How MarkAlarm works
The whole method, in the open — a monitoring service you can't inspect isn't worth trusting.
What we watch
Every business day the USPTO publishes its new trademark application data. MarkAlarm ingests it and extracts each application's mark text, owner, filing date, and international classes, then scores every one against the marks you're watching.
What we don't watch
State trademark registers, common-law (unregistered) use, business-name/DBA registries, foreign filings, and design-only marks with no word element. We monitor federal word marks in the USPTO application data — so you always know exactly what an alert means.
How similarity is scored
Each new filing gets a 0–100 score from four signals: spelling (Jaro-Winkler edit distance on normalized text), sound (double-metaphone phonetics, so KOZY and COZY collide), word overlap (shared significant tokens), and class overlap (same international classes as your watch). Class overlap weighs heaviest: a similar name in your class is the real risk, while an identical name in an unrelated class usually isn't.
Alert tiers
High (85–100) and medium (70–84) trigger an immediate email. Low (55–69) doesn't email you on its own — it's counted in your weekly summary so you know it was seen; below 55 isn't reported. Every alert shows its full breakdown, so you can see exactly why it fired and tune or remove the watch in one click.
False positives and negatives
No automated matcher is perfect. We tune toward catching real collisions even at the cost of an occasional loose match — which is why every alert is explainable and adjustable. A similarity score is information, not a legal conclusion about "likelihood of confusion," which is an analysis for an attorney.
A sample alert
Illustrative example with a fictional filing. Real alerts link the live USPTO record.