Brandable or keyword? The old naming fight assumed people would see your domain. In the age of AI search, that assumption is breaking.
Someone asks a chatbot for the best project-management tool for a ten-person agency. Three names come back. The person picks one, opens a tab, and types it in. At no point did anyone choose a domain. No ranking page was scanned, no URL was admired, and no address bar did any persuading.
For twenty years, the domain-name business has run on a single argument: should you buy the name that describes what you sell, or the name that sounds like a brand? Exact-match keyword domains versus brandables. It is a fight that has sustained brokers, naming consultants, forum threads and a great many overpriced .coms.
AI search doesn’t settle that argument so much as quietly change the question.
In the 2000s, a name like cheapflights-style keyword strings could rank on the strength of the words alone. Search engines treated the domain as a relevance signal, and an entire cottage industry grew around hoarding generic phrases.
Then Google acted. In September 2012 it announced an update aimed at low-quality sites that ranked mainly because their domain matched a search query. The keyword shortcut lost much of its force, and the conventional wisdom flipped: pick something distinctive, something people can say out loud and remember. Brandable won.
That verdict held for a decade. It rested on one idea, which is that the domain is a signal competing in a list of ranked results. Chatbots don’t present a list.
An AI answer synthesizes. It names some entities, sometimes attaches a few sources, and moves on. That creates two separate things a domain might influence: whether you get mentioned by name in the answer, and whether your site gets cited as a source. The data suggests these are surprisingly different events.
Semrush and Growth Memo’s Kevin Indig analyzed nearly 4,000 domain appearances across 115 prompts, 14 countries and four AI engines. About 62 percent were what they called “ghost citations”: the page was linked as a source, but the brand never appeared in the answer text. Only about 13 percent were both cited and named. Another quarter were named with no link at all.
A separate analysis of ChatGPT data across 1,094 U.S. categories, as reported by Search Engine Journal, found a slightly negative correlation between citation counts and brand mentions, and the most-cited domain was also the most-mentioned brand in only about a fifth of categories.
Read that carefully. Being the source and being the name are different games, and your domain plays them differently.
Here is the contrarian argument, and it is worth stating fairly.
When a chatbot does show sources, it shows them as small labels. A reader scanning a list of citations is making a snap judgment, and a descriptive domain does some of that work for free. “Source: bestmattressguide.com” announces relevance. “Source: zorvo.io” announces nothing.
There is a second argument for newcomers. A model has to categorize an unknown company before it can recommend it, and a name that literally describes the category might shorten that step. If you are a new entrant with no press coverage, a descriptive name could be a head start.
First, the domain’s string doesn’t seem to drive citation. Otterly.AI’s May 2026 analysis of more than a million cited URLs found that .com, .org, .io and .ai domains all averaged the same number of citations per URL, about 1.7. The extension was not a selection signal. (The analysis was reported secondhand via a domain registrar’s explainer, so it is worth verifying against Otterly’s original.)
Second, the AI-flavored name premium looks smaller than the hype suggests. Dynadot’s Q2 2026 aftermarket report, drawing on more than 54,000 completed sales, found names containing AI keywords averaged roughly 1.03 times the market price, with a median that matched the market exactly. Headlines about .ai names and the $70 million sale of AI.com describe a few extraordinary names, not a broad premium for sounding like the future.
Third, and most important, a lot of the time the domain isn’t shown at all, or isn’t what the user acts on. People ask for a name, then type or tap that name. The thing they carry from the chatbot to the browser is a brand, not a URL.
The standard defense of the brandable is that it is memorable and ownable. That is true but not the interesting part.
The interesting part is disambiguation. Language models learn entities from text: which names co-occur with which categories, how often, and in what context. A distinctive coined name is a clean unit to learn. A phrase like “best project tools” is just words, shared with every listicle and every competitor, and it is easy for a model to blur into the general noise of the category.
That is a mechanism argument, not a measured result, and it should be held lightly. No one has published a controlled test showing that coined names get mentioned more often than descriptive ones. But it fits how these systems are described to work, and it fits the voice-and-chat reality that people tell an assistant a brand and expect it to resolve.
The catch is that brandables have their own fragility. Unusual spellings that look clever on a pitch deck are miserable when spoken aloud to an assistant and then typed back in. The new ideal is a name that is easy to say, easy to spell, and unlike anything else in its category.
The most consistent finding across the vendor studies has little to do with domain strings at all.
Victorious’s Q1 2026 analysis of 177 brands across eight AI platforms found that domain authority correlated essentially not at all with AI visibility, and that about 90 percent of the brands had no AI mentions whatsoever. Authority was necessary, they argued, but nowhere near sufficient.
AirOps and Foundation Marketing studied 5.1 million AI responses and found only about 10 percent of cited links came from a brand’s own domain. For unbranded, intent-driven queries, the figure fell to roughly 2 percent. In other words, when someone is still deciding, the model mostly cites other people’s pages about you.
Ahrefs found the same pattern on the branded side. In its panel of 100 major brands, nearly two-thirds of exact-name searches triggered a Google AI Overview by the end of September 2026, and the outside source cited most often was Wikipedia, appearing in 41 of 57 overviews with source lists, followed by YouTube and LinkedIn.
The pattern is hard to miss. The model’s picture of your company is assembled from reviews, encyclopedias, forums and press, not from the string you registered. Your domain is how people reach you once they already know your name. It is not how the machine decides to say it.
There is one reassuring note for the unknown. The ChatGPT category analysis estimated that roughly 89 percent of AI search demand sits in categories with no clear brand owner. The field is open, which means the way to win it is to be talked about, not to be named cleverly.
The honest answer is that the keyword-versus-brandable debate has been demoted. It still matters at the margins, and it is worth getting right, but it is no longer the lever people thought it was.
A few practical takeaways follow from the evidence, with the caveat that most of it is vendor research from the last year and should be read as directional.
Pick for the mouth, not the eye. Say the name aloud, then spell it from memory. If a friend can’t type it after hearing it once, an assistant relaying it to a user will fail the same way.
Be distinctive within your category. The disambiguation argument is the strongest one for brandables. Avoid names that read as generic phrases, and avoid names that collide with an existing, better-known entity.
If you go descriptive, make it earn it. A keyword name can work as a trust cue and a categorization shortcut, but it will not get you cited or mentioned by itself, and it makes it harder to be a distinct entity.
Don’t overpay for an AI-themed name. The measured aftermarket premium is slim, and the extension appears not to matter for citations.
Spend the real effort elsewhere. Reviews, third-party coverage, a Wikipedia-grade presence, community mentions. That is what the studies keep pointing back to.
Current system time:2026-11-12 21:30:17(UTC+8) Privacy PolicyRegistrants' Benefits And Responsibilities SpecificationsRegistrant Educational Information
Copyright© 2026 GNAME.COM. All rights reserved. Non-Public Registrant Data (NPRD) Disclosure