IN SHORT
Automatic sorting starts with rules: "whatever comes from this sender, or has this word in the subject, goes to this category". Rules are predictable and cover most messages. Artificial intelligence helps with the rest, where there is no fixed sender or word, at the cost that it can be wrong and that the message is read by a model.
Rules: the foundation
A rule has two parts: a condition and an action. Outlook describes a rule as a name, a condition, an action and, optionally, exceptions (Outlook (opens a new page)). Gmail calls them filters: you write search criteria, see which messages match, and choose what happens to them (Gmail (opens a new page)).
The conditions that work best:
- A sender or a whole domain. Whatever comes from @accountant.gr goes to "Accounting". This is the most reliable condition.
- Words in the subject. "Invoice", "order", "quote".
- The address it was sent to. Whatever arrived at orders@ goes to "Orders".
- Exceptions. "Unless the subject says 'cancellation'."
Folders or labels?
A folder keeps a message in one place. A label, as in Gmail, lets one message belong to two categories at once, such as "Customer X" and "Invoices". Both work. What matters is that the categories match what you do with the messages (see organising email).
Common mistakes with rules
- Words that are too general. A rule for "account" catches the electricity account, the bank account and "you created an account".
- Greek with and without accents, in different cases. "Τιμολόγιο", "τιμολογίου", "ΤΙΜΟΛΟΓΙΟ". A rule that looks for one form misses the others. Check how your program handles accents and endings.
- Rules that clash. If a message matches two rules, which one wins? In Outlook, for example, rules run in the order of the list (Outlook (opens a new page)).
- Rules nobody checks. A rule that sends messages straight to the archive can hide something important for weeks.
A good habit: before saving a rule, see which of your earlier messages it would match. In Gmail you do this with the search before creating the filter (Gmail (opens a new page)).
Where artificial intelligence helps
Rules stumble when there is no fixed sign:
- a new customer writing for the first time, from Gmail,
- a message saying "here are the papers you asked for" without any keyword,
- a question that could be a sale or a complaint.
There, an AI model can read the text and suggest a category. Three things to know:
- It makes mistakes. It needs a way for you to correct it, and your corrections should count.
- It reads the message. To sort it, the text is sent to a model. Ask where this happens and what is kept. More in the guide AI in email.
- It is not needed for the predictable. Whatever a rule can decide is better decided by the rule: it is faster, costs nothing, and you always know why.
The best combination: rules first, AI only for what is left, and your own choices above everything.
A note on security
Rules are useful to intruders too: a hidden forwarding rule can send copies of your mail somewhere else. Microsoft lists it as a sign of a compromised account (Microsoft (opens a new page)). If you find a rule you did not create, see the guide I clicked a suspicious link.
Where Sortaki helps
In Sortaki, every category has rules with senders and words. The word "τιμολογ" catches "τιμολόγιο" and "τιμολογίου", with or without accents, and before saving you see which messages would go in. Whatever the rules cannot decide is looked at by AI, only if you have turned it on. Whatever you set by hand stays as you set it, and every message tells you why it went where it did. Nothing changes in your mailbox. See Sorting.
Frequently asked questions
Can I sort email automatically without AI?
Yes. Rules with senders, domains and subject words cover most recurring messages. AI is needed only for those without a fixed sign.
Do rules apply to older messages too?
It depends on the program. The new Outlook, for example, can run a rule on existing messages. Check what yours does before relying on it.
Is it safe to let AI sort my email?
It depends on where the text is sent and what is kept. Ask the provider whether the data is used to train models, where it is processed, and whether you can turn it off.