PressedMail AI
Auto-tag mail and check it for phishing without your own AI key, paid for with credits.
PressedMail AI runs auto-tagging, spam scoring and phishing checks. No AI account, no API key, no model to pick. We run the AI.
It is part of the paid editions. Free never sends mail to PressedMail AI.
Want drafts and summaries too? Those still use your own provider. See Set up and use AI.
Turn it on
- Open PressedMail’s admin settings and go to AI Tools.
- Find auto-tagging, spam detection or phishing detection.
- Change the dropdown next to it from Disabled to PressedMail AI.
- Save.
The dropdown has three choices:
- Disabled turns the feature off.
- Enabled (your AI provider) uses the provider you set up yourself, billed by that provider.
- PressedMail AI uses our AI. Auto-tagging spends one credit per email. Spam and phishing checks together spend one separate credit per email. Running all three on the same email costs two credits.
Tagging has its own choice, and phishing and spam checks share one. Tag with PressedMail AI and check spam with your own provider, or the other way round. Pick PressedMail AI and your balance, a Refresh button and a Buy credits link show up right next to the dropdown. The Licensing tab has the same balance, plus exactly what PressedMail AI sends and where.
Two switches sit above these choices. The site-wide Enable AI features switch turns off every AI feature on the site. Each person can also switch off checks on their own new mail with Check my new mail in their personal AI settings. Summaries and drafting only offer your own provider.
What a credit buys
Auto-tagging costs one credit per email. Spam scoring and phishing checks share one separate credit on the same email, so all three cost two credits. Running an analysis again costs credits again.
Email rules can run the same tagging and checks, and they spend credits the same way. See Email rules.
Paid plans come with credits, and you can buy more whenever you like. Numbers, packs and where to buy are in PressedMail AI credits.
Tag confidence
Every tag the AI can apply has a Confidence setting in its edit dialog. It is the probability the AI must go above before it adds that tag on its own, from 1% to 99%. The default is 70%.
Raise it for tags where a wrong match is annoying, like “Invoice”. Lower it for loose tags where catching more is worth a few misses. The AI scores every tag separately, so one strict tag does not hold back the rest.
When credits run out
Tagging and phishing checks pause. Nothing breaks and nothing is lost. New mail waits in the queue, and a notice on PressedMail screens tells you why.
Once credits are added, work picks up again within the hour. Want it now? Press Refresh next to the balance and the queues wake straight away.
Per-user monthly limits
Credits come from one shared balance per license. On a team, that means one busy inbox could eat the lot. The Team credits tab in AI Tools stops that:
- Default monthly limit sets a limit for everyone. Leave it blank for no limit. A 0 turns PressedMail AI off for everyone who has no limit of their own, so the page asks you to confirm it before it saves.
- Everyone your roles or Access Control let into PressedMail gets a row with the credits they have used this month. Give them a different limit, or switch PressedMail AI off for them. Someone let in only by another plugin’s own permission rule, or a network administrator who is not a member of the site or whose role there gives no PressedMail access, has no row and follows the default limit.
- Big team? Search by name or email, or page through the list. Changes you make on one page are kept while you look at another, and they all save together.
When someone hits their cap, their new mail skips auto-tagging and phishing checks until next month or until you raise the limit. Spam scoring can still be decided from mailbox history, without a credit. That mail isn’t picked up later on its own, so run tagging or a phishing check by hand on anything that matters. Everyone else carries on. People can see their own usage under the AI section of their personal settings.
Check many emails at once
Select messages in the list and use the AI actions in the bulk bar. Above 10 messages, the app warns you first. Your administrator sets a cap for each selection, 50 by default and 200 at most. Above the cap, the app asks you to confirm rather than refusing.
On Pro, Email Sweep runs tags, saved rules and the AI checks over a whole folder or a selection of messages.
What gets sent
Only what the AI needs to answer. For each email:
- sender name and address, and the subject
- the message body, without hidden text, cut down to its first 3,000 characters and its last 1,500. Email addresses in it become
[email]and runs of nine or more digits become[number], on a best-effort basis - for tagging, quoted replies and signatures are left out. Phishing and spam checks read the whole message on your site, quoted replies and forwarded text included, but only the head and tail leave it. Anyone can type an “On Monday, Ann wrote:” line above a scam, so a check cannot trust one
- your tag names and descriptions, for tagging
- for phishing checks only, the text and domain of up to ten links, and attachment file names
What your site works out about the sender stays on your site: SPF, DKIM and DMARC results, the findings of its own checks, and what your mailbox holds from them (whether you wrote to them, replied, saved them as a contact or junked their mail). The AI judges the email. Your site weighs the rest.
Mailbox passwords, attachment contents and everything else stay on your site. Requests go from your WordPress server to our Worker at ai.curb.software, which uses TypeSafe to produce the answer. Your license key, site instance ID, site domain, plugin version and a request ID go along too, so the right balance gets charged. Those five do not go on to TypeSafe.
| Who | Gets | Keeps |
|---|---|---|
| Your site | Your mailbox | Everything. Only the fields listed above leave it |
CurbSoftware (ai.curb.software, a Cloudflare Worker) | The fields listed above, plus the license, site and request details | No message content. It records credit balances, a credit ledger and spends keyed by a hash, and one usage log line per request with the model, counts and tokens, but no text. Cloudflare may keep its own request logs for the service, which hold request details such as the sending address and headers but not the body |
TypeSafe (api.typesafe.ai) | Sender address, sender name, subject, the first 3,000 and last 1,500 characters of the body, your tags for tagging, and links and attachment file names for phishing checks only | Set by TypeSafe’s privacy policy |
TypeSafe’s policy says (as checked on October 10, 2026) that its services are hosted in the United States and that it will not train or fine tune models on your input. There is no EU-only option for PressedMail AI today. If mail has to stay in a region, use your own provider there, or leave AI off.
More on where your data goes: Where your mail and data live and the Privacy Policy.
If something looks wrong
A tag that never appears is usually a confidence set too high, or a tag the AI is not allowed to apply. A queue that sits still is usually out of credits or over a user’s monthly limit. For anything else, see AI, rules, and security controls.