[BLOG] Most AI brand names fail one boring test

Problem

Founders spend three days generating names with AI. The list looks great. Then the .com check happens, and every good name is gone. Back to the prompt. Another list. Same result. Round three. Round four. Meanwhile the product ships without a name, or ships with a name nobody can spell.

The trap: AI generates from meaning, but availability lives in a registry that AI can't see. So the loop keeps producing names that sound perfect and are already owned. The founder falls in love with a name, spends an hour designing a mock logo, then checks the domain and starts over. That's not a naming problem. That's a filter problem.

What we tested

We ran a naming sprint for our own rebrand - solo, no agency. The setup was normal for round one: prompt an AI, get a list of candidates, feel clever. What we did differently was wire a live check into the loop before the emotional attachment kicked in.

The check itself isn't fancy. RDAP is the successor protocol to WHOIS - a JSON endpoint every registry runs. One curl call per domain tells you if it's registered right now. We scripted it so every candidate name got a green or red flag before it made it onto the shortlist. Nobody looked at a name until the flag was green.

Four rounds, three naming strategies:

• Round 1 - descriptive names. Words that literally describe what the product does. "Content", "Factory", "Fleet", "Operator" - combined and stacked. The prompt was "generate 40 SaaS names for a solo-operator content platform."
• Round 2 - easy-word combos. Two common English words smashed together. Think "Notion" + "Slack" energy. The prompt asked for simple, phonetic, five to eight letters.
• Round 3 - metaphor names. Names that don't describe the product at all. They describe how the buyer sees themselves. Ship metaphors, tool metaphors, weather metaphors. The prompt asked for names a specific person would want on a t-shirt.
• Round 4 - audience mirror. We stopped generating from scratch and started listening. Read three months of forum posts and podcast transcripts from people in the target audience. Pulled the exact words they used to describe their own work. Fed those back to the AI as raw material, not as a brief.

Every domain got checked against RDAP the moment it appeared. No candidate reached the shortlist without a green flag. If AI hallucinated a name that turned out to be taken, we saw it inside two seconds instead of two days.

The result

187 domains checked across the four rounds. 30 came back available. That's 16 percent. The strategies didn't perform equally.

The descriptive round was a graveyard. Every good literal name for a content platform, a solo-ops tool, or a marketing engine has been sitting in someone's portfolio since 2013. The 6.5% hit rate came from typos and hyphenations we wouldn't ship anyway.

Metaphor names cleared four times that rate. Nobody else was reaching for the same well because there's no obvious well to reach for. A name about being alone at sea doesn't collide with 400 competitors describing "automation" and "workflow" the same way.

Audience mirror was the surprise. It didn't just have the best hit rate - it produced the name we picked. When we listened to how the target buyer described their day, one phrase kept surfacing: they called themselves "a one-man fleet". Not lonely, not overwhelmed - a fleet. Multiple ships, one captain, still moving.

That was the whole naming exercise. Two hours, zero dollars. Not because we're fast, but because 84% of the work - killing names that were never available - happened before the emotional part started.

The takeaway

Four lessons came out of this, and three of them we didn't expect.

AI names from meaning. Availability lives in the registry. These are two different problems and the industry treats them as one. Every naming tool we tried generates candidates and stops. Availability is a separate tab, a manual check, a place where hope goes to die. Wire the check into the loop and the loop gets honest. Two lines of curl against RDAP is enough. You don't need a service, a subscription, or a domain broker.

Metaphor beats description because everyone describes the same product the same way. If your product helps founders ship faster, you and 200 competitors will both reach for "Ship", "Launch", "Fast", "Solo". Those .coms went in 2016. A metaphor doesn't collide because there's no shared frame to collide with. The upside: metaphor names also survive product pivots, because they describe the buyer, not the feature. When you add a second product, the name still fits.

Listen before you generate. The strongest names don't come from an AI prompt. They come from the audience describing themselves in their own words, then getting upgraded slightly. "One-man fleet" was in the forum posts before it was on our shortlist. We didn't invent it. We noticed it. AI was the tool that helped us test the variations - but the raw material came from listening.

Spelling matters more than cleverness. We killed a name we loved because one of the two words was routinely mistyped. If your audience can't dictate your domain to a friend at a bar, the name doesn't work. This is a boring lesson and it eliminated our second favorite candidate. The name that survived is spellable on the first try, even for someone who's had two drinks.

The thing that's hard to copy: the RDAP check is trivial once you know it exists, but nobody teaches it because naming agencies charge for the illusion that names come from creative genius. They come from filtering. The filter is the work.

Try this

Three things, in order, that a solo founder can do this week:

1. Add a live domain check to your naming loop. RDAP is a public JSON API. curl https://rdap.org/domain/yourname.com returns 200 if taken, 404 if available. Loop it over your candidate list before you look at the list. Names that come back red never enter your brain. Names that come back green get evaluated on merit, not on "well, we could add a hyphen or a .io".

2. Run three separate rounds, not one big one. One round for descriptive, one for word-combo, one for metaphor. Keep the results separate and count the hit rates. You'll see the pattern within an hour: descriptive is a wall, metaphor is a door. Prove it to yourself instead of taking our word. The numbers will tell you which round to spend the second hour on.

3. Read your audience for a day before you prompt. Pick one forum, one subreddit, or one podcast where your buyer talks. Read three months of it. Copy every phrase they use to describe their own work. Paste those into your naming prompt as source material. The name you pick will probably be a phrase you didn't write - you noticed it, filtered it, and put a domain around it.

The full sprint took two hours and cost nothing. The first hour was scripting the RDAP loop. The second hour was three rounds of generate-and-filter. If you cut out the availability check, the same sprint takes three days and produces a name you can't buy.

Every founder we know who hired a naming agency paid between $3,000 and $40,000. Every one of them came back with a name they still had to check for availability. The registry doesn't care who generated the name. It cares who registered the domain first.

That's the boring test most AI names fail, and it's the only test that decides whether a name becomes a business.