Five mistakes account for most of the disappointing AI output people complain about. None of them are about the model being bad.
Asking for the destination without the constraints
“Write a product description” leaves the model guessing at length, tone, and audience. State the constraints upfront. Our full prompt engineering guide has the reusable template for this.
Accepting the first answer
Starting over with a longer prompt throws away the model’s biggest advantage: it remembers the conversation. “Make the second paragraph shorter” beats a rewrite from scratch.
Trusting specifics without checking
Models hallucinate most on exact statistics, citations, and recent events, precisely the details that sound most authoritative when wrong.
Cramming a multi-step task into one message
Research, outline, draft, and polish in one prompt tends to produce a mediocre version of all four. Break it into stages and review each before the next.
Describing style instead of showing it
“Professional but friendly” means something different to every reader. Paste an example of writing you like and ask it to match that.
Most of these cost one extra sentence of context. That’s a small price for consistently better output.
See Anthropic’s own prompt engineering documentation for more depth.
