An AI email organizer that sorts your inbox without reading it
Updated · ClearInbox team
Plenty of tools sold as an AI email organizer now send your messages through a large language model. ClearInbox doesn’t. Its automatic email categorization is deterministic: a fixed set of rules reads headers and metadata (sender domain, mailing-list headers, automated-sender signals, subject, age and size) and places every message in one of thirteen smart categories. There is no model to read your mail, no model to train on it, and no guessing about why a message landed where it did.
Is ClearInbox’s AI email organizer actually AI?
Not in the machine-learning sense. ClearInbox has no language model and no trained classifier. It sorts mail with deterministic rules applied to headers and metadata, so the same message always lands in the same category. We use the “AI” label because that is what people search for, and we would rather say plainly what it means.
That choice was deliberate. A learned model is good at nuance and bad at explaining itself; a rule is the opposite. For inbox cleanup, where the output is “delete 2,000 of these”, explainability matters more than nuance. You should be able to look at a category and know exactly what qualifies a message for it.
It also means the sorting does not drift. Gmail’s own category tabs, for example, learn from the messages you drag between tabs, which is useful but means two people’s tabs behave differently over time. ClearInbox’s rules are identical for everyone and every account until the ClearInbox team changes them.
How does ClearInbox decide which category an email belongs to?
Each message runs through checks in a fixed order. First the sender’s domain is compared with curated lists of social, shopping, food delivery, travel, rideshare and finance senders. If none match, the subject line is checked for telltale wording. Then mailing-list headers, then automated-sender headers. Anything left is Personal. Age and size are added on top.
The first match wins, which keeps the result predictable: an order confirmation from a retailer’s domain is Shopping even if its subject says “sale”. Old Emails and Large Emails are overlays, so a three-year-old 8 MB airline confirmation shows up in Travel, Old Emails and Large Emails at once. None of these checks touches the message body.
| Smart category | Signals used to assign it | Typical example |
|---|---|---|
| Social Networks | Sender domain on the social list (Facebook, Instagram, LinkedIn, Reddit, Discord and others) | “You have 3 new connection requests” |
| Shopping | Retailer sender domain, or an order, shipping, tracking or receipt subject line | Order shipped notice from an online store |
| Food Delivery | Sender domain of a delivery service (Uber Eats, DoorDash, Deliveroo and others) | Delivery receipt |
| Travel | Airline, hotel or booking-site domain, or an itinerary, boarding pass or reservation subject | Booking confirmation |
| Rideshare | Sender domain of a ride or scooter service (Uber, Lyft, Bolt and others) | Trip receipt |
| Promotions | Subject wording such as sale, % off, coupon, promo or limited time | “Flash sale ends tonight” |
| Finance | Payment or banking domain, or a statement, payment, transaction or balance subject | Monthly statement ready |
| Newsletters | A List-Id header, or a List-Unsubscribe header on a non-transactional message | A weekly digest you signed up for |
| Notifications | Subject wording such as alert, reminder, password reset, security code or new login | Sign-in alert |
| Automated Messages | Auto-Submitted or Precedence: bulk headers, bulk-mailer signatures, no-reply style addresses | System report from a service |
| Old Emails | Date older than one year (added alongside the main category) | Anything from before last year |
| Large Emails | Message size over 5 MB (added alongside the main category) | A thread with photo attachments |
| Personal | No match above: a human sender with no bulk or automated signals | A note from a friend |
Why is rule-based sorting better for privacy than an LLM organizer?
Because it needs far less of your mail. An LLM-based organizer usually has to read message bodies, often by sending them to a model provider. Rule-based sorting needs only the envelope: who sent it, which list it came from, the subject, size and date. ClearInbox stores that metadata and a short preview snippet, never full message bodies.
Less data in means less that can leak, be retained or be repurposed. There is no model provider receiving your mail and no training pipeline, so the promise that your mail is never used to train AI is structural rather than a policy line that could change. Credentials are encrypted at rest with AES-256-GCM, and the app ships no third-party analytics or trackers.
There is a security angle too. A model that reads your mail can be steered by text inside a message, the attack OWASP lists first among LLM risks as prompt injection, with an email assistant among its examples. A rule that checks a List-Id header cannot be talked into anything.
To be fair to the other approach: language models can do things rules cannot, such as summarizing a long thread, spotting that a message needs a reply, or drafting one. ClearInbox does none of that. If those are what you want from AI inbox management, an LLM tool is the right category; just read where your message bodies go.
What do the one-click cleaning suggestions look at?
Suggestions are counts over the same categories and sender statistics. ClearInbox looks for unread promotions older than 30 days, emails larger than 5 MB, mail older than a year, newsletters you mostly leave unread, and noisy automated senders where keeping only the newest message makes sense. Each suggestion states the count and size before you act.
Nothing happens until you click. A suggestion to delete old promotions sends them to your provider’s Trash, where they stay recoverable until it is emptied, and every action is written to the activity log. If you like a suggestion enough to want it permanently, turn it into an Auto Clean rule and it will run on a schedule and on newly synced mail.
Does the same smart email sorting work beyond Gmail?
Yes. Because the rules only need headers, they behave the same on every provider. You can organize Gmail inbox clutter, a work Outlook or Microsoft 365 account, and Yahoo, iCloud, Fastmail or any IMAP mailbox with one set of categories, side by side in one dashboard. Gmail’s own tabs keep working; ClearInbox adds its view on top.
The practical benefit is that a rule written against a category, say “archive Notifications after two days”, means the same thing in every account. How to design that wider system is covered on the email organizer page; this page is only about how the sorting itself works.
Frequently asked questions
Does the AI read my emails?
No. There is no model reading anything. Classification uses headers and metadata only: sender, subject, size, date and mailing-list headers. ClearInbox stores that metadata plus a short preview snippet, never full message bodies, and your mail is never used to train AI.
Is there any machine learning in ClearInbox?
No. Categories and cleaning suggestions come from fixed, rule-based checks that the ClearInbox team maintains. The same message will always get the same category, which is what makes the results predictable and easy to audit.
What happens if a sender lands in the wrong category?
The curated domain lists cannot know every small shop or local service, so some mail lands in a broader bucket such as Newsletters or Personal. Write an Auto Clean rule on that sender or domain and it decides what happens to that mail, whatever its category.
How is this different from Gmail’s category tabs?
Gmail sorts into Primary, Social, Promotions, Updates and Forums using Google’s own signals and learns from your corrections. ClearInbox’s thirteen categories use published, fixed rules, work identically across every provider, and feed directly into bulk actions and rules.
Can the organizer act on mail by itself?
Only when you tell it to. Suggestions wait for a click. Auto Clean rules you create run on their own, and every action they take is recorded in the activity log. Pinned messages are protected from automation entirely.
Sources and further reading
- RFC 2919: List-Id, an identification header for mailing lists
- RFC 2369: URLs as meta-syntax for mailing-list commands (List-Unsubscribe)
- RFC 3834: Recommendations for automatic responses to email (Auto-Submitted)
- Gmail Help: Organize your emails into categories
- OWASP Top 10 for LLM Applications: LLM01 Prompt Injection