Can you use AI to clean up your email inbox?

Yes, with care. AI can sort your inbox, summarise long threads, draft replies and help you work out what to unsubscribe from. That part is genuinely useful. The care is about access, because an inbox is not just yours. It is full of messages written by other people, and connecting an AI to the whole lot hands it all of that too.

So here is the order to go in: unsubscribe, block and filter first, AI second. Much of the clutter in a messy inbox needs no clever software at all, only a bit of ruthlessness.

How to do it well

Start with the boring tools, because they work. Marketing emails have to give you a way to opt out, usually an unsubscribe link at the bottom. Use it. The ICO says organisations, charities included, must stop once you opt out. If there is no link, contact the organisation directly; a suitable address is usually in its privacy notice.

For senders you never want to hear from again, block them. In Gmail, blocking sends all their future emails to Spam; for promotional emails you do not want to block, Google suggests unsubscribing instead. Then filters (or rules) and labels do the rest: receipts land in one place, newsletters in another, and your inbox stops being a pile and starts being a list.

Now the tiring bits, which is where AI earns its place. A forty-message thread about a delayed delivery where you only need to know what was decided. A reply you know how to write but cannot face typing. A filter rule you cannot quite work out how to word.

Here is a prompt for the thread job. Paste in one thread only:

Below is one email thread. Please give me three things:
1. Decisions that have been made, as plain sentences.
2. Open questions that still need an answer, and who they are waiting on.
3. A short, polite suggested reply from me, in plain British English.
Do not invent details that aren't in the thread. If something is unclear, say so.

[paste thread here]

And one for filter rules. It uses sender names only, so you are not handing over any message content:

Here is a list of sender names that clutter my inbox. Group them into
sensible categories (for example receipts, newsletters, social
notifications, promotions) and suggest filter rules for each group,
explaining what each rule would do in plain English. I will set the
rules up myself.

[paste sender names here]

Same principle both times: you stay in charge. The AI proposes, you read, you press the button. Treat a drafted reply as a first draft and read it before it goes anywhere.

Where you can, use the assistant already built into your mail service, or paste in a single thread, before you think about granting a third-party app access to everything. Least access, always. If a task only needs one thread, give it one thread.

Rules and risks

The awkward bit first, because it is the reason for the “with care” in the verdict.

The National Cyber Security Centre (NCSC) has written about indirect prompt injection. That is where someone with no access to your account writes text that an AI later reads and treats as an instruction; the NCSC’s example is hidden text in a CV. It says current large language models do not enforce a security boundary between instructions and the data they are reading, and that this will remain a residual risk no product can fully remove.

In plain English: an AI with the power to send, forward or delete is reading text written by anyone who can email you. The NCSC’s advice is not to let an AI that processes emails from random external people have access to privileged tools, and sending and deleting are exactly that. Keep it to read-and-suggest, and never let it send or delete without you checking.

Second, scams. Phishing is when criminals use scam emails, texts or calls to trick people, often into visiting a website that downloads a virus or steals bank details. A tidy AI summary can strip out the urgency and odd details that give a fake message away, like a slightly wrong sender address or pressure to act today. Read suspicious messages yourself. The ICO advises against replying to emails from senders you do not recognise and against clicking adverts in spam, because both show your address is live. Block suspicious messages and report them to your email provider; you can also report scam emails to the NCSC for free. If marketing carries on after you have opted out, you can complain to the ICO using its complaint form.

Third, privacy, which cuts two ways. Take Google’s Gemini as an example of what to look for in any provider’s terms. Google says human reviewers, including trained reviewers from its service providers, review some data, and advises you not to enter confidential information you would not want a reviewer to see. With Keep Activity off, chats are still kept for 72 hours, and chats already reviewed are kept for up to three years, disconnected from your account. Work or school accounts may have different terms. And when you ask it to summarise an email from a named contact, Google notes it is processing personal data about other people.

For small-business owners this matters more. A business inbox holds customers’ personal data, and under UK GDPR the business is responsible for it. The ICO explains that a processor handles personal data on behalf of a controller and acts under its instructions, so before you connect any AI service to a business inbox, read the provider’s terms and the ICO’s guidance on controllers and processors. Do not just click “allow”.

The clever part is optional. The careful part is not.

This is general information, not legal advice.