The single biggest upgrade most beginners can make isn't clever wording — it's pasting the actual materials. The email thread you're replying to. The error message, complete and uncut. The paragraph you want rewritten. The AI can't see your screen, your inbox, or your codebase unless you put them in the prompt.
Then add one instruction that changes everything: "work only from what I've given you." Without it, the model happily fills gaps with plausible-sounding inventions. With it, the model tells you when something is missing instead of guessing.
Below is a client email. Draft a reply from me (project lead at a 12-person agency).
EMAIL THREAD: "Team — this is the second deadline that's slipped. We were promised the dashboard on the 3rd. I need to tell my VP something concrete today. — Dana"
FACTS YOU MAY USE (nothing else):
Tone: accountable, no excuses, no corporate filler. Under 120 words. End with one concrete next step Dana can forward to her VP.
My Python script crashes. Work only from the code and traceback below — if the cause isn't visible in them, say what else you'd need to see.
CODE (relevant function):
def load_user(payload):
user = payload["user"]
return user["email"].lower()
TRACEBACK:
KeyError: 'email'
SAMPLE PAYLOAD that triggered it:
{"user": {"mail": "[email protected]", "id": 7}}
Give: (1) the root cause in one sentence, (2) the minimal fix, (3) how to make the function fail with a clearer message next time.
Claude handles long pasted material well — put documents first and your instruction after them. For multiple documents, wrap each in XML-style tags like and ; Claude tracks tagged sections especially reliably.