AI chatbots are fast and helpful — but also dangerously confident. If you’ve ever been burned by a response that sounded correct but turned out to be wrong, this prompt is for you.
After testing AI tools like ChatGPT, Claude and Gemini daily, I’ve learned one thing: the biggest risk you can make is trusting AI too much. From unhinged recipes to definitions of fake idioms, AI will make up stuff without blinking.
That’s why I use one a trick that I call the “glitch” prompt — and it’s completely changed how I trust AI responses. It only takes a few seconds to run, but it forces any AI assistant to pause, double-check itself, and rewrite the answer more carefully — often surfacing mistakes it didn’t even realize it made. Here’s how the glitch prompt works, why it’s so effective, and when you should use it.
What is the 'glitch prompt?'
The glitch prompt is a self-audit command. It flips your AI assistant into correction mode instead of completion mode.
Instead of blindly accepting the chatbot’s first answer, you can use it to force a quick self-check:
The full glitch prompt:
Pause — I think there’s a glitch. Check your last answer for mistakes, missing steps, false assumptions, or made-up details. Then rewrite the answer more accurately, and add a confidence rating (1–10).
Short version: Re-check and rewrite for accuracy. Add confidence rating (1–10).
Why it works — and why it feels like an impressive cheat code
Most chatbots are designed to be fast, polished and helpful — not cautious.
That means when the question is complex, ambiguous or missing context, AI will often fill in the blanks with confident guesses based on patterns rather than admit "I don't know." The glitch prompt breaks that cycle immediately.
It works because it makes the AI:
- Evaluate its own answer instead of generating a new one
- Look for holes in logic or assumptions
- Admit uncertainty (the confidence rating helps)
- Improve clarity and structure automatically
It's essentially a mindset shift for the AI. You’re telling the model: Don’t simply perform the task, check your work.
When to use the 'glitch prompt'
You do not need to use this prompt every time you use ChatGPT. And, by the way, it works well with other chatbots too.
Here's when it's become extremely useful for me:
- When the answer sounds too confident. If the response feels too perfect, I treat that as a yellow flag. The glitch prompt is my way of saying, “Run that back, just in case.”
- When I’m about to spend money. AI can miss better deals, ignore key features or misunderstand your needs. I glitch-check every major recommendation before I buy.
- When troubleshooting tech. Some AI fixes only work on different devices or skip basic diagnostic steps. The glitch prompt helps catch these slipups.
- When the stakes feel higher. Whether I’m planning travel, writing something important, or job searching — glitch-checking is how I reduce risk.
Upgraded 'glitch prompts' that I actually use
Once you’re comfortable with the basic version, you can tailor it to the task.
-
Shopping: Pause — glitch check. You may be missing better options.
Re-do your recommendation with 3 alternatives, include tradeoffs and tell me what you’re least confident about. - Facts and stats: Glitch check: List every factual claim and label it Verified/ Unverified / Assumption. Then rewrite the answer using only verified info and clearly labeled assumptions.
- Troubleshooting: Glitch check: You might be solving the wrong problem. Ask me 5 diagnostic questions first, then suggest a fix.
- Writing and editing. Glitch check this draft: Find weak logic, vague phrasing, repetition or missing context. Rewrite tighter, in my tone.
The takeaway
I uses the 'glitch prompt' all the time, but it's important to know, hallucinations are not wiped out completely. A chatbot can still be wrong even when you specifically prompt it to avoid errors. This is especially true if you need real-time info or pricing or very specific advice that should be handled by a professional (i.e. legal or medical scenarios).
But what the glitch prompt does is reduce your chances of falling for a confident, wrong answer. It helps AI respond with more care and specifics. If you use ChatGPT, Gemini or Claude regularly — save this prompt. You’ll use it more than you think.
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.
