A quick diagnostic prompt to “sanity check” ChatGPT

PROMPT ENGINEERING & ACCURACY

TL;DR: Use a sanity check prompt to make ChatGPT critique its own output, exposing unverifiable claims, vague logic, and weak spots that need review.

Let’s be realistic. I can caution everybody as much as I can about AI, yet people are still going to use it as their go-to source for information now.

AI falls into what we call “satisficing,” which is where people settle for something that’s “good enough,” especially when the effort is too high.

ChatGPT often feels like “good enough,” even when we know it’s often bullshitting us. So how do we add just a bit of friction to our questions?

Here’s one way: sanity-check the output.

🧪 What Is a Sanity-Check Meta-Level Prompt?

I started seeing the instruction “sanity check” pop up in ChatGPT’s internal reasoning with the advent of ChatGPT-5 Thinking, so it’s evidently OpenAI’s working solution to attempt to minimize hallucinations. It’s not magic. But it does help, especially when trying to summarize technical or niche topics.

Here’s my simple prompt you can paste in after output you’re uncertain of:

Perform an in-depth sanity-check on the above output, avoiding…