How to choose a brand name you can actually own

Most naming advice ends with "check the trademark and the domain". The checking is where names are lost, and it is usually done wrongly: someone types the exact word into a search box, sees nothing, and carries on. A conflict is not decided by exact spelling. This guide shows how conflicts are judged, how three jurisdictions differ, and how to test a shortlist so the decision rests on evidence. It is general information, not legal advice.

Why "is it available?" is the wrong question

In the United States, registration can be refused when a mark so resembles one already registered, or one used earlier and not abandoned, that it is likely to cause confusion about the applicant's goods (15 U.S.C. 1052(d)). Examining attorneys compare the marks' appearance, sound, meaning and commercial impression, and filings in the Office's own case files state that similarity in any one of these can be enough. Two things follow. A sound-alike can block you. And the question is always asked about goods: the same word can sit with different owners because the goods are unrelated.

Goods are sorted into the Nice Classification, 45 classes of which 34 are goods and 11 services. The classes are a search tool, not a wall. Apple Corps and Apple Computer coexisted under agreements that split their fields (a 1981 agreement, widened in 1991). They went back to court when Apple Computer moved into music, and in 2006 the High Court found its iTunes store within its agreed field of use. The two settled in 2007. The lesson for a founder: clear the classes you are in now and the ones you plan to enter within a few years.

Pick the kind of name by the risk you carry

The US courts' scale comes from *Abercrombie & Fitch v. Hunting World* (2nd Circuit, 1976): fanciful, arbitrary, suggestive, descriptive, generic. Fanciful and arbitrary names are registrable at once. Descriptive ones need proof that the public knows them as a brand, and the Office may accept five years of substantially exclusive and continuous use as prima facie evidence (1052(f)). Generic names get nothing.

Names can also die of success. In *Elliott v. Google* (9th Circuit, 16 May 2017) the court held that GOOGLE had not become generic even though people use it as a verb. The test is whether the public primarily understands the mark as naming the producer or the category, for a particular kind of goods. The practical habit is to keep a generic noun beside your name ("Nova app", not "a Nova") so it never becomes the word for the product.

Search by sound, in the right classes

A knockout search is a fast first pass to eliminate names. Do it in this order:

1. Choose the classes. For a software or design business, classes 9 (software), 35 (advertising and business) and 42 (computer and scientific services) are typical starting points. Check the headings on the Nice pages for yours.
2. Generate the sound-alikes. Apply the rules below to each name.
3. Search each variant in the free WIPO Global Brand Database (international filings plus many national collections, with image search since 2014), in the national office of each market, and on the web and app stores. Unregistered use counts in the United States, which is why a plain web search matters.
4. Record owner, class, status, date and why it is or is not close. A shortlist without notes will be argued again in week three.

Rules like these give 15 to 23 variants for a four to six letter name. Search them all. The list on the picture below is our output for a made-up word, not a clearance.

The rules differ by market, and that changes the order of work

Details, with the sources we could read:

• EU. The General Court has restated that it is enough for an absolute ground to apply in part of the Union for refusal (Article 7(2) of Regulation 2017/1001). In the TOTALLY HANDS FREE case the EUIPO Board treated the English-speaking public as including states such as Denmark, Cyprus, the Netherlands, Finland and Sweden. The applicant disputed that and the court did not decide it.
• Indonesia. The system is first to file under Law No. 20 of 2016. Secondary sources say a bad-faith application is refused (Article 21) and a bad-faith registration can be invalidated in the Commercial Court without a deadline (Article 77). To oppose an application you normally need an earlier Indonesian application or registration, so a foreign owner without a filing has a weak position. Protection lasts ten years. A registered mark could be cancelled by a third party after three years of non-use. A Constitutional Court decision, number 144/PUU-XXI/2023, read that as five years, reported at the end of July 2024. Older sources say three. We could read only summaries of the ruling, so confirm the period with a local agent.
• Madrid. An international registration depends on the home application for five years. If that falls, all designations fall, though a WIPO working paper says third-party attacks on the home mark are rarely used in practice, and the owner can convert to national filings within three months.

Sound and fluency: useful as a tie-break

Pronounceability has measurable effects. Alter and Oppenheimer (PNAS, June 2006) found that stocks with easy-to-pronounce names and tickers did better right after listing, and that the effect faded within months as other information arrived. Do not trade distinctiveness for fluency. Sound also carries size. Klink (2000) found that names with front vowels read as smaller and quicker, back vowels as larger and slower; Yorkston and Menon (2004) showed that a made-up ice cream "Frosh" was rated creamier than "Frish". Lowrey and Shrum (2007) found front vowels preferred for sports convertibles and back vowels for SUVs. We saw these only through summaries and the studies are laboratory ones, so use them to choose between two otherwise equal names.

Meaning in other languages deserves a native speaker, and a sense of proportion. The famous claim that the Chevrolet Nova failed in Spanish-speaking countries because "no va" means "doesn't go" is a legend: Snopes notes that the Nova was sold under that name in Mexico and Venezuela between 1972 and 1978, and a linguist's post found no evidence it hurt sales. Real conflicts are legal and phonetic far more often than semantic.

Test with people, and know what a small test can show

Say the name once to each person, ask what it sells, and ask them to write it down the next day. Then type each spelling into a search box and watch the suggestions. The sample size sets what you can conclude:

We computed these in Python: the one-sided bound for zero failures is 1 minus 0.05 to the power 1/n, and the interval for failures is a Wilson interval. Ten people only find gross problems. The rule we use is a decision rule, not a statistic: drop any name that three of ten people misspell, then confirm the favourite with twenty.

Score the shortlist

Gates first, scores second. A name that fails a gate is out whatever it scores.

Gates, pass or fail

1. No identical or sound-alike mark in your classes in your markets (knockout search).
2. No problem meaning in your main languages, checked by a native speaker.
3. A domain and the main handles you need are possible without hyphens or numbers.

Scores, 0 to 2 each (our rubric, not a validated scale)

Pick a pass line before scoring, for example 9 of 12. The Letterforms line is real: Weyve's page describes the first letters of the name as one continuous line written with a pen of a single weight, and that line returns much larger behind everything the brand makes. A name is also a drawing. See Urbanova and Verden for names set as wordmarks beside their marks.

Where it goes wrong

Once the name is cleared, the strategy page (brand strategy for small business) and the difference between a logo and an identity decide what it has to carry. If you are choosing between names or weighing a rename, see when to rebrand, or talk to us.