TL;DR
- Profound's analysis of ~680M AI citations (Aug 2024–Jun 2025) found roughly 80% point to .com domains, with .org a distant second.
- That gap is a corpus-composition artifact, not a ranking signal. Google and the major AI providers have no stated preference for any TLD.
- Brand mentions across the open web appear to move AI citation odds far more than domain extension — one Ahrefs analysis put mention influence at ~3x that of backlinks.
- Newer TLDs (.ai, .io, .co) are treated as generic gTLDs by Google. The disadvantage they face is age and mention volume, not the string after the dot.
- AI citation patterns are volatile per prompt and per model. Any TLD-level rule of thumb should be held loosely.
The question comes up in almost every GEO audit: "We're on a .io — are we structurally disadvantaged in ChatGPT?" It's a fair worry, because when you actually count citations across LLM answers, the .com skew is enormous. But the causal story is more boring than it looks, and the practical implication for a .ai or .co brand is not "rebrand" — it's "earn more mentions."
What the citation data actually shows
The most-cited public dataset here comes from Profound, which analyzed roughly 680 million AI citations between August 2024 and June 2025. The headline: about 80% of cited URLs sat on .com domains, with .org taking a distant second place and everything else — .io, .ai, .co, .net, ccTLDs like .co.uk and .de — splitting the remaining sliver.
That number sounds like a smoking gun. It isn't. The web itself is heavily .com-weighted, and the sub-corpus that LLMs lean on for grounding (news sites, encyclopedic references, product review outlets, established SaaS blogs, government and NGO explainers) is even more .com-heavy than the web at large. If 75–80% of the plausible source pool is .com, an 80% citation share is roughly what any unbiased ranker would produce.
The .org share is a useful sanity check. Wikipedia, major nonprofits, standards bodies, and .gov-adjacent references are disproportionately .org. LLMs cite them a lot because they're trustworthy references, not because the models decoded the TLD string and awarded bonus points.
Correlation vs causation: why .com wins on paper
Three confounders swamp any direct TLD effect:
- Age. .com predates every alternative gTLD by decades. Age correlates with backlink profile, brand mention volume, Wikipedia presence, and training-data saturation — all of which are plausible AI citation signals. A 22-year-old .com site has had 22 years to accumulate the exact signals LLMs seem to reward.
- Category concentration. Media, retail, and reference sites — the categories AI answers pull from hardest — are overwhelmingly .com. Newer TLDs concentrate in specific niches: .io in developer tooling, .ai in ML products, .co in early-stage startups. Those niches produce fewer of the "definitive source" pages that AI answers cite.
- Mention volume. Ahrefs' 2025 work on AI citations, discussed here, suggests unlinked brand mentions correlate with citation frequency roughly three times more strongly than backlinks. Brands on .com have simply had more time and more press coverage to accumulate mentions.
Google's own documentation, going back years, is explicit that new gTLDs like .ai or .io get no ranking advantage or disadvantage versus .com. There is no published statement from OpenAI, Anthropic, Google DeepMind, or Perplexity that contradicts that stance for their retrieval layers either.
What this means for .io, .ai, and ccTLD brands
If you're on a non-.com, the practical picture is:
- Don't rebrand for GEO reasons alone. The domain-migration cost — redirects, lost link equity, brand confusion, entity re-consolidation in knowledge graphs — will almost certainly outweigh any citation lift, and there's no evidence the lift is real.
- Close the mention gap. If the driver behind .com dominance is mention volume, that's a lever you can pull. Digital PR, podcast appearances, being quoted in industry research, Reddit and Hacker News discussion, and Wikipedia-worthy notability all move this needle. See our note on brand mentions vs backlinks for the mechanics.
- Consolidate your entity. Make sure your Organization schema,
sameAslinks, Crunchbase, LinkedIn, and Wikidata entries all agree on the canonical domain. Fragmented entities lose citations regardless of TLD. - ccTLD nuance. Country-code TLDs (.de, .co.uk, .fr) do carry a geographic association in Google's search index. Whether that affects AI answer geography is untested publicly, but if your audience is single-country, a matching ccTLD is not a handicap — it may actually help for locale-specific queries.
Where the .com premium might be real (a little)
I'd flag one narrow scenario where the extension itself could nudge model behavior: user-facing citation display. Some AI interfaces render the domain next to the citation. If a user sees "acme.com" vs "acme.xyz," click-through and trust perceptions differ — Nielsen Norman-style trust research has consistently found that unfamiliar or novelty TLDs read as less credible to mainstream users. That's a downstream engagement effect, not a citation-selection effect, but it can feed back into the training loop over time via click and dwell data.
For B2B developer audiences, .io and .ai are familiar enough that this penalty is probably negligible. For consumer categories, it's worth thinking about.
How to actually measure this for your brand
Don't rely on TLD-aggregate stats to make a call about your specific site. Instead:
- Track your citation share across ChatGPT, Perplexity, Gemini, and Google AI Overviews for a defined prompt set.
- Benchmark against three to five competitors — ideally a mix of .com and non-.com in your category.
- If your citation share trails a category peer with a similar mention profile, the gap is unlikely to be TLD-driven. If it trails a peer with a much stronger mention profile, focus on mentions, not the domain.
CiteFlow does this tracking natively; you can also assemble a rough version with manual prompt logs plus a scraper.
FAQ
Should a startup pick a .com over a .ai for GEO reasons?
No. Pick the domain that fits your brand, budget, and category conventions. Any AI-citation gap between TLDs is explained by age and mention volume, both of which you build regardless of the extension.
Do AI models actually parse the TLD as a feature?
There's no public evidence that retrieval or ranking layers in the major LLMs use the TLD string as an explicit feature. The observed skew is corpus-driven. Google has publicly stated that new gTLDs get no ranking preference in classic search.
Will migrating from .io to .com boost my citations?
Almost certainly not enough to justify the cost. You'd lose link equity during the redirect period, fragment your entity across two domains in AI training snapshots, and confuse existing brand recall. Fix mentions and entity consolidation first.
