For the last twenty-some years, domain valuation has run on one basic assumption: if a name gets typed into Google a lot, or sits close to keywords people search for, it’s worth something. Exact-match .coms built entire fortunes on that logic. Buy the keyword, rank for the keyword, sell the traffic (or the domain) later.

That assumption is starting to crack, and most of the domain investing world is still pricing names like it’s 2015.

Over the last year I’ve watched a small but real shift in where visitors actually come from on some of my own sites—not huge numbers, but enough to notice. A trickle of traffic now shows up as referrals from ChatGPT, Perplexity, and lately Claude. People aren’t typing a search term into a search bar and clicking through ten blue links anymore. They’re asking an AI a question, and the AI is picking which sites to mention, summarize, or send them to.

That’s a completely different game than SEO. And it changes what “brandable” should even mean.

Back in 2023, CircleID covered the basics of AI in domain valuation—mostly registration tools and simple appraisal algorithms. What’s changed since then isn’t the tools. It’s where the traffic itself is coming from.

Old brandable vs. new brandable

Under the old model, a brandable name mostly needed to sound clean, be easy to spell, and not scare off a trademark lawyer. Whether an AI model could reliably attach that name to a specific business or concept wasn’t a factor, because AI models weren’t doing the referring.

Now it matters. If a name is too generic, or sounds like ten other startups, a language model has no strong reason to associate it with your specific brand when someone asks a related question. It’ll just talk about the category instead, or point to whichever competitor got mentioned more in its training data and web sources. A distinct, slightly unusual, but still pronounceable name gives the model something to actually latch onto and repeat back correctly.

I’ve got a handful of names in my own portfolio that sit on opposite ends of this—one is a very literal, exact-match style name tied tightly to a single niche (grief support platforms, in this case), and a couple are invented, made-up brandables with no dictionary meaning at all. Under the old valuation logic, the exact-match name should win every time—it’s got built-in search intent. But when I’ve tested how AI tools describe or reference each type, the invented names come back cleaner and more specifically attributed. The exact-match one gets swallowed into generic category language almost every time. Nobody’s citing “griefplatform” specifically—they’re just describing grief support services in general and maybe listing a few well-known players.

That’s the opposite of how these two types of names have historically been priced against each other.

Why this matters beyond just domainers

This isn’t only a domain-investing curiosity. If AI answer engines become a real referral channel—and every signal points that way—then domain valuation tools that are still built entirely on historic sales comps, keyword volume, and backlink profiles are going to be measuring the wrong thing more and more often. A domain can look “cheap” by every traditional metric and still be a poor long-term brand pick because it can’t survive being paraphrased by a model that’s summarizing a whole category instead of naming a specific business.

There’s also a policy angle worth raising, and I think this is where it gets interesting for the ICANN and registry crowd specifically. A lot of dispute policy and valuation precedent (UDRP cases, trademark disputes, even some registry pricing tiers) still leans heavily on traffic and search demand as the measure of a domain’s worth. If a meaningful chunk of future “traffic” becomes invisible—happening inside an AI chat window instead of a browser with a referrer header—some of that evidentiary logic gets shakier. How do you argue bad-faith registration based on traffic diversion when the traffic in question never touches a search results page at all?

What I’d actually tell someone buying names right now

Don’t throw out the fundamentals—length, pronounceability, and clean history still matter a lot. But I’d start weighting distinctiveness over exact-match intent more than most valuation tools currently do. A name that’s slightly harder to guess the meaning of, but easy to say and impossible to confuse with a competitor, is probably aging better in an AI-mediated world than a keyword-stuffed exact match that used to be a slam dunk five years ago.

Nobody’s built a solid appraisal tool around this yet, as far as I’ve seen. The existing AI-powered valuation tools are still mostly trained on historic sales data and SEO-era signals—which means they’re predicting the past pretty well and the next few years not so much.

I don’t think this fully replaces keyword-based valuation. But it’s the first real crack I’ve seen in a pricing model that’s held steady since the early 2000s, and I think it’s worth domain investors—and honestly ICANN policy folks too—paying closer attention to before it gets priced in everywhere at once.