Identical vs confusingly similar

A trademark does not have to be identical to yours to block you: both the USPTO and the EUIPO refuse marks that are confusingly similar to an earlier one, judged on sound, appearance, and meaning together with how related the goods are. This is the single most misunderstood fact in startup naming, and it is why an exact-spelling search feels thorough while missing the conflicts that end up in oppositions. Here is the distinction, with worked examples.

The standard is confusion, not duplication

Trademark law exists to stop consumers from mistaking one source for another, so the question an examiner asks is not “is this string already taken” but “would a consumer seeing this mark on these goods think it comes from the earlier company”. In the United States that runs through the likelihood-of-confusion factors, with the similarity of the marks in sound, appearance, and commercial impression weighed against the relatedness of the goods and channels. In the EU it is framed as a global assessment, but the substance is the same. Two consequences follow. An identical string in an unrelated category can be fine. And a different string in your own category can be fatal, if it sounds or reads like the earlier mark.

Worked examples, from tightest to loosest

The names below are invented for illustration; the collision patterns are the ones that recur in real refusals.

• Identical, same class. You want “Lumora” for a SaaS analytics product; a live registration for “Lumora” covers downloadable software. This is the clearest possible conflict and the fastest disqualifier. No judgment call is required, which is why exact matching alone feels sufficient right up until it is not.
• Phonetically identical, spelled differently. You want “Klariti” for a project management tool; the register holds “Clarity” for business software. Out loud the two are the same word, and confusion is judged by ear as well as by eye: a customer who hears the name in a podcast cannot see your spelling. Sound-alike conflicts in the same class are refused routinely, and no exact-match search will ever surface them.
• One edit apart. You want “Vantik” for payments infrastructure; the register holds “Vantic” for financial software. Short names one or two characters apart look near-identical at a glance, and at six letters a single character is a seventeen percent difference in spelling and roughly zero difference in impression. The shorter the name, the harder near-spellings collide.
• Similar, distant class. You want “Orvella” for developer tooling; a registration for “Orvela” covers cosmetics. Here the similarity is real but the categories are unrelated, the channels do not cross, and coexistence is the normal outcome. This is the case where a naive similarity search over-warns; the class scoping is what separates a caution from a conflict, as the page on using a name trademarked in another industry explains.
• The reach of famous marks. The exception to the class logic: a famous mark is protected against dilution well beyond the categories it trades in. Naming your accounting tool after a household-name brand fails even though the classes never touch. If everyone has heard of it, treat it as blocked everywhere.

How a real search implements this

The distinction above dictates the method: any search that stops at exact spelling answers the duplication question, not the confusion question. A search built for confusion runs three passes, exact spelling, phonetic encoding so “Klariti” and “Clarity” collide, and string distance so “Vantik” and “Vantic” collide, then weights every hit by class overlap so “Orvela” the cosmetics brand is discounted rather than treated like a blocker. That is precisely what the Trademark Check here runs, with the thresholds documented in the open on the methodology page: the US (USPTO) check is free on any name, and the EU (EUIPO) register is added on the paid check. Running the same grid by hand is entirely possible, and the knockout search guide shows how; the tool exists because doing it rigorously for a whole shortlist is an afternoon of queries.

Where the line sits, and who draws it

None of the examples above are mechanical rules. Similarity is a spectrum, examiners weigh the whole factual picture, and close cases go either way on evidence about markets and consumers that no register search contains. What a founder controls is the distance: names that need a lawyer to argue they are different enough are expensive names, and names with no close neighbor in their class are cheap ones. Screening for phonetic and near-spelling conflicts early is how you buy that distance before you are attached to the name. What a screen does not do is draw the legal line: an automated registry search surfaces the neighbors, it is not legal advice, and it never certifies that a name is free of conflicts. The judgment on a close call belongs to a trademark attorney, made once, on the finalist, after the obviously blocked candidates are already gone. The full checking guide covers that sequence end to end.

Questions, answered

Related: what a knockout search is, the free US trademark search, and trademark vs domain vs company name.

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.