Most GEO advice assumes an open contest. Improve your schema, restructure your content, earn some citations. It is a reasonable model and it describes traditional search well enough.
It does not describe AI search. The distribution of citations across the web is not merely uneven — it is oligopolistic in a way that has no real precedent in the SEO era, and understanding the shape of it determines whether your GEO program is aimed at something achievable.
The concentration
Analysis from BrightEdge and Ahrefs indicates roughly 40 to 55 percent of ChatGPT Search and Perplexity citations flow to fewer than 1,000 domains. For context, there are over 200 million active registered domains. A set representing well under a thousandth of a percent of the web captures around half of all AI citations.
Reddit is the single largest beneficiary. Ahrefs data puts Reddit in roughly 18% of all ChatGPT Search answers. That is one domain appearing in nearly one in five generated responses.
Two things drive it. OpenAI holds a formal data partnership with Reddit, which guarantees access and likely weighting. And Reddit's native format is structurally close to ideal for retrieval — a question, multiple competing answers, and visible community validation through voting. A retriever handling "which CRM is best for a 20-person team" finds a thread where five practitioners argue about exactly that, with signal about which answer other practitioners endorsed.
Reddit did not optimize for AI citation. It happens to have been built, fifteen years ago, in the exact shape a retriever wants to read.
What this means for your content calendar
If half of citation supply is spoken for by a thousand domains, and your brand is not one of them, then the strategic question is not "how do we get cited more." It is "which of the three available routes are we actually running."
| Route | What it looks like | Effort to start | Who it suits |
|---|---|---|---|
| Borrow the authority | Earn presence on domains already inside the set — authentic participation in relevant subreddits, accurate Wikipedia and Wikidata entries, answers on Stack Overflow, coverage in trade press | Low | Almost everyone. This is the default and most teams underinvest in it badly. |
| Publish what cannot be sourced elsewhere | Original research, proprietary benchmarks, survey data, pricing transparency. A retriever cannot pull a number from Wikipedia if the number only exists on your site. | High | Brands with data access and the willingness to publish it. |
| Own the narrow query | Specific, long-tail, high-intent questions where Reddit has a thin thread and Wikipedia has nothing. Less volume, far less competition, better buyer intent. | Low–medium | B2B and technical categories especially. |
Notice what is absent from that table: publishing more blog posts about your category. Broad category queries are effectively closed. If a buyer asks a general question about your space, that answer is being assembled from the thousand, and another well-optimized page on your own domain does not change the arithmetic.
The platform picture also moved
A second finding worth building into your planning. Single-platform GEO strategy has stopped being defensible, and the change happened fast.
ChatGPT lost 26.5 points of B2B referral share in eight months. Not because it shrank — it reached roughly 900 million weekly active users by February 2026, up from 400 million a year earlier — but because the other surfaces grew faster off a smaller base.
Claude's move from negligible to 18.5% is the part that should change how you allocate. As covered in the measurement framework, Claude's B2B referral share runs roughly seven times its global consumer share. If your prompt set weights by consumer market share, you are systematically undertesting the platform your buyers use for long research sessions.
The cost of being outside the set
One more number that frames the urgency. Non-branded informational query traffic is reported down 15–30% across content sites. That is the measurable cost of citation concentration to everyone outside the thousand: the queries that used to send traffic are now being answered without a click, sourced from domains that are not yours.
This is why I keep pushing clients away from "publish more" as a first move. Publishing more into a saturated citation market produces content that is read by crawlers and cited by nobody. The three routes in the table above are the ones with a mechanism behind them.
What I would do first
Audit where the thousand are winning your queries. Run your 25-prompt set and record not just whether you appear, but which domains do. If Reddit is taking four of your top ten commercial queries, that is a concrete, addressable finding — and it points at community participation rather than another blog post.
Find one thing only you can publish. Pricing you are willing to make transparent, benchmark data from your own operations, a survey of your customer base. One genuinely original dataset outperforms a quarter of category content, because it is the only route that creates citation supply rather than competing for it.
Rebalance testing across four surfaces. If your reporting is ChatGPT-only, you are blind to roughly 37% of the referral market and to the fastest-moving platform in it.
See which domains are taking your citations.
The free 5-prompt snapshot reports your citation rate across all four surfaces and names the domains winning the queries you are not. 48-hour turnaround, no commitment.
Get the free snapshot Full GEO auditHow to cite this
If you quote a figure from this piece, keep its source and its denominator attached. That is the whole request.
ReferenceSummers, J. C. (2026). The Citation Oligopoly: Why Fewer Than 1,000 Domains Own AI Search. Sable Search. https://sablesearch.com/blog/citation-concentration-ai-search/In a sentenceSummers (Sable Search, 2026), citing BrightEdge and Ahrefs data, notes that roughly 40 to 55 percent of ChatGPT Search and Perplexity citations flow to fewer than 1,000 domains.
Frequently asked questions
Analysis from BrightEdge and Ahrefs indicates roughly 40 to 55 percent of ChatGPT Search and Perplexity citations flow to fewer than 1,000 domains, dominated by Reddit, Wikipedia, Stack Overflow, and major news outlets. Most brands are not competing against the open web — they are competing for the citation slots that set has not already claimed.
Ahrefs data indicates ChatGPT Search cites Reddit in roughly 18 percent of answers. OpenAI holds a formal data partnership with Reddit, and Reddit threads provide multi-perspective answers to commercial investigation queries in a format retrievers extract cleanly — a question, several competing answers, and visible community validation through voting.
Yes, but rarely by publishing more brand-owned content. The three viable routes are earning presence on domains already inside the set, publishing original data that cannot be sourced elsewhere, and owning narrow queries where high-authority domains have no specific answer. Broad category queries are effectively closed to new entrants.
Significantly. Goodie panel data measuring B2B AI referrals found ChatGPT's share fell from 89.1 percent in May–August 2025 to 62.6 percent by March–April 2026. Claude rose to 18.5 percent and became the second largest source, Gemini reached 10.6 percent, and Perplexity 7.3 percent.
Jason C. Summers
13+ years enterprise technical SEO and digital product, with weekly LLM citation monitoring running across five platforms in production. Related reading: the GEO measurement framework and the 7 mechanisms behind AI citation.
Sources
- BrightEdge and Ahrefs. AI search citation domain concentration analyses, 2026.
- Ahrefs. AI Search Study — Reddit citation frequency in ChatGPT Search.
- Goodie. 2026 AI Search Traffic Report — Wave 1 (May–Aug 2025) and Wave 2 (Mar–Apr 2026) B2B referral panels.
- Similarweb. AI search traffic data, 25.77B visits, January–April 2026.
- OpenAI. Weekly active user figures, February 2026.