All plays › Other

86’d · What not to do
OtherChatGPT

Don't send AI-cited sources you never checked

What happened

as reported

$5,000 court sanction

Two New York lawyers filed a brief citing court cases ChatGPT had made up. A federal judge fined them and their firm and ordered letters to every real judge named in the fake rulings.

Who did it

Steven A. Schwartz and Peter LoDucaLevidow, Levidow & ObermanMata v. Avianca, sanctions order (S.D.N.Y.) · June 2023

Try it now

The safe way to do this instead:

  1. Treat every AI citation, statistic or quote as unverified
  2. Open and read each original source yourself
  3. Ask the AI to list its sources separately so you can check each
  4. Have a second person check anything going to a court, regulator or client
  5. If you find an error, correct it in writing right away
Copy-paste prompt
You are my fact-check assistant. Below is a draft I plan to send to a client or file officially. For every factual claim, citation, case name, statistic, quote or regulation in it, make a numbered table with: (1) the claim as written, (2) the specific source it depends on, (3) whether you are certain the source exists, (4) what I must look up to confirm it. Mark anything you aren't 100% sure exists as UNVERIFIED. Don't invent or guess sources. Finish with a list of everything I must check by hand before sending.

Draft: [PASTE DRAFT]

Want this set up on your own site?

Build it with Page

What to watch out for

From 2023: an older lesson. Asking the same chat 'is this real?' isn't a check; it can confirm its own invention. Your name on the document makes you responsible, whoever wrote it.

Did it work?

PageMotor keeps this site free.