A five-day starter plan for using AI chat tools at work

Monday: the email you've been putting off

Everyone has one. Paste the facts in a few bullets and ask:

Write a short, polite email from me to [who] about [what].
Facts: [bullets]. Under 120 words. Don't add anything
I didn't mention.

Read it out loud. Change anything you wouldn't say.

Tuesday: messy notes into a to-do list

Paste your scribbled notes from a call or meeting and ask: "Turn this into a to-do list. One line per task, with who does it, if my notes say. Put anything unclear at the bottom as a question."

Wednesday: explain something confusing

A letter from the bank, a term in a contract, a vendor's invoice. Ask: "Explain this like I'm new to the job. Then tell me what question I should ask the person who sent it." You're not taking its word as final. You're getting ready to ask better questions.

Thursday: plan a crowded day

List what has to happen tomorrow and how long each thing usually takes. Ask: "Put this in an order that makes sense. Flag anything that won't fit." You decide the final order. It just gives you a starting draft.

Friday: keep or drop

Look back at the four jobs and answer two questions for each:

• Did it save me time, after fixing its mistakes?
• Would I do it this way again next week?

Keep the ones with two yeses. Drop the rest without guilt. One job that works every week is worth more than five you tried once. If you want to bring the rest of your team along later, see how to get staff using AI tools.

If something goes wrong

It will get something wrong this week. That's normal. Chat tools sometimes state things that aren't true, so check names, dates, and numbers before they leave your desk. Here's a five-minute way to fact-check an AI answer.

Under the hood

Why these five tasks: each one is a different capability with a different failure mode.

• Drafting (Mon): fluent but may add facts. Constrain with "don't add anything I didn't mention."
• Extraction (Tue): good at structure, weak on unstated owners. Route unknowns to a question list.
• Explanation (Wed): useful for orientation, unreliable as authority. Convert the output into questions for a human.
• Planning (Thu): reasonable ordering, no real knowledge of your constraints. Treat it as a first draft.
• Review (Fri): the metric is net time saved after corrections, not output quality in isolation.

Data minimization matters more than tool choice in week one. Consumer plans differ in whether conversations can be used for training; check the setting in your account before pasting anything about customers.

Short close

One small real job a day, nothing private in the chat, read everything before it goes out, and keep only what saved you time.