Set up Claude in Outlook before you let it touch anything

I learnt this one the hard way. When Claude first landed in Outlook I did exactly what most people do — switched it on and started prompting randomly. I stayed up until 1am going down a rabbit hole, asking it to do this and that, getting mixed results and no real system out of it.

The three prompts below are what I wish I had run from the very start. Do these before you ask Claude to reply to anything or move a single email, and you get something the random approach never gives you: a proper inbox system, built on your actual mail.

Why setup comes first

The trick is prompt chaining — the output of each prompt feeds into the next. Instead of asking for one clever thing, you have Claude look at your real inbox, decide how it should be organised, and then design the structure to match. By the third prompt you are not guessing; you are working from a picture of how you actually use email.

Skip this and you get what I got at 1am: a pile of individually-impressive answers that do not add up to a way of working.

The three prompts

1. Diagnose the inbox. Have Claude analyse the last 90 days of your sent and received mail and tell you what is really going on in there — who you talk to most, what keeps coming back, what you start and never finish, where the time goes. Almost everyone is surprised by their own diagnosis. You cannot fix an inbox you have never honestly looked at.

2. Build the management strategy. Feed that diagnosis into a second prompt that decides the rules: what stays in the inbox, what gets archived, what needs a follow-up, and what your daily and weekly routine should be. This is where you build “touch it once” habits — the discipline of dealing with a message rather than re-reading it four times.

3. Design the Outlook setup. The third prompt turns that strategy into the actual furniture — folders, categories, flags and rules — and, importantly, gives each proposed rule a risk rating so you know which changes are safe and which deserve a closer look before you switch them on.

By the end you do not have a clever demo. You have a system, built from your own email, that a normal person can actually run.

Keep it simple enough to stick

The biggest trap here is over-engineering. A twelve-folder taxonomy with fifteen rules is a system you will abandon in a fortnight. When Claude proposes the setup, prune it. The best inbox system is the simplest one you will still be following in three months — not the most sophisticated one you built on a Sunday.

Two honest caveats, because this touches real mail:

• Claude will not get every number perfectly right. Read its diagnosis and its proposed rules before you act on them — treat them as a smart draft, not gospel.
• If your organisation manages Microsoft add-ins centrally, check with your IT team before connecting Claude to your work email. That is a five-minute conversation that saves a very awkward one.

Then, and only then

Once the system is in place, everything else in the Claude in Outlook toolkit works better — the drafting, the triage, the hidden skills all have a structure to plug into. Setup is the unglamorous half nobody demos, and it is exactly the half that decides whether this sticks or joins the pile of AI things you tried once.

If you would rather not build the system alone — or you want it rolled out across a whole team with the habits and guardrails in place — that is what our workshops and the AI Execution Sprint are for.