AI, Explained
Plain keyword rules are great, but they can't tell urgent from routine, or a real person from a sales blast. That's what MailPrism's AI adds: it reads each email and turns its understanding into signals you can act on — in rules, in tracking, and across your inbox.
When you sign up, you agree to let MailPrism use AI providers (Google Gemini as the primary provider, with OpenAI as the fallback) to analyze your email. You can turn AI off at any time in Settings → Privacy → AI Data Processing. While it's off, nothing is sent for AI analysis. See AI privacy & consent.
AI analysis is included on paid plans (Starter and up). The Free plan runs rule-based automation. Bring-your-own-key (BYOK) is available on Business, Enterprise, and Lifetime.
The one exception: the AI rule drafter works on every plan — Free plans get 10 AI drafts a day.
What the AI does
When AI is on (and your plan includes it), MailPrism reads each incoming email and produces a small set of structured signals — a category, an urgency level, a sentiment, and a handful of yes/no flags. Those signals become available everywhere a decision needs to be made:
- In rules — every signal is a condition you can match on, so automations can react to what an email means, not just what words it contains.
- In response tracking — AI helps decide which threads need your action or are awaiting a reply. See Response tracking.
- In the inbox — the analysis is surfaced on the email itself, so you can see at a glance why something was flagged.
The signals at a glance
For each email, MailPrism can detect:
| Signal | Values | Good for |
|---|---|---|
| Category | What kind of email it is (see the full list) | Sorting and routing |
| Urgency | high medium low | Surfacing what needs you now |
| Sentiment | positive neutral negative | Spotting unhappy senders early |
| Is spam | true / false | Extra spam filtering |
| Is automated | true / false | Separating machine mail from people |
| Is cold outreach | true / false | Filtering unsolicited sales pitches |
| Needs response / action | true / false | Driving follow-ups and tracking |
Each signal also carries a confidence score, so your rules can require the AI to be sure before acting. For the full breakdown of every signal and its meaning, see Classification signals.
Where AI shows up
| Area | What AI contributes |
|---|---|
| Rule conditions | Match on category, urgency, sentiment, spam, automated, cold outreach, needs-response, and more — see Conditions reference |
| Response tracking | Detects threads that need action or are awaiting a reply |
| AI Insights | A dashboard widget surfacing recent AI activity — see AI Insights |
| AI Categories settings | Create your own custom categories and train the AI from corrections |
Which providers are used
By default, AI analysis runs through MailPrism's managed AI. Google Gemini is the primary provider, with OpenAI as the fallback — MailPrism selects the provider and model for each task, so you don't choose one. None of these providers use your data to train their models.
If you'd rather use your own account and API keys, you can bring your own — but BYOK supports OpenAI and Anthropic only (Gemini isn't available as a BYOK provider). See Bring your own key (BYOK).
MailPrism picks an appropriate model for each task. Specific model names can change as providers release new versions, so we don't list them here — what stays stable is the signals the AI produces.
A note on accuracy
AI is excellent on typical email but not perfect. Sarcasm, unusual phrasing, and highly technical content can be misread. For important automations, combine an AI signal with a concrete condition (like sender relationship). See Best practices.
→ Next: Classification signals · Custom categories · AI privacy & consent