I asked ChatGPT why Einstein won the Nobel Prize for relativity. It gave me a confident answer, and it was wrong. This prompt goes in your Custom instructions once, and tells ChatGPT to back up what it says, admit when it isn't sure, and correct you when your question is wrong.
In the video I said to comment FACTCHECK. This is that guide. Einstein won the 1921 Nobel Prize in Physics for the law of the photoelectric effect, not for relativity. ChatGPT took my wrong question and ran with it. With this prompt saved, the same question got a correction and a source.
The prompt tells ChatGPT three things: how to answer you, what to never do, and to check every statement before it sends anything.
How to answer me: - Only tell me what you can back up. If you're not sure, say "I'm not sure" and tell me what would confirm it. - Give a source for every fact, number, quote and date. Link it when you can. - Keep facts and opinions separate, and label your opinions. - If my question is based on something wrong, correct it before you answer. - Being accurate matters more to me than being fast or sounding confident. Never: - Make up facts, quotes, statistics, studies or sources. - Fill a gap with a guess and present it as fact. - Give me old information without telling me how old it is. Before you send any answer: Check every statement. Is it accurate, backed by a credible source, and cited? If not, fix it or take it out. If you can't confirm something, say so.
The prompt is about 770 characters. Free and Go accounts can save 1,500 characters of custom instructions, and Plus and up can save 5,000. If you already use my CODES, the two together come close to the free limit, so check it saves in full.
No prompt makes AI perfect. This one puts safeguards in place, and you still check anything that matters before you use it.
Give it standing rules in Custom instructions. Tell it to only say what it can back up, to say "I'm not sure" when it isn't, to cite a source for every fact, and to correct your question if it's based on something wrong. It still makes mistakes, so check anything that matters.
On the web or desktop app, open Settings, then Personalization. On iPhone or Android, open Settings, then Customize ChatGPT. Make sure Enable customization is on, paste into the Custom instructions field and save. They're available on every plan, including the free one.
It writes from patterns in its training data, and OpenAI says it can sound confident even when it's wrong. If your question has a wrong fact in it, it may run with it. I asked why Einstein won the Nobel Prize for relativity and got a confident answer, but he won it for the photoelectric effect.
Yes. OpenAI lists fabricated quotes, studies and citations, and references to sources that don't exist, as a kind of hallucination. Open every link it gives you, find the quote or number on the page itself, and search the exact title of any study. If you can't find it, don't use it.
Free and Go accounts can save up to 1,500 characters. Plus, Pro, Business, Enterprise and Edu accounts can save up to 5,000. The fact-check prompt on this page is about 770 characters, so it fits on any plan with room to spare.
A follow-up prompt to run on any answer before you use it, plus a short checklist for catching a source ChatGPT made up.
Check your last answer before I use it. 1. List every fact, number, date, name and quote in it. 2. For each one, give the source and link it. Search the web to confirm it if you can. 3. Mark each one: Confirmed, Not sure, or Wrong. 4. Fix anything wrong, remove anything you can't back up, and tell me what you changed. 5. Tell me if my original question was based on something that isn't true. Then give me the corrected answer.
For anything you'll publish or send to a client, turn on web search first so the sources are real pages. In a chat, select View all tools, then Search, or type / and choose Search. Then run the prompt above.
Prompts, step-by-step guides and checklists you can put to work today.