Does your domain name matter for AI search?

Honestly: nobody outside the answer engines knows — and right now, nobody outside them can know. Google has described the machinery — retrieval-augmented generation and query fan-out, riding its ordinary Search ranking — but not the weighting: which of the retrieved pages gets cited, and why. OpenAI, Anthropic, and Perplexity haven’t published even that much. And no study has isolated the domain name as a variable. That makes this question different from its traditional-SEO sibling, where keyword domains are well and truly debunked: there, Google has answered on the record; here, the record is mostly silence. So instead of manufactured certainty, here’s what’s officially known, what the best evidence shows (and who sells it), what cannot be known yet, and what to do anyway.

Does ChatGPT or Google care about your domain name?

Google has come closest to an actual statement, and it says nothing about domain names. Its Search Central documentation is explicit: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary. To be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet.” Its July 2026 optimization guide goes further on mechanism — retrieval-augmented generation, plus a query fan-out that issues related searches behind a single question — while still describing the whole thing as “rooted in our core Search ranking and quality systems.” Domain names appear nowhere in any of it — not as a factor, not as a non-factor.

There is one named selection lever, and it belongs to the reader, not to you: Google’s Preferred Sources, announced in May 2026, lets users nominate sites they want to see more of, and Google has said it is working toward using that as a signal. The way to be preferred is for someone to type your name on purpose — which is a branding problem, not a markup one.

The others say even less. OpenAI confirms ChatGPT search can include inline citations but publishes no weighting formula for why one source wins over another. Anthropic’s documentation covers how Claude’s citations are formatted, not how sources are chosen. Perplexity has published nothing on ranking at all; outside researchers describe a pipeline favoring freshness, topical relevance, and clean quotable pages — inferred behavior, not documentation.

Hasn’t Google already said domain names don’t matter?

For classic Search, yes — repeatedly and bluntly. John Mueller, Google’s Search Advocate: “A keyword domain name is not going to give you any recognizable SEO advantage on Google.” On extensions: “The TLD is not something we take into account” — the question we unpack in whether exotic TLDs hurt SEO. On age: “No, domain age helps nothing.”

Every one of those is about classic Search, made before or outside the AI-answer context. The tempting inference — AI features run on core Search systems, so the same non-factor status carries over — is reasonable, and we’d bet on it, but it’s an inference, not a quote. The reverse claim circulating in GEO content, that keyword domains boost AI citations, has no source at all: the old exact-match-domain myth transplanted with no new evidence.

What does the best available evidence actually show?

Correlations from a conflicted source, and benchmarks that never tested domain names. The most-cited numbers come from Ahrefs — which sells Brand Radar, a paid AI-visibility tracker, giving it a direct commercial interest in this topic seeming measurable and important. Two findings.

First, volatility. Ahrefs’ own analysis found the overlap between AI Overview citations and top-10 organic rankings dropped sharply between two of its own studies — though it attributes part of that shift to changes in its own measurement method rather than to Google. We won’t quote the percentages: they illustrate volatility, not stable fact.

Second, brand signals. A correlational Ahrefs study found brand-mention and brand-search signals correlated more strongly with AI Overview appearance than backlink counts did, across ~75,000 already-established brands (all with Domain Rating above 40). That’s a correlation among sites that had already cleared an authority bar, not evidence any single factor causes citation — and it isn’t a study of domain names: “Domain Rating” is an authority score, unrelated to what a domain is called. Ahrefs’ own write-up states correlation is not causation.

The cleanest sources are academic. “GEO: Generative Engine Optimization,” a Princeton study presented at KDD 2024, found that content interventions — adding citations, quotations, statistics — could raise a source’s visibility by up to ~40% in the authors’ own benchmark harness. A 2026 critical survey notes those gains are conditional on a source already being present in a fixed context, and establish “neither organic discoverability nor durable traffic effects.” Mind the term trap, too: when the paper reports that efficacy “varies across domains,” it means topic categories — medicine versus law — not domain names. A lot of GEO content quietly conflates the two.

The largest controlled study to date, presented at SIGIR ‘26, ran roughly 252,000 trials across six models and eighteen factors, and found topical relevance and position in the retrieved list dominate, while formatting-only edits barely register. Its design is the most telling detail for our question: it anonymizes brands to strip source identity out as a nuisance variable. The best experiment anyone has run deliberately erased the thing you’re asking about.

Should you add an llms.txt file?

It’s not a naming question, but it’s the one everybody asks next — and the answer is not because anyone has shown it does anything. Google’s documentation now says so directly: you don’t need “new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn’t use them.” Gary Illyes has confirmed Google has no plans to support it, and John Mueller was blunter: “AFAIK none of the AI services have said they’re using LLMs.TXT (and you can tell when you look at your server logs that they don’t even check for it).” Asked whether the file does anything at all, he answered: “I don’t think anyone knows — it’s purely speculative for now.” Neither OpenAI nor Anthropic has confirmed consuming it. (RSL, the licensing standard that has overtaken llms.txt in adoption, answers a different question entirely — payment, not citation.)

Full disclosure: domain.yoga publishes an llms.txt file itself. It takes minutes to maintain and, on Google’s own account, neither helps nor harms — but we’re not going to pretend it’s proven to do anything. Widespread adoption is not evidence of efficacy, and that applies to our adoption too.

What can nobody actually know yet?

Four things — stated plainly, because the gaps are the story.

• No engine has published its citation weighting. Google describes the machinery but not which retrieved page gets picked; everything past “be indexed and crawlable” is outside inference.
• The best experiment deliberately removed the variable. The SIGIR study anonymized brands precisely so source identity wouldn’t contaminate its results — sound method, and a reminder that nobody has yet run the inverse study that would answer this question.
• The systems change faster than the measurements. Even Ahrefs, with every commercial reason to pin this down, attributes swings in its own numbers partly to its own methodology changes. When the measurers can’t cleanly say what moved, any snapshot has a shelf life of months.
• Answers are personalized and non-deterministic. None of the primary sources reviewed for this piece describe how login state, location, conversation history, or simply re-asking changes which sources get cited — a confound under every outside study.

So what should you do about your name?

Pick a name whose value doesn’t depend on the answer — then do the three things that are actually documented.

1. Be indexed and snippet-eligible. This is the only requirement Google states, so treat it as the whole technical checklist: if a page can’t appear in Search with a snippet, it can’t be cited. Nothing exotic required.
2. Let the crawlers in. OpenAI documents OAI-SearchBot as the agent behind ChatGPT search results; blocking it in robots.txt is the one lever known to remove you. Check you haven’t done it by accident.
3. Measure yourself instead of buying someone’s tracker. Search Console’s Generative AI performance report shows your own AI-surface traffic, free, from the primary source. You can’t know the algorithm — but you can know your own numbers, which is more than any vendor study gives you.

Beyond that, the case for a distinctive, brandable name over a descriptive one never rested on ranking signals — it rests on memorability, ownability, and word-of-mouth. To the extent official material gestures anywhere, Google’s quality guidelines emphasize experience, expertise, authoritativeness, and trust (E-E-A-T) — concepts tied to who is behind the content, not what its URL is. But Google has not stated that brand recognition substitutes for or outweighs a domain name in any AI ranking system. What’s in your control is the who: a name people remember and search for on purpose (what makes a name brandable covers the ingredients), distinct enough to be recognized even when the domain doesn’t match it character-for-character, via the process in how to name a startup.

We won’t tell you Domain Yoga will get you cited by AI engines — after everything above, distrust anyone who claims theirs does. What it actually does: generates around 250 availability-checked name ideas per search, ranked for brandability with a scoring method we publish, at $2–$5 per search — no subscription, and credits never expire. A distinctive, brandable name earns its keep on human grounds alone. That’s the whole case, and it’s fortunate, because it doesn’t need the AI question answered — which nobody can do yet.