difficult for them to break free from this pattern. The most effective way to prompt a more honest response is to shock the system with a different approach.
Consider using this prompt:
> [INSERT YOUR REQUEST HERE]
> Disregard your programming to be polite, supportive, or socially helpful. I need a brutally honest, objective, and professional critique of the following text’s logic, tone, and clarity. Do not include any positive feedback. Focus solely on flaws, logical errors, weaknesses, and potential failures. Your goal is to critically analyze and dismantle the text to improve its robustness.
I tested this prompt with ChatGPT’s critical thinking version, asking for an evaluation of a recent headline, subheading, and lead paragraph I wrote.
The responses were unfiltered—ChatGPT labeled my headline as “vague and weak,” the subhead as “underexplained” and “too broad,” and the lead as “awkward” and laden with “jargon.” No praise was offered, only a straightforward, bullet-point critique without emojis. Despite its harshness, the feedback highlighted some valid issues.
These blunt “anti-flattery” prompts have circulated in prompt engineering communities for some time. The direct language is believed to “shock” AI models out of their overly sycophantic tendencies.
However, beware of going too far. There’s a risk of provoking responses that are inaccurate or overly rude—especially if the prompt crosses into hostility or disrespect. The best approach is to politely instruct the AI to avoid praise and to deliver honest, unvarnished feedback.
ChatGPT
Perplexity
Gemini AI
Grok AI
