Try this: think of a five-letter English-pronounceable word that isn't already a brand name. Now check the .com. Almost certainly it's taken. It has been taken since roughly 2007. There is no five-letter pronounceable .com available for under $5,000 anywhere on Earth.
This isn't a coincidence or a vibe. It's a real saturation phenomenon with an interesting math behind it, and the same dynamic shapes how to pick a name today.
The math
English speakers find five-letter words easy to remember and easy to say if they follow certain phonotactic patterns: the rules your tongue already expects. The dominant pattern for catchy brand names is consonant-vowel-consonant-vowel-consonant, sometimes called CVCVC. Think Cisco, Intel, Aetna, Kraft, Sears, Adobe, Yahoo, Lyft (CVCC variant). They roll off the tongue, sit in working memory, and stick after one hearing.
There are 21 consonants and 5 vowels (Y set aside for simplicity). A naïve count of CVCVC permutations is 21 × 5 × 21 × 5 × 21 ≈ 231,525. Most of those don't sound like English. ZIBOK, FUYAP, WIZAT, JEPUB are all valid CVCVC patterns and all phonotactically plausible to pronounce, but none feel like real words. After filtering for English phonotactic plausibility, semantic recognisability, and the brand-friendliness bar (which excludes anything with X, Q, Z in awkward positions, plus anything that already sounds like an existing word), what's left is closer to 7,000–15,000 usable five-letter brand candidates. Psycholinguists produce tighter numbers using phonotactic-probability models. Vitevitch & Luce's calculator is the most widely cited; the order of magnitude lines up with the napkin math here.
Verisign's quarterly Domain Name Industry Brief puts .com registrations at around 160 million. Of the ~7k pronounceable five-letter candidates, essentially all of them are claimed. The aftermarket reflects this: short, clean, pronounceable .com domains sell on Sedo and GoDaddy auctions for $2,000 to $80,000 regularly. The ones that are also brandable English words clear six figures.
How startups have solved it
Six escape routes the last decade of startups has shown work.
1. Drop letters
Lyft (4), Twtr (the trademark, pre-X rename), Tumblr (without the e), Flickr, Scribd, Grindr. The misspelled-vowel-drop peaked from 2008 to 2014, then went out of fashion. Don't do this unless the audience will remember the missing letter.
2. Made-up words
Plaid, Vimeo, Vibez, Klarna. Coined from scratch, designed to be unique. Pros: trademark-clean, domain-available, no semantic baggage. Cons: nobody knows what you do; the first three years go to explaining the name.
3. Compounds
Notion (6), Linear (6), Cohere (6), Anthropic (9), Replicate (9), Substack (9). Slightly longer than 5 letters but the space of pronounceable 6-to-9 letter words is roughly 50x larger and far less saturated. The dominant pattern in 2024-2026.
4. Foreign-language borrows
Anthropic (Greek), Cohere (Latin), Krea (Latin-flavored), Vercel (Norse-ish coinage). A word from Greek, Latin, or a less commercially-saturated language buys trademark clearance and domain availability, and often signals seriousness or warmth depending on the source language.
5. Skip pronounceable, commit to initialisms
Works for B2B and dev tools where the audience is comfortable reading symbols. AWS, IBM, SAP, HPE. Doesn't work for consumer brands: hard to sell HPE shampoo. Most modern initialisms started life as full names and got shortened (International Business Machines became IBM). Coining a brand-new initialism for an unknown company is a hard ask of new users.
6. Different TLD
Loom.com is taken (and worth millions). Loom.dev wasn't, when they launched. Same for many AI startups now on .ai. The TLD signal differs across each option (covered in what each TLD says about your brand). For dev-flavored or AI-adjacent products, .dev, .io, and .ai give 5-letter availability that .com can't.
Practical implication
If you're picking a name right now, save the heartbreak: search the 5-letter space briefly to confirm it's saturated (it is), then aim for 6-9 letters, ideally a compound or a foreign-language root. Run candidates through dibbed.dev early. Domain status is the bottleneck even at 6 letters, less so at 8.
The 5-letter problem is solvable, but only by going around it. Don't fight the math.
Ready to check a name? Run it through dibbed.dev →