My inbox used to be the thing I opened first thing and immediately regretted. Two hundred unread, half of them newsletters I never signed up for (or think I never signed up for), a few things from actual humans buried somewhere in there. AI can help with this. Not by taking it over completely, but by doing the boring bits, so you can do the bits that need a human.

This guide is about a workflow, not a magic trick. Rules and labels first, AI second. That order matters and I’ll explain why.

What you’ll be able to do

By the end of this you’ll be able to sort a backlog without reading every email individually, get a fair summary of a long thread instead of scrolling through fifteen replies, and draft responses that sound roughly like you rather than a corporate robot. You won’t be handing your entire inbox over to an assistant and hoping for the best. That’s not the goal, and honestly it’s not a good idea either.

The tools I’m talking about here are the ones already sitting in the products a lot of us use: Gmail with its Gemini features, Outlook with Copilot, Claude connected to Gmail, and ChatGPT’s connectors. None of them are perfect. All of them are useful if you set them up properly and know their limits.

Before you start

Right, before you connect anything to anything, think about what you’re actually giving away. An AI assistant that reads your email reads your email. All of it, or whatever slice you grant it access to. That includes attachments, names, addresses, anything sensitive sitting in threads from months ago that you’d forgotten existed.

Check the data-use terms for whichever plan you’re on. This differs between free and paid tiers, and between personal and business accounts, so don’t assume yesterday’s answer still applies. Gmail’s Gemini features and Outlook’s Copilot behave differently depending on whether you’re on a consumer plan or a Microsoft 365 plan with Copilot licensed. Claude’s Gmail integration is a connection you set up deliberately, not something switched on by default. ChatGPT’s connectors to Gmail and Outlook are available on some plans, not all, so check yours before you get excited.

Here’s the bit I actually care about: keep client work and anything medical off consumer AI tools entirely. If you’re a freelancer with client correspondence, or you’ve got health-related emails in there, don’t feed those into a general-purpose assistant just to save yourself a read. The convenience isn’t worth it. Some threads are for you and maybe a doctor, not for a language model.

Get set up

Do the boring, deterministic stuff first. Rules and labels. This isn’t glamorous but it does more heavy lifting than people expect, and it’s free.

In Gmail, set up filters that automatically label and archive newsletters, receipts, and anything from senders you don’t need to see immediately. In Outlook, rules do the same job, sorting by sender or subject line into folders before anything even reaches your main view. Get this right and you’ve already cut your visible inbox by half, before AI touches anything.

Then, connect an assistant. In Gmail, Gemini summaries and “Help me write” are built in, no separate setup beyond having access to the feature on your account. In Outlook, Copilot can summarise a thread and draft a reply on eligible Microsoft 365 plans, check with your admin if you’re on a work account and unsure. If you use Claude, you can add Gmail as an integration through Claude.ai, which lets you ask it questions about your mail directly in a chat. ChatGPT has connectors to Gmail and Outlook too, on some plans, so again, check what you’ve got before assuming it’s there.

Email’s only half the pile, though. If your chats are the real mess, a group thread or a week of WhatsApp, you can get an assistant to summarise your messages so you catch up without scrolling through 400 of them. That’s the bit I’d do next, because the same connect-and-ask habit carries over.

Pick one. Don’t try to run all four at once, that’s a recipe for confusion and duplicate work. I use one for summarising and one for drafting, and that’s plenty.

Try it yourself

Here are three prompts that each do a different job. Copy them, adjust the bracketed bits, and see what comes back.

Sorting a backlog, once you’ve got labels set up and want a human-readable digest of what’s sitting there:

Look at my unread emails from the last 7 days. Group them into: needs a reply today, can wait, and can probably be archived. For each group, list the sender and subject line only. Don't draft any replies yet.

Summarising a long thread, useful for anything with more than five or six replies:

Summarise this email thread in under 150 words. Include: what's being asked of me, any deadlines mentioned, and who else is involved. Don't include pleasantries or small talk from the thread, just the substance.

Drafting a reply, when you know roughly what you want to say but can’t be bothered to write the polite version:

Draft a reply to this email. Keep it under 100 words, friendly but direct, and confirm [specific detail, e.g. "the meeting time of 3pm Thursday"]. Don't add any extra offers or suggestions I haven't mentioned.

Small tweaks make a big difference here. Word limits stop the AI padding things out. Telling it what not to do (no pleasantries, no extra offers) stops it inventing content you didn’t ask for.

Check the result

Don’t trust a summary or a draft just because it reads confidently. Confidence and accuracy are not the same thing, and AI is very good at sounding sure of itself.

For summaries, pick a thread you already know well and run it through the assistant. Does the summary match what you remember happening? Did it miss a deadline, or invent one? Check names, check dates, check anything that looks like a number, because numbers and dates are exactly the kind of detail that gets quietly wrong.

For drafts, read the reply out loud. Does it sound like something you’d actually say, or does it sound like a template? If every draft comes back with the same three phrases, that’s the AI’s house style bleeding through, not yours. Edit it or ask for a different tone.

For sorting, spot-check the “can probably be archived” pile before you actually archive anything. This is the group most likely to have an important email wrongly demoted, because urgency isn’t always obvious from a subject line.

One more test, worth doing weekly: ask your assistant what you ignored.

Look at emails older than 5 days that are still unread. List anything that looks like it needed a response and didn't get one. Be specific about what was being asked.

This one’s caught things for me that I’d genuinely have missed. It’s the closest thing to a safety net in this whole workflow.

If it doesn’t work

Sometimes the summary is vague. Sometimes the draft sounds nothing like you. Sometimes you just don’t trust it, and that’s fine, trust your gut here.

If summaries keep missing detail, the prompt is probably too open. Be more specific about what you want included, deadlines, names, decisions, and tell it explicitly what to leave out.

If drafts come back sounding stiff or oddly formal, that’s usually a tone problem rather than a content problem. Give it an example of how you actually write, a previous email of yours, and ask it to match that style. Still won’t be perfect. It rarely is on the first go.

If you’re worried about privacy, that worry is doing its job, listen to it. Don’t connect an AI assistant to an inbox that has client contracts, medical correspondence, or anything you wouldn’t want summarised by a third party. If in doubt, leave it out. There’s no prompt that fixes a decision you shouldn’t have made in the first place.

And sometimes the answer is just: don’t use AI for this one. Some emails need a human brain and nothing else. That’s not a failure of the tool, it’s just not what it’s for.

Questions people ask

Which AI assistant should I use to manage my inbox?

The useful options are Gmail with its Gemini features, Outlook with Copilot, Claude connected to Gmail, or ChatGPT's connectors. Pick one rather than trying to run all four at once, that's a recipe for confusion and duplicate work.

Do I need to set up filters before connecting AI to my email?

Yes, do the boring, deterministic stuff first. Gmail filters or Outlook rules that label and archive newsletters, receipts and non-urgent senders can cut your visible inbox by half before AI touches anything, and it's free.

Is AI email management free or do I need a paid plan?

Rules and labels are free. Beyond that it depends on your plan: Gmail's Gemini features and Outlook's Copilot behave differently on consumer versus Microsoft 365 plans, and ChatGPT's connectors are available on some plans, not all, so check yours.

What emails should I not let an AI assistant read?

Keep client work and anything medical off consumer AI tools entirely. Don't connect an assistant to an inbox holding client contracts, medical correspondence, or anything you wouldn't want summarised by a third party. If in doubt, leave it out.

How do I check an AI email summary is accurate?

Run a thread you already know well through the assistant and see if the summary matches what you remember. Check names, dates and anything that looks like a number, because those are exactly the details that get quietly wrong.

Keep exploring

If the drafting side is where you lose the most time, have a look at “How to get AI to send emails for me” (the rule there is the same one as here: AI drafts, you send). And if the drafts keep coming back sounding nothing like you, “How to get AI to sound like me” goes into fixing that properly.

Sources and review notes

Review date: 15 September 2026 — every source above was opened and checked against the text on that date.