How the Trademark Check works

The Trademark Check screens a name against the live US (USPTO) and EU (EUIPO) trademark registers with exact, phonetic, and fuzzy matching, weighted by the Nice classes your product would file in. The US (USPTO) check is free on any name, no login and no card; the EU (EUIPO) register is added on the paid check. This page documents the mechanism end to end: where the data comes from, how matching is scored, how results are cached, and where the method stops.

Updated August 2026

The sources: two live registers, named

Every check searches the United States federal trademark register maintained by the USPTO and, on the paid check, the European Union trade mark register maintained by the EUIPO. Candidate marks are retrieved live at search time, either through a trademark data service that queries both offices in one normalized call or directly through the offices' own interfaces, the USPTO Open Data Portal and the EUIPO Trademark Search API. The retrieval step is deliberately broad: it pulls every plausibly close mark, not just exact spellings, so the scoring step has real material to work with.

Marks that are no longer live are dropped before scoring. A registration that has been withdrawn, abandoned, cancelled, expired, or refused does not block your name the way a live mark does, so it does not belong in a knockout result. Pending applications stay in, because a pending application that registers later can block you retroactively from your perspective: you would have built on a name someone else was already claiming.

The matching: exact, phonetic, fuzzy

Registers protect against confusion, not just duplication, so an exact-spelling lookup misses most real conflicts. Each retrieved mark is compared to your name three ways, and the three signals combine into a single closeness value between 0 and 1.

• Exact. Both strings are normalized (lowercased, punctuation and spacing stripped) and compared. An identical string scores 1.0, the tightest possible collision.
• Phonetic. Both strings are encoded with Double Metaphone, the standard phonetic algorithm, and compared by sound. This is what catches “Klariti” against “Clarity” or “Fonik” against “Phonic”: different spellings, same name to the ear, which is how examiners and courts hear them too.
• Fuzzy. Jaro-Winkler string similarity, which weights shared prefixes the way brand recognition does, plus Levenshtein edit distance, which catches short names one or two characters apart. A six-letter mark one edit away from yours is a genuine collision candidate even when it neither looks identical nor encodes to the same sound.

The thresholds are conservative in both directions. A mark below 0.55 closeness is not reported at all, because burying you in weak look-alikes would hide the matches that matter. At the top end, a mark at or above 0.92 closeness inside your own class is flagged as a likely conflict. Between the two sits the caution zone: real similarity worth reading before you commit, not an automatic disqualifier.

Nice-class weighting

Trademark rights are granted per class of goods and services, the 45-class Nice system both registers use, so the same closeness value means different things in different classes. The check infers your likely classes from what you tell it you are building: software lands on classes 9, 42, and 35 as the floor, and the description adds more, class 36 for fintech, 44 and 5 for health, 41 for education, 39 for logistics, and so on. Every collision is then weighted by class overlap: a match inside your inferred classes counts at full weight, and a match in a distant class is discounted to under half. This is why a sound-alike selling something unrelated shows up as caution rather than conflict, and the same sound-alike in your own category can stop the name. The page on identical versus confusingly similar marks walks through worked examples of exactly this distinction.

What comes back

The check rolls the scored collisions up into a band and a 0 to 100 score, and reports the closest matches individually: the mark text, which register it sits in, its Nice classes, its status, its owner where the register provides one, a short description of the goods and services it protects, and a deep link to the official record at the USPTO's TSDR or EUIPO's eSearch so you can read the primary source yourself. The bands mean exactly this:

• Green. No direct or confusingly similar live mark surfaced in the registers and classes searched, on the date searched.
• Yellow. Similar marks exist, in distant classes or below the likely-conflict threshold. Read them before you commit.
• Red. A likely conflict: a live mark at or above the top threshold inside your own classes.
• Unknown. No register could be reached, so nothing was screened and no verdict is claimed.

What is free and what is paid

Generating names and checking .com availability are free, with no login to start. The US (USPTO) trademark check is free on any name, no login and no card. The EU (EUIPO) register is the paid upgrade: every paid check screens the US and EU registers together. The $49 Launch Pack covers 40 full US and EU checks, top-ups run $19 for 10, $39 for 25, and $79 for 60, Pro is $149 per month for 150 checks with API and MCP access, and Ultimate is $479 per month for 1,000 checks. Full detail sits on the pricing page, and developers can call the same engine through the MCP server and HTTP API.

Freshness, caching, and the honesty rules

Results are dated. Every report carries the date its search ran, because registers change daily and an undated verdict slowly turns into a false one. A repeat check of the identical name, class set, and scope can be served from a cache for up to 60 days; anything that changes the question, a different name, a different class set, or upgrading from a US-only to a US plus EU screen, is a fresh search. A US-only result is never silently reused to answer a US plus EU request.

The failure behavior is the part most tools leave unstated. If a register does not respond, the check reports that register as not screened rather than guessing, the overall band becomes unknown rather than green, and the failed result is never cached. On a free check, the EU line reads “not screened on this check” because the EU register is the paid upgrade, which is a different statement from unreachable, and the report says so in those words. The rule underneath all of this: the check never reports a clear result for a register it did not search.

What this method does not do

This is a knockout search, built to eliminate obviously blocked names early, and that is the boundary. It does not search unregistered common-law marks, company registers, US state trademarks, or the national registers of individual EU member states. It compares words, not logos, so figurative similarity is out of scope. It does not weigh evidence of use, reputation, or the full legal factors an examiner or court applies. And it is not legal advice, not a clearance opinion, and never a statement that a name is free to use: full clearance before filing is a job for a trademark attorney. What it gives you is the fast, honest first pass: the same search you would run by hand across the USPTO and the EUIPO, done in seconds, with the method in the open.

Questions, answered

Related: what a knockout search is, how to check a name by hand, identical vs confusingly similar, and searching the EU register.

Trademark results are an automated database search against the USPTO and EUIPO registries, not legal advice and not a clearance opinion. Registries change daily; results are dated. Before filing, have counsel run full clearance.