Skip to main content

Classification Signals

When AI is turned on, MailPrism reads each email and produces a set of signals. This page explains every signal, the values it can take, and what each value means. All of these are available as rule conditions.

Requires AI

These signals only exist when AI features are enabled. See AI, Explained and AI privacy & consent.

Category

The category is MailPrism's best guess at what kind of email this is. These are the built-in categories:

urgent important personal work financial newsletter promotional cold_email spam social notification transactional system
CategoryMeaning
urgentNeeds your attention right now
importantHigh-priority, but not time-critical
personalFrom friends, family, or your personal life
workJob- or project-related
financialInvoices, billing, receipts, statements
newsletterSubscriptions and recurring digests
promotionalMarketing, sales, and offers
cold_emailUnsolicited outreach you didn't ask for
spamUnwanted or junk mail
socialNotifications from social networks
notificationAutomated alerts from apps and services
transactionalOrder confirmations, shipping, account actions
systemSystem-generated mail (e.g. delivery reports)
You can add your own

Categories aren't fixed. On Pro and above you can create custom categories and even edit the built-in ones. See Custom categories.

Urgency

How time-sensitive the email is.

ValueMeaning
highAct soon
mediumWorth attention, not pressing
lowNo rush

Sentiment

The emotional tone the AI reads in the message.

ValueMeaning
positiveFriendly, appreciative, upbeat
neutralMatter-of-fact
negativeFrustrated, unhappy, or critical

Great for spotting an unhappy customer before it escalates.

Yes / No signals

These are simple true/false flags. Each is usable as a rule condition with an is true / is false operator.

SignalMeans it's true when…
Is spamThe email looks like junk or unwanted mail
Is automatedThe email is machine-generated with no human sender (receipts, notifications, system alerts)
Is cold outreachThe email is an unsolicited sales pitch or cold outreach
Needs response (detected)The email looks like it needs a reply from you
Requires action (detected)The email needs an action outside email — sign in, approve, update payment, etc.
Sender expects reply (detected)The sender appears to be waiting on a response from you
Cold outreach vs. needs-response

An email can't be both "cold outreach" and "needs a response" at the same time — by design, flagging something as cold outreach keeps it out of your needs-action queue.

The last three signals — needs response, requires action, and expects reply — also feed response tracking, which decides a whole thread's state rather than judging a single message.

Confidence

Every classification carries a confidence score from 0.0 to 1.0 — how sure the AI is about its answer. You can use this to keep automations conservative:

If Category is cold_email and confidence is greater than 0.8, archive it.

Two related, optional values may also be present:

  • Alternative category — the AI's runner-up guess, when it was torn between two.
  • Category confidence — confidence specific to the category choice.
Combine AI with a plain condition

For anything destructive (archive, delete, forward), pair an AI signal with a concrete condition like sender relationship, and require high confidence. See Best practices.

Where these appear

SignalAs a rule condition
CategoryAI Category
UrgencyAI Urgency
SentimentAI Sentiment
Is spamAI: Is Spam
Is automatedAI: Is Automated Email
Is cold outreachAI: Is Cold Outreach
Needs responseAI: Email Needs Response
Requires actionAI: Requires External Action
Sender expects replyAI: Sender Expects Reply
Thread state confidenceAI: Thread State Confidence

See the full Conditions reference.

→ Next: Custom categories · Smart rules