IN SHORT
Artificial intelligence in email does three things well: it suggests a category for a message, writes summaries and draft replies, and answers questions about your mail. To do that it reads your messages, and it can be wrong. Before turning it on, find out where the data is sent, what is kept and whether you can turn it off.
What it does in practice
Sorting. It reads a message and suggests where it belongs: customer, supplier, accounting. Useful where rules are not enough, such as for new senders. See the guide Automatic email sorting.
Summaries. A long thread becomes three lines: what was asked, what was agreed, what is left.
Help with writing. A draft reply, a politer or shorter text, corrections. You read the final version and you send it.
Questions about your mail. "What did the accountant send me this week?" A good tool answers only from your own messages and shows where the answer came from.
Where it goes wrong
An AI model writes convincingly, even when it is wrong. In email that means:
- The wrong category. An important message may land where you will not look. You want a way to correct it, and your corrections should count.
- A summary that drops a detail. An amount, a date, an "unless". For anything with consequences, read the original.
- A reply that promises something. Never let AI send replies that commit the business on its own.
There is also something AI should not do at all: decide on payments. A convincing scam email can fool a model too. For fraud checks, rules that explain their reason are more reliable than a "feeling" (see email scams).
What to ask about your data
To do any of the above, AI reads your messages, and with them the personal data of customers, partners and colleagues inside them. Before turning such a feature on, ask the provider:
- Where is the text sent? To which model provider, and in which country is it processed?
- What is kept, and for how long? Are copies of the messages or answers kept?
- Is it used for training? Do your messages become training material for models?
- Can I turn it off? And what keeps working without it?
- Who sees what? In a team, AI should see only what the person asking is allowed to see.
For businesses in the EU, when a provider processes personal data on your behalf, the General Data Protection Regulation requires a written contract with it, covering what Article 28 sets out (GDPR (opens a new page)). Ask for it. For questions about your obligations, the Hellenic Data Protection Authority (opens a new page) publishes guidance, and it is wise to speak with whoever advises you on data protection.
One simple rule
AI is good at "suggest" and bad at "decide on your own". Use it to save time on what you read and write, not to leave decisions unchecked.
Where Sortaki helps
In Sortaki the fraud check uses no AI: it uses rules, and every warning gives its reason. In sorting, your rules decide first, and AI looks only at what they cannot tell apart, only if you have turned it on. Whatever you set by hand stays. The Assistant answers only from your own messages, shows where the answer came from, and a colleague sees through it only what they are allowed to see. Sortaki reads a copy: it does not move or delete anything in your mailbox. See Sorting.
Frequently asked questions
Does AI read all my email?
An AI feature reads whatever messages the program gives it to do its job. Ask the provider which messages are sent, where, and what is kept.
Are my emails used to train AI models?
It depends on the provider and its agreement with the model provider. It is one of the first questions worth asking, and the answer should be in writing.
Can AI tell whether an email is a scam?
It can help, but a convincing scam email fools models too, and AI does not always explain why it decided something. For payments and IBAN changes, always confirm by phone.