Startup naming guide
How to Come Up With a Company Name
March 2026
9 min read
Coming up with a company name is the part of building a company most founders underestimate, then get stuck on longer than anything else. This is the practical guide: six approaches to generating candidates, five tests every shortlisted name must pass, and the single mistake that kills most founder-generated names before they reach a single customer.
Before you generate anything
The most common naming mistake is starting with names. Before you write a single candidate, answer three questions. Your answers will eliminate the wrong approaches and give every name you generate somewhere to aim.
What position does this name need to occupy? Is the company a precision tool, an accessible companion, a disruptive challenger, or an authority figure? A name like Stripe occupies a completely different position than a name like Hims. Both are good names. They would not be interchangeable.
What is the primary context? A name that will live in developer documentation has different requirements than a name that will appear on pharmaceutical packaging or a consumer app. The contexts where your name will actually appear -- business cards, URLs, spoken in meetings, printed on contracts -- determine which properties matter most.
What does the name not need to do? Many founders try to make the name do too much: describe the product, communicate the mission, hint at the technology, and be memorable. Names that try to say everything say nothing. The best brand names create a single strong impression and let the company fill in the rest over time.
"The best brand names do not describe the company. They create the impression the company is designed to leave."
The six approaches to generating company name candidates
No single approach produces consistently excellent names. Founders who end up with strong shortlists typically run multiple approaches in parallel and combine the outputs.
Approach 1
Category pivot
Name the adjacent metaphor rather than the product. Instead of naming what you do, name what you replace, what feeling you create, or what you make possible. Slack named a feeling -- the loosening of workplace friction. Stripe named a visual property of something streamlined and fast. Neither name describes payments or messaging software.
How to apply: write down the three most common adjectives your best customers use to describe your product. Generate names from those adjectives rather than from the product category itself.
Approach 2
Morpheme construction
Build names from meaningful parts rather than whole words. Most durable invented brand names are not random letter combinations -- they are constructed from roots that carry recognizable weight even when the word is new. "Anthropic" uses Greek roots. "Vercel" combines the velocity of "ver-" with the clean Latin suffix "-cel." "Figma" assembles a functional consonant cluster with a minimal vowel profile.
How to apply: list the five core concepts your product embodies. For each, find the Latin, Greek, or Germanic root. Combine roots in pairs and listen to what the combinations sound like when spoken aloud.
Approach 3
Single-word abstraction
Find a common word that is off-category for your industry but carries the right impression. Zoom is a word from aviation and cinema. Notion is a philosophical term. Linear is a mathematics term. Each was "borrowed" from outside the software industry and applied to software products. The off-category origin is part of the value -- the word arrives with existing associations that transfer to the brand.
How to apply: for each property your product has (fast, precise, clear, fluid, expansive), generate a list of common English words that carry that property in a completely different domain. Every word on that list is a naming candidate.
Approach 4
Phoneme-first construction
Start with sound properties rather than meaning. Decide what the name should feel like when spoken -- sharp and quick (plosive consonants, short vowels), warm and approachable (nasals and liquids, open vowels), or precise and minimal (fricatives, closed vowels). Then build invented words that carry those sound properties. This is the approach large naming agencies use for pharmaceutical and consumer brand names, where the visual and semantic associations must be built from scratch.
How to apply: type three existing company names in your category that you admire into a phoneme analyzer. Note the shared sound properties. Use those properties as constraints when generating new invented names.
Approach 5
Constraint narrowing
Set explicit structural constraints and generate within them. Constraints produce variety, not uniformity -- a rule like "must end in a vowel" or "must be two syllables" creates a focused generation space that produces names a blank-slate session never would. Lexicon, the agency behind Pentium, Febreze, and PowerBook, begins most engagements by identifying what the name must not be (not descriptive, not metaphorical, not category-adjacent) before generating what it should be.
How to apply: write three explicit constraints (structural, phonetic, or semantic) before your generation session. Examples: "No more than 3 syllables," "Must not contain the word 'smart' or 'flow' or 'hub'," "Must be a word that exists in English but is off-category."
Approach 6
Category audit and inversion
Map every naming convention in your competitive space, then deliberately invert two or three of them. If every company in your space uses a two-syllable made-up word ending in "-ly" or "-io," do not do that. The name that stands out in a category is usually the one that breaks a convention that everyone else is following unconsciously. Stripe, in a payments landscape full of "Pay-" and "Square"-style names, did this by being a single common word with no payments connotation at all.
How to apply: list 20 competitors. Find the three most common naming conventions (structure, suffix, semantic territory). Treat each as a constraint to avoid.
The six types of company names
Every company name falls into one of six structural categories. Each has different tradeoffs for trademark clearance, memorability, and brand flexibility.
Real word
Notion, Stripe, Linear
Strong recall. Harder trademark clearance. Works best when the word is off-category.
Invented word
Figma, Vercel, Anthropic
Strongest trademark protection. Requires phoneme design to carry meaning through sound.
Altered spelling
Fiverr, Tumblr, Flickr
Era-dated quickly. Creates persistent "how do you spell that?" friction. Avoid unless domain-driven.
Compound word
Salesforce, Dropbox, HubSpot
Descriptive and clear. Limits brand expansion as company scope grows. Common in B2B SaaS.
Acronym / initialism
IBM, SAP, GE
Works only after the company is large enough that the acronym carries recognition. Do not start here.
Founder / place name
Ford, Dell, Goldman Sachs
Strong for services businesses where founder reputation is the brand. Limits sale or transition.
The category that produces the most durable names for modern startups is the invented word with conscious phoneme design. It maximizes trademark clearance probability, eliminates semantic constraints on future brand expansion, and gives the company complete control over the associations the name accumulates over time.
The mistake that eliminates most founder-generated names
The most common naming mistake
Anchoring on a name before testing it in context
Founders generate a name they like, check that the .com is available, and stop. The problem surfaces months later: the name is difficult to pronounce consistently, creates a trademark conflict, sounds like a competitor in a different language, or becomes a different word when spoken at normal speech speed ("Rate Us" becomes "Ray-tus"). The test that catches all of these is simple: put the name in the 10 actual contexts where it will appear and read it out loud in each one. Most shortlisted names fail at least two of these tests before anyone has spent a dollar on them.
The five tests every name candidate must pass
Run these tests on every name before it reaches the final shortlist. The order matters -- later tests are more expensive, so eliminate early and cheaply.
01
The spoken clarity test
Say the name to five people and ask them to spell it back. If more than one person writes it differently from how it is spelled, or hesitates on how to pronounce it, it will create friction at every touchpoint where it is heard before it is seen. This catches altered spellings and phonemically ambiguous invented words.
02
The context performance test
Write the name in five real contexts: a WSJ headline, a TechCrunch launch sentence, a business card job title line, a spoken sentence in a sales call ("We use [name] for our billing"), and a dotcom URL. A name that reads well in all five is rare. Most names that fail are eliminated here before they reach expensive legal clearance.
03
The cross-language check
Look up the name in Spanish, Mandarin (phonetic approximation), Japanese (romanization), and German. Check for unintended meanings, negative connotations, or existing brand conflicts in your top three markets. The Chevy Nova in Spanish-speaking markets (Nova = "doesn't go") is the classic cautionary example. Even B2B software companies reach international customers earlier than expected.
04
The trademark knock-out search
Run a free search on the USPTO TESS database before investing further in any candidate. Search the exact name and phonetic equivalents in your primary classes (International Class 42 for software; Class 35 for business services). A name with an existing registration in your class is a material legal risk regardless of how strong it is on every other dimension. This search takes five minutes and eliminates roughly 30% of strong candidates.
05
The phoneme fit test
Ask whether the sounds in the name create the right impression before anyone reads the description. High-energy companies (Zoom, Stripe) use plosive consonants and short vowels. Precision tools (Linear, Figma) use fricatives and minimal vowel profiles. Warm or accessible brands (Hims, Ro) use nasals, liquids, and open vowels. A name whose sounds contradict the brand impression creates a friction that compounds over time -- customers can feel the mismatch even when they cannot articulate it.
The practical process from blank page to shortlist
1
Answer the three positioning questions
What impression does the name need to create? What context will it live in? What does it not need to do?
2
Run three generation approaches in parallel
Pick any three from the six above and generate 20-30 candidates from each. Aim for 75-100 raw candidates before filtering.
3
First filter: eliminate on structure
Remove altered spellings, compound words that are too descriptive, and names that follow the dominant convention in your category. You should cut roughly 60% at this stage.
4
Second filter: run the five tests on what remains
Spoken clarity, context performance, cross-language, trademark knock-out, phoneme fit. Run them in that order. Stop testing a name the moment it fails.
5
Shortlist to 3-5 finalists
The names that clear all five tests are your shortlist. If you have fewer than three, go back to generation -- the pool was too small or the approach too narrow.
6
Full trademark clearance on finalists only
A formal clearance opinion from a trademark attorney costs $800-$2,000 per name. Do this only after the five filters have been applied -- not before. Most founders get this order backwards and spend $5,000 clearing three names before any of them have passed the spoken clarity test.
How many candidates do you actually need?
The answer surprises most founders: a lot more than you think. The typical founder generates 8-12 names and treats that as a complete process. The typical naming agency generates 200-2,000 candidates internally and presents 8-15 to the client. The difference is not effort for its own sake -- it is that the distribution of name quality is highly right-skewed. Most of the good names are not obvious. They are discovered by exhausting the obvious space first and then going further.
A manual generation process that produces 75-100 candidates is workable. Below 50, the shortlist is almost certainly over-anchored on the names that came to mind first, which are usually the most expected names in the category. The best names require getting past the obvious.
"You cannot evaluate your way to a great name. You have to generate your way there first."
What to do when you are stuck
The most common stuck state in company naming is having a 6-10 name list that no one on the team can agree on, combined with the feeling that every obvious name is taken. This is a generation problem, not a selection problem. The list is too short, not too hard to choose from.
The reset that works: temporarily remove all constraints. Ignore domain availability. Ignore trademark. Ignore whether anyone likes the names. Set a 20-minute timer and generate 50 names using only one approach (category pivot or morpheme construction work best here). Do not evaluate any of them until the timer is done. The quality of the second 25 names is almost always higher than the first 25, because the obvious space has been exhausted.
After that session, apply the five tests. The right name usually appears in the second 25.
Check your name candidates against 14 dimensions
The free phoneme analyzer scores any name across Energy, Authority, Warmth, Precision, Innovation, and 9 more dimensions. See where your shortlisted names land before you commit to one.
Analyze any name free →
Flash proposal $499 · Studio $4,999 · 30-day money-back guarantee