Free USPTO federal register screen
Is my startup name trademarked?
A fast first-pass screen against the USPTO federal register — sound-alikes, look-alikes, and same or related classes, in plain English. No signup, no email, no cost.
Watch this name for new filings
Get alerted when a new USPTO filing appears that is confusingly similar to your name. We screen everything already on the register first, so old marks never alert — only filings that appear after you start watching.
Your watches
Your watches are tied to your email — nobody else can see them. Enter your email to get a 6-digit code, then enter the code to see and stop them. No password, no account.
Your results, side by side
Screen a name on the left — the USPTO verdict, domain availability, and EU & UK marks appear here together, no scrolling needed. One screen covers all three, free while we're testing.
Domains — is the name free to register?
.com, .io, .ai, .co, .app, .dev, .net, .org against the registries (RDAP). A free domain is not trademark clearance.
Outside the US — EU & UK marks
Same scoring, run against TMview (EUIPO, offices EM + GB). Best-effort: TMview sometimes refuses automated requests — if it does, this says so instead of failing.
This checks the USPTO federal register for sound-alikes, look-alikes, and same/related classes, plus domain availability and EU/UK marks above. It does not search common-law use, state registers, social handles, or non-EU/non-UK registers. A clear screen does not mean a name is safe to use or register. Get a trademark attorney for a full clearance before investing in a brand or filing (USPTO fee: $350 per class — fees change, see the official schedule). Data can lag live filings by days — very recent applications may not appear yet.
Common questions
Is this a trademark clearance?
No. Trademark Screen is a first-pass screen against the USPTO federal register. It flags names that look or sound similar in the same or related classes, in plain English. A clear screen does not mean a name is safe to use or register — a full clearance needs a trademark attorney.
Does it check state registrations or common-law use?
No. It checks the federal register only, plus domain availability and EU/UK marks. State registers, common-law use, and social handles are not searched.
How do trademark watches work?
When you watch a name, we screen everything already on the register first, so old marks never alert. Every Tuesday, after the USPTO Gazette publishes, we check again and email you only if a new filing appears that is confusingly similar at or above the alert level you chose.
Is it really free?
Yes. Screening is free while we're testing — no daily limit right now — and watches are free. No signup and no account — the watch list is private to your email, unlocked with a 6-digit code.
What do the risk levels mean?
High risk means a live registered mark is very similar in the same or a related class. Medium-high and medium mean a similar mark exists but the classes or the similarity are weaker. No conflicts found means no conflicting federal mark was found in this screen. The score is a similarity measure, not a legal opinion.
Do you store what I search?
No — a screen is not stored. We count screens only (how many, per day), never the name or what you sell. A watch is different: it has to remember the name and your email to check the register and alert you, so those are stored until you stop the watch — then they are deleted. See the Privacy Policy for retention and deletion.
How it works
1. You describe the name and what you sell. We work out the Nice classes your goods or services fall in. If we can't work them out from what you sell, we say so instead of guessing — pick the classes yourself and screen again.
2. We search the USPTO federal register for marks that sound alike, look alike, or share meaning, in the same or related classes.
3. Each candidate gets a similarity score — appearance, sound, and spelling, weighted by how closely the classes relate and whether the mark is live.
4. You get a plain-English verdict and the top hits, with the reasons, so you can see exactly what the score is based on.
Where the data comes from. The USPTO Trademark Search public backend — the same data the official search at tmsearch.uspto.gov uses. Records can lag live filings by days. Screen only, not a clearance. About the data.
How accurate is it?
We ran a 30-name spot check on Nov 2, 2026 (v0.5.9): 15 famous marks that should be flagged, and 15 invented names that should come back clear.
15 of 15 conflicts flagged HIGH. Every famous mark — NIKE, APPLE, COCA-COLA, MCDONALD'S, GOOGLE, AMAZON, SAMSUNG, ADIDAS, STARBUCKS, MICROSOFT, DISNEY, TOYOTA, PEPSI, NETFLIX, FACEBOOK — came back at the top score (1.0), with the famous mark itself as the top hit.
15 of 15 clear names came back with no conflict. Twelve had no hits at all; the other three had only weak hits below the threshold (best 0.35).
See the full 30-name list
Should conflict — all flagged HIGH (1.0): NIKE (clothing), APPLE (software), COCA-COLA (soft drinks), MCDONALD'S (restaurants), GOOGLE (software), AMAZON (retail), SAMSUNG (electronics), ADIDAS (footwear), STARBUCKS (coffee), MICROSOFT (software), DISNEY (entertainment), TOYOTA (vehicles), PEPSI (soft drinks), NETFLIX (streaming), FACEBOOK (social networking).
Should be clear — no conflict: NOVABREW 0, FERNWICK 0.35, DUSKTUNE 0, ZORBLAX 0, KELVOR 0, QUILTBERRY 0.30, VEXMONT 0, BRAMBLEQUILL 0, NIXOLARA 0, PORTHAVEN 0.30, JUMBLEWICK 0, TALLOWMERE 0, ORPHICAN 0, ZELMARA 0, WEXFORDALE 0.
Adversarial test — 10 names from an external review (Nov 2, 2026, v0.6.0). A reviewer ran 10 tricky names — typos of famous marks, the same word in different classes — and found two bugs: class detection defaulting to software classes, and typo variants missing vowel swaps. Both are fixed, and the re-run passes 9 of 10 as expected: NIKE, NIKEE, GOOGLE, STARBUCKZ, APPLE (software), GOOGEL (HIGH at 0.917 via GOOGLE), KLEANEX (HIGH at 0.869 via KLEENEX), DELTA airline (HIGH on Delta Air Lines) and DELTA faucets (HIGH on Delta Faucet). The tenth, APPLE for fresh fruit, came back medium-high (0.84) on APPEL, a live class-29 mark — not on Apple's software marks. That flag is under verification: if APPEL is a real live mark in the same class, the tool is right and the "coexists" expectation was incomplete.
Four more names from the same review ran the same day: FRESH BAKERY came back medium-high (0.794) on a real "FRESH BY MELAO BAKERY" mark; ZEN, ZORVANTLE and QUILMERA had unmapped goods ("consumer goods"), so the screen correctly returned "classes not confirmed" instead of a clear verdict.
New names, not tuned against (Nov 2, 2026, v0.6.0). Six typo names the tool had never been tested on: COCA KOLA (soft drinks), ADIDOS (footwear), MICROSOFTT (software), NETFLIKS (streaming), TESLLA (vehicles), GOGLE (search software). All six came back HIGH on the famous mark — COCA KOLA 0.931 on COCA-COLA, ADIDOS 0.908 on ADIDAS, MICROSOFTT 0.950 on MICROSOFT, NETFLIKS 0.868 on NETFLIX, TESLLA 0.917 on TESLA, GOGLE 0.915 on GOOGLE (GOGGLE 0.915 edges it by 0.001). Honest note: the first run was 3 of 6 — COCA KOLA, NETFLIKS and GOGLE were missed, because the variant generator lacked a K→C swap (COCA KOLA vs COCA-COLA), an X↔KS swap (NETFLIKS vs NETFLIX), and the double-letter expand (GOGLE vs GOOGLE) that an earlier change had wrongly dropped. All three were fixed the same evening and the retest passed 6 of 6.
Precision round (Nov 2, 2026, v0.6.0). Three invented names came back too hot: ZORVANTLE got Medium from lookalikes (ZORVANE, ZORVANYL, ZORVEM), FRESH BAKERY got Medium from unrelated "___ BAKERY" marks, and QUILMERA was mixed with QUILT-based marks. Three fixes: candidates now need a real near-match (edit distance 1–2 or a phonetic match) instead of just sharing a prefix; the verdict is driven by the strongest relevant hit — same or related class and similarity at least 0.5 — so a 0.4-similarity mark or an unrelated exact match cannot lift it to Medium; and descriptive words are down-weighted harder (an all-descriptive name caps at 0.40, below the Medium bar). Diagnostics after the fix: FRESH BAKERY is now clear (top hits 0.40, shown but not driving the verdict); QUILMERA's QUILT marks are gone, leaving only genuine 1-edit near-matches (QUIMERA 0.794, QUILTERA 0.789); ZORVANTLE's lookalikes are real live marks scattered across classes (ZORVANE in clothing and software services, ZORVYNIX in pharmaceuticals), so whether it flags depends on the class screened — which is the correct behaviour for a near-identical live mark. The fresh typo set re-ran 5 of 6 HIGH after the fix (COCA KOLA 0.931, ADIDOS 0.908, MICROSOFTT 0.950, NETFLIKS 0.868, TESLLA 0.917); GOGLE is pending the backend rate-limit.
Still to run: the invented-name precision set (VELTRAX, FERNWICK, NOVABREW and more), GOGLE, and the 30-name set re-run on the current code. The USPTO backend rate-limits this server, so the remaining tests go in small batches — this section will be updated with the new numbers.
The 30-name set above was run on v0.5.9; it is being re-run on v0.6.0, and this section will be updated with the new numbers. The USPTO backend rate-limits this server, so the re-run goes in small batches.
A 30-name spot check is proof the scorer works on obvious cases, not a guarantee. The register changes daily, and a name can be clear here and still collide with a common-law mark we don't search. The full test data is in the project's repository.