The Reality of the GEO Era — Is AI Citing Your Site?
The Reality of the GEO Era — Is AI Citing Your Site?
You worked your way to page one of Google. ChatGPT still does not know your site exists. In 2026, “being found in search” and “being chosen by AI” have become two entirely different games.
#1 on SEO, ignored by AI?
Working in Japan recently, I came across an interesting dataset. Over the course of 2025, COOD tracked which Japanese domains the major generative AI systems — ChatGPT, Perplexity, and others — cited as “trustworthy primary sources.”
#1 Wikipedia (28.4%), #2 note.com (19.2%).
As an IT engineer, my intuition said “for tech blogs, isn’t Qiita #1?” Qiita came in 8th. The creator platform note was crushing the tech-specialist Qiita. Why?
Pulling on that thread, the outlines of a different game from SEO (Search Engine Optimization) began to emerge. That game is GEO — Generative Engine Optimization.
What is GEO?
GEO (Generative Engine Optimization) is the practice of optimizing content so that AI-based search engines such as ChatGPT, Perplexity, and Google AI Mode will cite or mention it.
The concept was first formalized academically in a 2024 KDD paper, “GEO: Generative Engine Optimization” (a joint study from IIT Delhi and Princeton). The researchers built a benchmark of 10,000 queries (GEO-bench) and found that applying GEO techniques improved AI search visibility by up to 40%.
The key finding: including statistical data increases AI visibility by 22%, and inserting quotations increases it by 37%. Factors entirely different from the keyword density and backlinks that traditional SEO prized are what drive AI’s choices.
What the numbers say: the gap between SEO and GEO
“We’re on page one of Google, so of course AI cites us, right?” — The data refutes this assumption.
Google rank ≠ AI citation
According to Ahrefs’ global analysis, only 12% of pages cited by ChatGPT appear in Google’s Top 10. ChatGPT cites pages ranked 21 or lower on Google about 90% of the time.
Google AI Overview has somewhat higher overlap with organic results (about 54.5%), but even there nearly half of citations come from outside the Top 10. BrightEdge’s 16-month tracking dataset confirms it.
The collapse of clickthrough rate
Seer Interactive’s analysis of 42 organizations and 3,119 informational queries returned a shocking result.
- For queries that trigger AI Overview, organic CTR dropped 61% (1.76% → 0.61%)
- Paid ad CTR dropped 68% (19.7% → 6.34%)
- Even on queries that do not trigger AI Overview, CTR fell 41% — meaning users are migrating to ChatGPT and Perplexity entirely
Meanwhile, sites cited in AI Overview saw organic clicks rise 35% and paid clicks rise 91%. Whether AI chooses you or not now decides the life and death of your traffic.
Publishers’ traffic in freefall
According to a Press Gazette/Chartbeat study, global publishers’ Google search inbound traffic fell by a third in the year through November 2025. Business Insider’s organic search traffic was down 55%, and HuffPost lost half.
The global AI citation map: who gets chosen
So where is AI actually pulling information from? Combining Semrush’s analysis of more than 230,000 prompts and over 100 million citations with SurferSEO’s analysis of 36 million AI Overviews and 46 million citations yields the following.
Top citation sources by platform (global)
| Rank | ChatGPT | Google AI Mode | Perplexity |
|---|---|---|---|
| 1 | |||
| 2 | Wikipedia | YouTube | |
| 3 | NIH (U.S. National Institutes of Health) |
Reddit, Wikipedia, YouTube, LinkedIn, and Google’s own properties — these five domains account for 38% of all AI citations. The probability that an individual company’s official site is cited directly is vanishingly small.
Worth noting: in September 2025, ChatGPT’s citation patterns shifted abruptly. Reddit citations crashed from about 60% of responses to 10%, and Wikipedia citations from 55% to below 20%, with PRnewswire, Forbes, and Medium picking up the slack. AI platform citation policies can change at any time.
AI Overview presence by industry (BrightEdge, 2025→2026)
AI Overview is spreading rapidly across every industry.
- Healthcare: 88%
- Education: 83% (up from 18% the prior year)
- B2B technology: 82% (up from 36%)
- Restaurants: 78% (up from 10%)
- Insurance: 63%
The AI Overview appearance rate against all queries climbed from about 31% in February 2025 to 48% in February 2026 — a 58% year-over-year increase. The direction is unmistakable. A world in which more than half of all searches are answered directly by AI is arriving.
The distinctive terrain of the Japanese market
Working in Japan, I cannot ignore the gap between global and Japan-local data. Here is what Ahrefs Brand Radar found analyzing four AI search products in the Japanese market from May to September 2025.
Top 3 citation sources in Japanese AI search
| AI engine | #1 | #2 | #3 |
|---|---|---|---|
| ChatGPT | Reddit (228K) | Wikipedia EN (166K) | Ameblo (132K) |
| Google AI Mode | YouTube (92K) | Google.com (35K) | Wikipedia JP (33K) |
| Copilot | Wikipedia JP (126K) | Bing.com (93K) | note (65K) |
| Perplexity | Yahoo!知恵袋 (519K) | YouTube (277K) | Wikipedia JP (252K) |
There is overlap with the global trend (Wikipedia, YouTube, Reddit), but Japan-specific platforms hold a strong presence.
- Yahoo!知恵袋 — Perplexity leans on it overwhelmingly. 519,000 citations is two to five times the count seen for other AIs. That a Q&A platform analogous to Korea’s Naver Knowledge iN occupies this much weight for an AI says a great deal.
- Ameblo (Ameba Blog) — ChatGPT cites it as the #3 platform in Japan. Japan’s largest blog service, with rich lifestyle and experience-based content.
- note — #3 in Copilot, #2 overall in the COOD study. A creator platform full of “stories only that person can write” is being recognized by AI as a primary source.
- PR TIMES — #5 in the COOD study. A corporate press release platform that is becoming a trusted source for AI.
Meanwhile, one of the COOD study’s central findings: “Even sites with high human traffic are ignored by AI all year long if they merely repackage information.” Traffic and AI citation are separate games.
Why note > Qiita
Why the gap — Qiita 8th, note 2nd?
- Breadth of subject matter — Qiita specializes in IT/programming (vertical); note spans every field (horizontal). AI must answer all kinds of questions, so it cites broader platforms more often.
- The uniqueness of primary information — According to the COOD study, AI is reluctant to cite AI-generated content. It prefers irreplaceable context — “lived trial and error, independent reflection, on-the-ground experience.” note is dominated by essay and personal-experience writing; Qiita’s tech tips and code snippets are comparatively replaceable.
- Domain-level authority — note.com is one of Japan’s largest creator platforms, used by tens of millions monthly, and the domain itself carries authority.
How do you measure GEO?
“OK, GEO matters — but how do you measure it?” The metrics in industry use today are roughly these.
Core GEO metrics
| Metric | Description |
|---|---|
| AI Share of Voice | Share of AI responses to a relevant query that mention your brand |
| Citation Rate | Frequency with which your site is cited as a source in AI responses |
| AI Visibility Rate (AIGVR) | Share of AI-generated responses in which your brand appears |
| Content Extraction Rate | Share of your content actually reflected in AI answers |
| Share of Model (SoM) | Share of AI appearances by your brand vs. competitors |
The measurement tool ecosystem
Tooling in this space is exploding. llmrefs.com lists more than 200 AEO/GEO/LLMO tools.
Enterprise-grade:
- Ahrefs Brand Radar — Tracks 320M prompts, covers six AI platforms. 699/month
- Semrush Enterprise AIO — Multi-platform monitoring of ChatGPT, Google AI Mode, Perplexity
- Profound — Tracks ten-plus AI engines (ChatGPT, Claude, Perplexity, Gemini, Copilot, DeepSeek, Grok, etc.). $499+/month
Specialist tools:
- Otterly.ai — Over 20,000 users; GEO audits with 25+ evaluation factors. 189/month
- SurferSEO AI Tracker — Database-driven, tracking 36M AI Overviews and 46M citations
Free tools:
- HubSpot AEO Grader — Free brand presence analysis across ChatGPT, Perplexity, Gemini
This is not a mature market. Different tools track different AI platforms and metrics, with very different data collection methodologies. But the direction is clear. The era of only tracking “Google rank” is ending, and the era of measuring “how often AI cites you” is opening.
In the GEO era, academic research is revealing AI’s biases
Research on what information AI selects is accumulating. A few findings worth noting.
The “Matthew effect,” amplified
A paper presented at NAACL 2025 (“Large Language Models Reflect Human Citation Patterns with a Heightened Citation Bias”) analyzed citation patterns in 166 CS papers (AAAI, NeurIPS, ICML, ICLR) for GPT-4, GPT-4o, and Claude 3.5. The finding: LLMs cite already-heavily-cited papers more strongly than humans do. The “Matthew effect” — the rich get richer — is amplified by AI.
This carries over to corporate content. Already-authoritative domains (Wikipedia, YouTube, etc.) get cited more by AI, which reinforces their authority — a virtuous cycle for them, a vicious one for latecomers.
AI search is “more diverse, but more concentrated”
A study published on arxiv in December 2025 compared 55,936 queries across six LLM-based search engines and two traditional search engines. LLM-based search cites 37% more diverse domains than traditional search, but 80% of responses converge on fewer than 10 unique URLs. The breadth widened, but the depth concentrates on a small set of “trusted sources.”
Also notable: 82% of Grok responses and 38% of Gemini responses contain no source website at all.
So what should you do?
After the data is laid out, the question returns: “What needs to change in practice?“
1. Your own site is not enough
Rather than visit “your site” directly, AI pulls information about you from authoritative third-party platforms. Globally, that role is played by Reddit, Wikipedia, YouTube, and LinkedIn. In Japan, add note, Yahoo!知恵袋, Ameblo, and PR TIMES.
Action point: Invest not only in your own site’s SEO, but systematically build presence (mentions, citations, reviews) on these third-party platforms.
2. Make “irreplaceable primary information”
What AI ignores: information scraped and repackaged from elsewhere, AI-generated content, shallow keyword-stuffed posts.
What AI selects: original data, analysis grounded in real experience, on-the-ground voices, structured expert information.
Borrowing the COOD study’s phrase: “Trial and error, independent reflection, on-the-ground experience — context only you can write” is the condition for content AI will choose.
3. Structured data is table stakes
For AI to extract your content correctly, technical readiness matters. Schema.org markup, FAQ structuring, clear heading hierarchy — these mattered in SEO too, but in GEO they map directly onto “Can AI parse my content?” and so they have a more direct impact on outcomes.
4. Start measuring
“You can’t improve what you don’t measure.” GEO is no different. Whether you start with the free HubSpot AEO Grader or run an Otterly.ai GEO audit, at minimum you need to know “how often AI is currently citing our brand.” That is the first step.
One realistic caveat: AI search is still a “research channel,” not a “conversion channel.” Per BrightEdge data, direct conversions from AI search are near zero. AI citation is a path to brand awareness and trust — it is not yet a direct path to revenue. Make your investment calls with that understood.
Closing: from “being searched” to “being chosen”
Ten years ago, “are you found on Google?” decided the life and death of online business. Now, that paradigm is shifting. “Are you chosen by AI?” is becoming the new standard of visibility.
Zero-click searches have crossed 60% (and 77% on mobile), and the AI Overview appearance rate has reached 48%. ChatGPT handles 2.5 billion requests a day, and Perplexity has grown 370% year over year. The direction cannot be reversed.
But this shift does not mean “SEO is dead.” 54.5% of Google AI Overview citations still come from top organic search results. SEO is the foundation, and GEO is a new layer stacked on top.
As a Korean engineer working in Japan, I feel this change daily in both countries. Japan’s note, Yahoo!知恵袋, PR TIMES; Korea’s Naver Knowledge iN, Namuwiki, Tistory — research into how each country’s distinctive platform ecosystem participates in AI citation is still in its infancy. But what is certain is that the era of “build great content and they will come” is over.
To be chosen by AI, you have to know where AI is looking. And you have to be there.
References and sources
Academic papers:
- Aggarwal et al., “GEO: Generative Engine Optimization” (KDD 2024) — arxiv.org/abs/2311.09735
- “Large Language Models Reflect Human Citation Patterns with a Heightened Citation Bias” (NAACL 2025) — arxiv.org/abs/2405.15739
- “Source Coverage and Citation Bias in LLM-based vs. Traditional Search Engines” — arxiv.org/abs/2512.09483
Global research:
- Semrush, “The Most-Cited Domains in AI: A 3-Month Study” — semrush.com
- Ahrefs Brand Radar, “The 10 Most Mentioned Domains Across 78.6M Searches” — ahrefs.com
- SurferSEO, “AI Citation Report 2025” — surferseo.com
- Seer Interactive, “AIO Impact on Google CTR” — seerinteractive.com
- BrightEdge, “AI Search Visits Surging / Rank Overlap After 16 Months of AIO” — brightedge.com
- Press Gazette, “Global Publisher Google Traffic Dropped by a Third” — pressgazette.co.uk
Japan market research:
- COOD, “AIに指名される企業 — 引用元ランキング日本版” — prtimes.jp
- Ahrefs, “AI検索が引用する情報源トップ10【日本版】” — ahrefs.com/blog/ja
- Ahrefs, “日本におけるAI検索エンジン4社の引用元を徹底分析” — commercepick.com
GEO measurement tools:
- Ahrefs Brand Radar — ahrefs.com/brand-radar
- HubSpot AEO Grader (free) — hubspot.com/aeo-grader
- Otterly.ai — otterly.ai
- Profound — tryprofound.com
- llmrefs.com, “200+ AEO/GEO/LLMO Tools” — llmrefs.com