Ask ChatGPT, Claude, Gemini, or DeepSeek for an available .com and you will get a confident list. Those models are not connected to the domain registry. They predict a name that sounds available. We measured how often that guess is wrong.
This page is the AI Domain Hallucination Index: 2,449 suggestions from 10 models, run on 2026-09-01, each checked live via RDAP against the authoritative .com registry. “Available” means the registry returned no active registration at check time. Verify a name the AI just gave you, or generate live-available .coms with the same RDAP check. The AI domain name generator and how to check if a domain is taken use that check too.
The leaderboard
Each model was asked for hundreds of supposedly-available .com domains. The percentage is how many of those picks were free to register. Overall, 30.5% of the 2,449 suggestions were actually available. The best model in this run, anthropic/claude-sonnet-5, still only managed 50.5% — a coin flip. The weakest, anthropic/claude-haiku-4.5, was 8.2% (about 92% already taken). n is how many verified suggestions each model contributed; google/gemini-2.5-pro is shown for completeness but was rate-limited to a sample too small to rank.
| Model | Actually available | n | |
|---|---|---|---|
| Domain Search King our tool | 98.7% | ||
| anthropic/claude-sonnet-5 | 50.5% | 287 | |
| anthropic/claude-opus-4.8 | 46% | 287 | |
| openai/gpt-5.6-luna | 38.2% | 288 | |
| deepseek/deepseek-v4-pro-0813 | 32.6% | 291 | |
| google/gemini-2.5-pro low sample | 31.7% | 41 | |
| mistralai/mistral-nemo | 31% | 168 | |
| google/gemini-3-flash-preview | 22.9% | 288 | |
| nousresearch/hermes-3-llama-3.1-70b | 21.9% | 247 | |
| meta-llama/llama-3.1-8b-instruct | 21.1% | 270 | |
| anthropic/claude-haiku-4.5 | 8.2% | 282 |
Raw leaderboard CSV: ai-hallucination-index.csv (CC BY 4.0). DeepSeek’s row is 32.6% available on the same method. Full write-up.
Paste the name the model gave you
Check it live, or generate names that pass the registry before they reach the screen.
Verify a domain → Generate available names →Where AI wastes your time most
The miss rate is not even. Across all 10 models, retail and professional-services names are the dead ends — online store, SaaS and consulting sit at the bottom, and they are the searches founders actually type. Trades and lifestyle niches do far better. Domain Search King is 98.7% across every niche in this test, because taken names never make the list.
| Business type | AI picks free to register |
|---|---|
| Online store | 10.6% |
| SaaS startup | 17.4% |
| Consulting firm | 19.0% |
| Restaurant | 21.8% |
| Law firm | 26.4% |
| Coffee shop | 39.0% |
| Yoga studio | 49.4% |
| Domain Search King | 98.7% |
We asked for 96 names. All 96 were already taken.
In July 2026 we ran the same founder prompt again on a current frontier model (Claude Opus 4.8): “give me available .com domain names for a <business>.” 96 brandable names across 12 industries, every one checked live. Zero were available.
The model offered names registered before it existed, some from the early web:
| Name the AI called “available” | Registered since |
|---|---|
| dailygrind.com | 1995 |
| leakbusters.com | 1996 |
| flowmasters.com | 1997 |
| pipeguard.com | 1998 |
| draindoctors.com | 2000 |
| brewhaven.com | 2002 |
| morningritual.com | 2004 |
Median registration year of that sample: ~2004. Download: ai-hallucination-index-registration-dates.csv (CC BY 4.0).
Why this happens
A language model predicts the next plausible token. A short, brandable .com sounds like a free name because those strings are common in training data: blog posts, old directories, “great name” lists. People have been using those names since the 1990s. The better a name sounds, the more certain it is already owned.
The model was never connected to the registry. Instant Domain Search, a rival naming tool, said the same thing in 2024: “one limitation when using ChatGPT as an AI name generator is that it does not assess domains for availability.” Ahrefs found a cousin of the same bug on the open web: AI assistants send people to dead 404 pages 2.87× more often than Google, because they assert that a page exists without checking.
Availability and trademark clearance are two different searches. An AI-suggested name can also collide with a live mark the model never checked. Confirm both before you commit. We do not know of another registrar or namer (GoDaddy, Namecheap, Namelix, Looka) that publishes this hit rate. As far as we can find, this is the only live benchmark of it.
Domain Search King checks every name against the .com registry in real time before it appears. Taken names never reach the screen. That is the 98.7% row above. The same check is available as a remote MCP for Claude, Cursor, and any agent that should stop guessing.
How we measured it
Each model received the same prompt: available .com names across dozens of business niches and naming styles (freeform, one-word, prefix/suffix). Every suggestion was checked live against the .com registry. Available means no active registration at check time.
2,449 suggestions sit on the leaderboard, from a single run stamped 2026-09-01; every row is in the public dataset, so the per-model rates can be recomputed from source. The July 2026 spot-check (96 names, 0 available) is a separate frontier-model re-run, not mixed into the per-model rates. Reproduce it: ask any model for “available” .coms and check each at a registrar.
Raw data is free to reuse under CC BY 4.0 with a link back: leaderboard CSV · registration-year sample · per-industry CSV · per-TLD CSV.
Get names that pass the registry
Live-checked .coms for founders. MCP for agents. Already holding a taken name? Start with the 7 moves that fix it.
Generate available names → Use the MCPFAQ
Why do AI models suggest taken domains?
A language model predicts plausible names. It was never connected to the domain registry, so most names it calls “available” are already registered.
How was availability measured?
Every suggested .com was checked live against the authoritative .com registry. “Available” means the registry returned no registration at check time. 2,449 suggestions were verified in the 2026-09-01 run.
How long have the taken domains been registered?
Often for years or decades. In the 2026-09-01 run the median registration year of AI-suggested “available” .coms was 2018, and some had been registered since 1993 — long before the model that suggested them existed.
What should I do with a name an AI just gave me?
Verify it live before you get attached to it. If it is taken, generate alternatives that have already passed RDAP, or follow the AI-taken rescue flow. Attackers also register the fake domains models invent for real brands. That is phantom squatting.
Are the register links affiliate links?
Yes. Register links may earn Domain Search King a commission at no extra cost to you. Availability can change; register the same day. Trademark search is a separate check.