List every assumption you made while writing that, and mark each one as “from the files” or “my guess”. Then tell me what in the files contradicts or weakens what you wrote.
Trust, but check: reading AI output like a PM
The assistant is a confident junior who never says “I don’t know”. Where it fails, the five-line check, and when not to use it at all.
Page 6 of 7 · Last checked
The assistant is a confident junior who never says “I don’t know”. The PM’s job does not change: verify before you sign. This is the short page, with the checklist.
Where it fails
- Invented numbers and dates, in the same confident voice as the real ones.
- Invented citations - “PMBOK section 4.3.2 says” - it may not.
- Confident tone on thin context: the less you gave it, the more it fills in.
- Agreeing with you. Ask a leading question and you get a leading answer.
- Drifting from your format late in a long chat.
The five-line check
Ask for its working
Two follow-ups that catch most errors
A second opinion, cheaply
Paste the draft into a fresh chat - or into a different assistant - and ask it to find the mistakes. Vendor-neutral, and surprisingly effective: the second reader has none of the first one’s momentum.
Own the output
The PMI Code of Ethics has no AI exemption. The report with your name on it is yours, however it was drafted. The dilemma of the week already teaches this instinct; this page is where it meets the tool.
When not to use it at all
- Legal wording and contract clauses.
- Contract numbers, prices and anything with a signature line.
- Anything about a named person’s performance.
- Anything on the red list.
Last checked by Petar Trajchevski. Product names and settings paths were walked through on that date; the concepts do not move. Report a change