Spam scoring
Score incoming mail for spam, read the reasons behind a score, and act on it with rules.
Spam scoring is a paid feature. It reads a message and returns a confidence from 0 to 100 together with the reasons behind it, so you can act on suspicious mail without reading every one. The confidence says how sure the check is that the message is spam: 0 is certainly not, 100 is certainly. It is not a chance you can quote as a percentage, and a low number does not prove a message is genuine. Every check that completes ends in a number, so you will not see “uncertain” or “could not tell”. A check that fails reads “Check failed, try again” and has no number.
Turn it on
An administrator must enable the site-wide AI master setting and the spam feature configuration before this workflow is available. Turning the AI master off also disables spam scoring.
Spam scoring runs on whichever engine is selected for it in the plugin’s AI Tools settings:
- Disabled turns scoring off.
- Enabled (your AI provider) uses the provider configured on the site. Provider charges are separate, and these checks do not use PressedMail AI credits.
- PressedMail AI uses our engine. Spam scoring costs one credit per email, the same as auto-tagging or a phishing check. Spam and phishing share a credit when they run on the same email, so all three cost 2, and running a task again costs another.
Each AI feature has its own dropdown, so spam scoring can run on PressedMail AI while tagging or phishing detection uses your own provider, or the other way round. See PressedMail AI for what a check sends and PressedMail AI credits for the allowances.
Read the confidence
The list, the reading pane and the report show it as “Spam confidence 86%”, with the band beside it in the tooltip and the report. A confidence falls into one of four bands. The interface and the rules compare the band, not the bare number.
- Clean (0 to 29) found nothing worth flagging.
- Newsletter or promotion is bulk mail scoring under 50. Wanted bulk mail keeps its own band and is never treated as a little spammy; it ranks with clean.
- Suspicious is ordinary mail scoring 30 to 74, or bulk mail scoring 50 to 74. Worth a second look.
- Spam (75 to 100) found strong spam signals.
Open the message to see the confidence and the reasons behind it. The reasons name the signals that counted, so a result you disagree with tells you what to add to your rules or your block list.
How the number is made
With PressedMail AI, the model answers a handful of questions about the message: what kind of mail it is, what it sells, whether it claims an earlier conversation, whether it promises too much or hides its words. A second model, fitted on thousands of real and public messages, weighs those answers together with what your site can check about the sender: whether the message is signed by the sender’s own domain, its list headers, whether you have written to the sender, and what your mail provider decided. No single answer can mark a message as spam on its own, and nothing a sender writes about itself earns more than a limited credit.
Marking mail Not spam teaches your site that the sender is fine. A sender you marked as Junk counts strongly against their mail.
Use it in rules
Paid plans can build email rules on the result:
spam_scorecompares the confidence number, 0 to 100. The rule editor calls it Spam confidence.spam_bandcompares the band, so a rule can match “suspicious or worse” without naming a number.spam_categorymatches the category the check assigned, such as bulk mail or a commercial pitch.
A rule needs a result to compare. A message nobody has checked yet matches none of these conditions, negative ones included. Neither does one whose check could not decide: it still shows a number, but rules treat it as unknown.
These work alongside ordinary rule conditions. See Email workflows for how rules run and when.
What it is not
Spam scoring is a separate tool from phishing analysis. A message can be spam without being a phishing attempt, and a phishing warning does not make a message spam; Phishing reports covers that tool. Scoring itself never moves or deletes mail. It reports, and your rules decide what happens next.
One opt-in exception: under Settings > Preferences > AI, When a check flags an email can move flagged mail to a folder, tag it, or both, once a check reaches the level you pick. It is off until you turn it on, an administrator can switch it off, your own rules run first, and every move can be undone.