7 AI habits that actually work
They're the boring stuff you'd tell a new hire. Turns out the chatbot needs the same talk.
01 Define done first
PROMPT
Task: [what you need]. Done means: [length], [format], [what each part must include].
WHY Anthropic’s first rule is “be clear and direct.” OpenAI tells you to spell out what a good answer looks like.
02 Show, don’t describe
PROMPT
Here are 3 examples of what I want: [paste]. Match the style and format, not the content.
WHY Anthropic calls examples one of the most reliable ways to steer format and tone.
03 Turn thinking on
PROMPT
This is a hard problem. Think it through carefully before answering, then check your answer.
WHY Asking for reasoning boosted math scores in 2022 tests. Newer models do it built in, so switch thinking mode on first.
04 Ask for the other side
PROMPT
Here’s my plan: [paste]. Give me the 3 strongest arguments against it. Don’t soften them.
WHY Anthropic’s own research found AI assistants often lean toward what the user already believes.
05 Run a pre-mortem
PROMPT
Here’s the project: [paste]. Imagine it’s 6 months from now and it failed. List the most likely reasons why.
WHY Psychology research: imagining the failure already happened draws out more reasons it could go wrong.
06 Make “I don’t know” OK
PROMPT
[paste document] Using only this, answer: [your question]. If it’s not there, say “I don’t know.” Quote the line behind each claim.
WHY All three moves come straight from Anthropic’s guide to reducing hallucinations.
07 Verify in a fresh chat
PROMPT
Search the web and check each claim below separately. For each, give a source or mark it unverified: [paste claims]
WHY Checking claims one by one, apart from the original answer, cut errors in tests. It still misses things.
SAVE THIS. TRY ONE TODAY.
Pick the one you skip most. Use it on your next prompt.
Talk to us