Custom Categories
The built-in categories cover most inboxes, but yours is unique. In Settings → AI Classifications you can create your own categories, adjust the built-in ones, and train the AI to match how you think about email.
Creating custom categories and editing the built-in (system) ones requires a Pro, Business, Enterprise, or Lifetime plan. Everyone can view their definitions and accuracy metrics. For current plans, see the in-app pricing page.
The manager is organized into tabs:
| Tab | What it does |
|---|---|
| Definitions | Your categories — built-in and custom |
| Discover | AI scans your inbox and suggests brand-new categories |
| Optimizations | Suggestions to improve accuracy, learned from your corrections |
| History | Your past classification corrections |
| Metrics | How accurate the classifications have been |
Definitions
Each category is shown as a row with its icon, name, and description. Badges tell you what kind it is:
- System — a built-in MailPrism category.
- Customized — a built-in category you've edited.
- Disabled — turned off and not used for classification.
From here you can Edit any category, Reset a customized one back to its built-in defaults, or Delete a category you created. (Built-in categories can be edited or reset, but not deleted.)
Creating a category
Choose Add Category and fill in the form. A category has these parts:
Basic information
| Field | What it's for |
|---|---|
| Classification type | Whether this defines a category, an urgency level, or a sentiment |
| Display name | The human-friendly name (e.g. Important Client) |
| Category key | A short machine name, auto-generated from the display name |
| Icon & Color | How it looks in the app |
| Description | A short line shown in the UI |
AI definition
This is what actually teaches the AI when to apply the category.
| Field | What it's for |
|---|---|
| Definition prompt | A clear description of when an email belongs in this category. Be specific. |
| Keywords | Comma-separated words that hint at this category (e.g. urgent, deadline, priority) |
| Sender patterns | Comma-separated email patterns (e.g. @important-client.com, ceo@) |
Learning examples
| Field | What it's for |
|---|---|
| Positive examples | Subjects that should match (one per line) |
| Negative examples | Subjects that should not match (one per line) |
Behavior settings
| Setting | What it does |
|---|---|
| Confidence threshold | The minimum AI confidence (50–95%) required before this category is applied |
| Priority | Higher-priority categories are evaluated first (1–100) |
| Enabled | Whether the category is used when classifying email |
The definition prompt is the most important field. Describe the intent — "emails from clients about active projects that need a timely reply" — not just keywords. Keywords, sender patterns, and examples sharpen it.
Once saved, your custom category behaves like any other: it can drive rule conditions and appears in your metrics.
Discover: suggested categories
The Discover tab asks the AI to look at a sample of your recent inbox and propose new categories you don't have yet. Each suggestion comes with evidence — example emails, the patterns it noticed, and the domains involved — so you can judge it before accepting.
Discover runs analyze a batch of emails and are subject to monthly run limits and a cooldown between runs (shown in the tab). Limits depend on your plan.
Optimizations: learning from corrections
When you correct a classification — for example, recategorizing an email — MailPrism remembers it. The Optimizations tab scans your correction history for repeating patterns and suggests concrete improvements, such as adding a keyword or sender pattern to a category.
For each suggestion you can Apply it (the change is made to the category) or Dismiss it. Use Find Patterns / Refresh to re-scan.
The system needs at least a handful of similar corrections before a reliable pattern emerges — so suggestions appear once you've given it something to learn from.
History
The History tab lists your past corrections. This is the raw material the Optimizations tab learns from, and it's where corrections can be reviewed.
Metrics
The Metrics tab shows how the classifications are performing:
- Accuracy rate, emails classified, average confidence, and corrections this month at a glance.
- A confusion view showing which categories get mixed up with which.
- A weekly accuracy trend.
- Accuracy by category, with an up/down trend arrow per category.
The more you correct, the better these numbers get — your feedback feeds directly back into the AI's accuracy.
→ Next: Smart rules · Classification signals