One word in your list is doing all the work

Take six US states and five building trades, cross them, and check all thirty .com domains. Six come back available. That is 20%, which sounds like a pattern with a bit of room left in it.

Here is what the 20% actually was.

Every free domain was a plastering domain. utahplastering.com and nevadaplastering.com are sitting there unregistered while utahstucco.com, utahseptic.com, utahirrigation.com and utahfencing.com are all taken, along with the same four for every other state we checked. The state you pick makes no difference at all. The trade makes all of it.

It happened in all four patterns

We ran the same 6x5 grid on three more pairs from the word lists that ship with the tool. The concentration showed up every time, and it was never subtle.

Crossing six prefixes with five AI words left four survivors out of thirty, and three of them start with flash: flashcopilot.com, flashcognitive.com, flashdiffusion.com. Every combination of my, your, in and new with any of those five words is registered. All of them. That is twenty for twenty.

Nature elements crossed with motion words came out at 50% and looked healthier until you sort it by element. Granite was untouched, five for five. Reef gave up four. Delta gave up nothing.

Adjectives crossed with brandable nouns is the pattern with the most room, at 70%, and it flips which side of the pair matters: there the noun decides. bridge is gone in five of six combinations. remix and dominion are completely unregistered.

Why the grid matters, and where it fails us

Sampling as a grid rather than thirty independent draws is what made the concentration visible. It is also a real limitation, and it cuts against the headline percentages: thirty cells built from eleven words carry a lot less information than thirty random domains would. Treat the per-pattern rates in this post as rough. The per-word splits are what the data actually supports, because each of those is six clean observations of the same word.

Two other things worth knowing before you lean on this. We capped generated labels at 24 characters, which quietly drops the longest combinations and may flatter availability slightly. And this is .com only, on one day.

What to do with it

The instinct when a pattern comes back mostly registered is to change the pattern. Our numbers say to look one level down first. A pattern that reports 20% is not 20% picked over: it is 80% exhaustively farmed and 20% barely touched, and the line between the two runs along a single word in one of your two lists.

So run the pattern, then sort the results by the word rather than reading them as a list. If four of your five terms come back with nothing, that is not a failing pattern. That is four terms somebody has already worked through and a fifth nobody has, which is more useful than any overall percentage was ever going to be.