Why Startups Are Ditching Keyword Domains for Invented Names

The Pattern in 2025-2026's Best-Funded Startups

Look at where venture capital actually went in the last 18 months, and a naming pattern jumps out: the biggest rounds are going to companies with names that say nothing about what they do.

Y Combinator's Fall 2025 batch shows the same pattern at the earliest stage — short, abstract company names like Rivet, Item, Bear, Moss, Iris, and Dome, rather than descriptive ones (Smart Branding). This isn't confined to AI, either — Notion, Figma, Loom, Linear, Vercel, and Supabase set the precedent a few years earlier.

Get the History Right: What Google's 2012 Update Actually Did

The usual shorthand — "Google killed keyword domains in 2012" — overstates it. On September 28, 2012, Matt Cutts announced Google's Exact Match Domain (EMD) update, after warning webmasters about it in a March 2011 video (Search Engine Journal). Google's own figures put the impact at roughly 0.6% of English-language queries — and the target was specifically low-quality EMDs paired with thin or spammy content, not exact-match domains as a category. Moz found about 41 EMDs fell out of the top 10 results afterward, with non-.com EMDs hit harder than .com ones.

The precise, current-day (2026) status: Google treats keyword-in-domain as a minor contextual signal, not a ranking boost (SEO Melbourne, DomCop). EMDs still show up disproportionately at the top of some search results — but that's mostly because owning the exact phrase correlates with genuine topical focus, not because the domain string itself gets a boost. What changed in 2012 wasn't "EMDs stopped working." What changed is that a keyword domain stopped being a shortcut that could substitute for actual content quality and backlinks.

The Real Driver: Trademark Math, Not Just Taste

The strongest, best-evidenced reason behind the invented-name shift is legal, not aesthetic. David Placek of Lexicon Branding — the naming firm behind BlackBerry, Swiffer, Azure, Vercel, Windsurf, and Sonos — puts a number on how much the landscape has changed: US Class 9 (technology) trademark registrations have gone from roughly 23,000 forty years ago to well over 2 million today (Focus Lab / Lexicon, Jan 2026). Clear a name across roughly 35 global markets, and the collision odds compound fast.

This isn't just a crowding problem — it's a legal-strength problem. Trademark law treats descriptive names ("Denver Coffee Roasters," "Fresh Juice Co.") as inherently weak marks: harder to register, easier to challenge, and offering thin legal protection even when registered (Stemer Law, L4SB). Arbitrary and fanciful names — invented words, or real words used with no descriptive connection to the product — sit at the strong end of that spectrum and are far more defensible. Placek's advice for founders: start naming and trademark clearance 2-3 years before launch if you're building for venture scale and global markets.

The Twist: "Invented" Isn't an Automatic Trademark Shield

Short and abstract doesn't mean safe. In December 2025, the Ninth Circuit affirmed a trademark injunction against OpenAI's "io" product line in a dispute with IYO Inc. over a confusingly similar mark (IPWatchdog) — a company with essentially unlimited legal resources still collided.

The risk is worse for founders relying on AI name generators without following up with real clearance. Documented 2025-2026 cases include a startup that received a cease-and-desist roughly three months after launching a name containing "Lumina" (about $25K in rebrand costs), another that absorbed roughly $350K in total rebrand costs after a trademark notice, and a Series A that was delayed five months (~$185K in costs) over a phonetically similar mark (IPRightsHub, 2026). There's also an irony worth flagging: AI-adjacent invented-sounding words — Nova, Synth, Flux, Nexus — have been used so often in tech naming that Focus Lab now says they "feel generic and fail the distinctiveness test completely." Invented is not automatically distinctive; it still has to actually be distinctive.

When a Descriptive Domain Still Makes Sense

The invented-name trend is a venture-scale, global-ambition phenomenon — it's not a universal rule. For local and hyper-niche service businesses, a descriptive domain can still do real work: it pre-qualifies clicks by telling a visitor exactly what the business does before they land, functioning like a 24/7 self-explaining ad in a way an invented name can't without separate brand-building spend. Some sources cite `japan.travel` as an example of an exact-match domain holding a strong ranking for "japan travel" queries (Baslon Digital), and local SEO practitioners report a perceived (if not Google-confirmed) edge for geographic or service-keyword domains in local search specifically (local SEO commentary).

Worth flagging plainly: most of this counter-evidence comes from domain-registrar and SEO-vendor blogs with a direct commercial interest in "domains still matter" being true. Treat it as plausible, not settled — the same scrutiny applied to the naming-agency claims above.

A Practical Decision Framework

Venture-scale, global ambition, planning to raise real capital: lean invented or arbitrary. The trademark-clearance math favors it, and Google hasn't rewarded keyword domains in over a decade regardless.

Local business or hyper-niche service, long-tail search intent: a descriptive or hybrid name is still defensible — the self-qualifying-click benefit is real even if the SEO boost is mostly folklore.

Either path: a domain-availability check is not a trademark check. An AI generator (including this one) can tell you a name is registerable as a domain — it can't tell you whether someone else already owns the trademark. Run an actual USPTO search before you get attached to any name, invented or descriptive.

Frequently Asked Questions