AI can and will hallucinate.
AI can make something up and sound completely sure about it. That’s why I ask for sources in my first message. If I’m asking AI what happened on a project, I want the exact line it’s relying on right there in the answer.
Fictional example
What I ask up front
“I uploaded a project email, meeting notes, and a status update. Put the events in date order. Cite the document and quote the line that supports each entry. If a date was proposed and later changed, show both. If you can’t find when something actually happened, say so.”
What a good answer looks like
What I check myself
March 17 shows up in the documents, but it was only a proposal. The March 24 status update is what tells me the kickoff actually happened.
Always, always check your work.
Trust but verify.
Copy the prompt
Use only the documents I’ve provided to answer [question].
Put a numbered footnote after each key claim. In each footnote, give the document name, the section, page, or date if available, and a short exact quote that supports the claim. Link to the source if I supplied a link or you can verify it. If I pasted the text without a link, use the heading and the exact quote.
Separate what the documents explicitly say from what you infer. If they don’t answer the question, say “I couldn’t find that in the documents you gave me” and tell me what to confirm. If two passages conflict, show both.
Before you return the answer, double-check every key claim against the original documents. Make sure each quote matches the source exactly, each reference points to the right place, and each quote actually supports the claim beside it. Correct or remove unsupported claims. Flag anything you still can’t verify.
[Add your documents or source links.]