ChatGPT at work: a cheat sheet for facts, format and privacy

Give it the facts

• It knows the current chat, its general training and what you set up. Your current prices and policies have to come from you.
• Number your facts, so you can match each claim to one.
• Add the rule "Don't invent details or promise anything that isn't in the facts."
• Start a new chat for a new task.

Set the format and length

• Cover Role, Context, Task and Format: who is writing, the facts, what you want back and how it should look.
• Say the length, like "under 120 words", instead of trimming a 300-word reply by hand.
• Ask for a table when you need one, with "Not stated" for gaps.
• Ask for specific changes, like "a little warmer and under 120 words", not "make it better".

Check every promise

Check that reply against my facts. Make a table of every promise or claim in it and the fact that backs it up. Flag anything the facts don't support.

From the Reply Faster lesson. Still compare the table with your own facts.

Words to know

Hallucination

Something false or made up, stated as fact in the same confident voice as the true parts.

Knowledge cutoff

When its training data ends. It can't know newer prices or offers unless you tell it or a tool looks them up.

Data controls

A section in Settings. Turn off Improve the model for everyone so a personal account's chats aren't used for training.

Temporary Chat

A one-off chat with no history or memory, not used for training. Anything you paste still goes to OpenAI.

Keep private data out

• Remove card numbers, account numbers or health details first. Only paste what the task needs.
• Training off doesn't mean private: chats are still processed and stored to run the service.