NED SHIPS ITNED SHIPS ITNED SHIPS ITNED SHIPS ITNED SHIPS ITNED SHIPS ITNED SHIPS ITNED SHIPS IT
A free guide from the Ned Ships It video
How to Stop AI Hallucinations: 3 Free Prompts
Your AI guesses because guessing gets rewarded. These three copy-paste prompts make it admit when it doesn't know.
Commented TRUTH? This box is yours.
From
Ned, @nedshipsit
To
You
In the box
3 free tools
Checked
October 7, 2026
Why your AI makes things up
OpenAI explained it in its own research: models are trained and tested like students on a multiple-choice exam. A wild guess might be right. "I don't know" is guaranteed zero points. So the model learns to guess.
Their example: ask a model for someone's birthday it doesn't know. If it guesses "September 10", it has a 1 in 365 chance of being right, and on a scoreboard that still beats saying "I don't know".
On OpenAI's SimpleQA test
Said "I don't know"
Right
Wrong
A model that almost always guesses (o4-mini)
1%
24%
75%
A model that admits it (gpt-5-thinking-mini)
52%
22%
26%
The careful model got about the same number right, and was wrong a third as often. The fix is to give your AI the same permission.
picture a kid guessing on a test because a blank answer scores zero. Same thing.
Quote the exact source for every claim you make. If you can't find a quote that supports a claim, take the claim back.
Try it
Paste a long article and ask:
You, to Claude
Summarize this in 5 points. For each point, quote the sentence it comes from.
Heads up This comes straight from Anthropic's own guide to reducing hallucinations: ground answers in direct quotes, and retract anything without one. More
Quotes only help if you glance at them, so spot-check one or two.
Open a fresh chat (or a different AI), paste the first answer, then add:
You, to Claude
Check every claim in the answer above. List anything that isn't supported or that you think is wrong, and explain why.
Try it
Run it on the last important answer you got from AI, like a summary you were about to send.
Heads up A fresh chat matters because the first chat is already committed to its answer. More
In the same 16-model test, a second, cheaper AI checking the answers caught 38 of 49 made-up values and wrongly flagged none of the correct ones.
Keep them on for good
Paste this once and every new chat gets all three rules:
Paste this once
If you're not sure about something, say so. Do not guess. It's fine to say "I don't know".
When you state a fact, tell me where it comes from. If you can't point to a source, say it's unverified.
For anything important, remind me to double-check it in a fresh chat.
ChatGPT: click your profile picture, open Settings, then Personalization, and paste it into Custom instructions (the box about how ChatGPT should respond). Make sure it's turned on for new chats.
Claude: click your initials in the lower left, open Settings, and paste it into the box "What preferences should Claude consider in responses?". Claude reads it at the start of every new chat.
set it once, and your AI gets a little more honest in every chat after that.
Questions people ask
Why does AI make things up?
OpenAI's own research says training and testing reward a confident guess over saying 'I don't know'. A guess might score points, 'I don't know' always scores zero, so models learn to guess.
Will these prompts stop hallucinations completely?
No. They cut made-up answers down a lot, but no prompt makes AI perfect. Anthropic says the same about its own techniques. For anything important, check the source yourself.
Do these work in ChatGPT, Claude and Gemini?
Yes. They are plain instructions, so they work in any chat AI. The 'keep them on for good' steps below cover ChatGPT and Claude.
Which prompt should I use first?
Prompt 1. It's one line, costs nothing, and in one test it cut made-up answers from about 71% to 20%.
Where Ned does his homework
The tools are free. Knowing what to ask them for is the real skill. Professor Glitch teaches it from zero, in order, inside the community: 8 courses and 200+ lessons. These two are where Ned started.
