How to come up with a brand name — the 7-step process.

Step 2 — Name the buyer's cohort

Who buys this? Not the persona deck. The cohort with a vocabulary. Infrastructure engineers between thirty and forty-five who read Hacker News. Marketers at Series B SaaS who live in Notion. Founders of seed-stage consumer companies who watch twenty-second TikToks. The cohort dictates the linguistic register the name lands in.

Linear was briefed against a cohort of infrastructure-leaning software engineers. The name had to feel like a Unix tool — short, hard consonants, one-syllable energy even at two syllables. Cursor was briefed against the AI-tooling cohort that already uses keyboard-first editors. Same lineage. Different cohort would not have picked either name.

Step 3 — Pick one emotion

At first encounter, what should the name make the buyer feel? Competence. Calm. Speed. Warmth. Gravity. Curiosity. Pick one. The phonemes encode the feeling — hard plosives (k, t, p) read fast; liquids and nasals (l, m, n, r) read warm; long vowels read grand. The Sapir & Köhler line of sound-symbolism research, summarized in Sound Symbolism for Brand Naming, gives you the empirical map.

Stripe is fast: hard /str/ onset, hard /p/ release. Notion is calm: nasal /n/ onset, soft sibilant close, no plosives. Cursor is precise: hard /k/, hard /r/, the /sɔːr/ that lands the eye exactly where the cursor lands on screen. The match between phoneme and feeling is not coincidental; it is the encoded craft of the naming agencies that priced these decisions for thirty years.

Step 4 — Pick a linguistic style

Real word, coined word, compound, Latinate root, mythology, deliberate misspelling, found-object: each style trades different properties. The Operative Words taxonomy gives nine clean buckets, and the choice between them is the most consequential decision before the brainstorm.

Real English word (Linear, Stripe, Notion, Cursor, Plain, Loops). Maximizes meaning. Minimizes domain availability. Maximizes §2(e)(1) descriptiveness risk if the word is too on-the-nose. Linear works because it is a property (linearity) that is metaphorically related to the product, not the product itself.

Coined word (Stripe is borderline; Mintlify, Tigris, Convex, Sardine, Anthropic, Persona). Maximizes distinctiveness. Maximizes trademark room. Minimizes first-encounter comprehension. Coined names cost more in category education and earn more in legal defensibility.

Latinate root (Linear, Vercel, Modal, Loops, Convex, Falcata). The current SaaS canon leans heavily on Latin: it sounds grown-up, clears trademark easily, and reads cleanly in English and most Romance languages. The current Y Combinator cohort has noticed.

Pick the style first. Then ask the brainstorm for candidates inside the style. A brainstorm without a style brief returns sixty candidates that span all of them, and the founder ends up comparing apples to oranges to coined Latinate apricots.

Step 5 — Write the constraints down

Before brainstorming, write the hard constraints. .com required, or aftermarket acceptable up to $X? USPTO clearance required in which Nice classes? Social handles acceptable as @brand or acceptable as @brandhq? Year-one and year-two launch markets — does the name need to survive Brazilian Portuguese, Mandarin, Hindi, Arabic?

Written constraints prevent the most expensive failure mode: discovering, in week fourteen, that the name does not pass a constraint nobody wrote down. The constraint sheet is the rebrand-prevention layer. Tape it to the wall.

Step 6 — Brainstorm against the brief

Now brainstorm. Sixty candidates is the right order of magnitude — not ten, not three hundred. Six rounds of ten. Each round opens a different drawer:

• Round 1 — Direct. Words that map directly to the positioning. These will mostly be too descriptive to clear trademark, but they map the territory.
• Round 2 — Metaphor. Properties of the product re-expressed as concrete nouns. Linear is a metaphor (linearity).
• Round 3 — Mythology. Greek, Roman, Norse, Egyptian. Apollo. Hermes. Athena. Falcata. Maps onto a cohort that reads books.
• Round 4 — Latinate. Latin and Romance roots. Vercel, Modal, Convex. The current SaaS register.
• Round 5 — Found-object. Stripe, Bolt, Loom, Plaid. A real word from an unrelated domain, deliberately repurposed.
• Round 6 — Coined. Synthesize sounds. Mintlify, Tigris, Sardine. The expensive distinctiveness move.

Read every candidate aloud. Words live in the mouth, not on the page. A word that looks good in twelve-point Inter and feels awkward on the third syllable is a word that will lose every sales call. Score each candidate against the five-word positioning, the cohort, the emotion, the style. Keep the top twelve.

Step 7 — Verify every finalist across five axes

The twelve finalists go through the five-axis verifier: trademark, domain, cultural meaning, sound symbolism, pronunciation resilience. The verdict is what decides which candidate ships. Every flag traces to a record number — the USPTO serial, the Wiktionary entry, the audio file of the misheard variant.

In our corpus, the median twelve-finalist shortlist contains two PROCEED candidates, five STRATEGIC candidates, three STRATEGIC candidates, and two ABANDON candidates. You will not know which is which until the verifier returns the verdict. The cost of running twelve candidates is fifteen seconds. The cost of skipping the step is the week-16 envelope and the $40K legal bill.