What 109 Questions and 439 AI Responses
Reveal About Getting Cited
Who does AI search cite? We ran a GPR (Generative PR) experiment: 109 questions to ChatGPT, 439 responses observed, 4,365 citation URLs collected across 440 unique domains. Here are the 5 conditions that determine whether AI recommends your site.
109
Questions asked
439
AI responses observed
4,365
Citation URLs collected
440
Unique domains
Why we ran this experiment
We had 34 articles on our LP and blog. From an SEO perspective, that should be plenty. But the GPR experiment results were brutal.
Generic query citation rate: 0%
For queries like "companies with proven AEO results" and "how to get cited in AI search" -- our core topics -- we received zero citations and zero mentions. 34 articles, and AI used none of them.
Brand queries ("What is Sighted") showed 100% mention rate and 38% citation rate. AI knows we exist, but doesn't trust us enough to recommend for generic questions. To understand why, we ran a systematic experiment with 109 questions.
Methodology
GPR (Generative PR) experiments send structured questions to AI search engines and statistically analyze citation URLs and mentions in responses. Each question is repeated 3-5 times to observe response variance (non-determinism).
10 fixed seed queries (5 Japanese, 5 English) + 99 AI-generated variants. Brand and generic queries separated by design.
3-5 trials per question. Citation URLs, mention text, and response context all recorded. 439 total observations.
4,365 citation URLs, 440 unique domains. Domain dispersion, category-level citation rates, and citation patterns analyzed.
Finding 1: There is no dominant player
Even the most-cited domain accounts for only 10.6% of all citations. With 440 domains cited, there is no "default" source for AEO and AI search optimization content.
| # | Domain | Citations | Share |
|---|---|---|---|
| 1 | youtube.com | 461 | 10.6% |
| 2 | reddit.com | 221 | 5.1% |
| 3 | note.com | 117 | 2.7% |
| 4 | prtimes.jp | 100 | 2.3% |
| 5 | webanalytics.speee.jp | 61 | 1.4% |
| 6 | lp.sighted-aeo.com | 59 | 1.4% |
| 7 | customs.go.jp | 48 | 1.1% |
| 8 | crevia-ts.com | 44 | 1.0% |
| 9 | 0120.co.jp | 43 | 1.0% |
| 10 | acquia.com | 43 | 1.0% |
What this means
Even the top domain (YouTube) holds only 10.6%. No site dominates. This means there is still time to claim the default position in your Question Space. For most industries, the battle for AI search share hasn't started yet.
Finding 2: What AI actually cites
Categorizing 4,365 citation URLs reveals the content types AI prefers.
Video content (10.6%)
YouTube videos lead all citations. Step-by-step tutorials with visual demonstrations outperform written guides. "I tried it" content beats abstract explanations.
Community discussions (5.1%)
Reddit threads rank second. AI treats real user experiences and comparison discussions as third-party validation.
Personal experience reports (2.7%)
Individual blog posts with first-hand results rank third. AI values "I did this, here's what happened" over corporate marketing copy.
Press releases (2.3%)
Official announcements with data serve as AI information sources. Siteimprove appeared in AI responses from a single press release.
The pattern is clear: AI cites content with specific experience, data, and procedures. Generic "What is X" explainers and product feature lists rarely get cited.
Finding 3: Self-referential content doesn't get cited
Looking at our 34 articles explains why our generic query citation rate was 0%.
- AEO optimization logs #001-#007 (our improvement diary)
- Sighted practical guide (how to use our tool)
- AEO framework (abstract methodology)
All about "us." No general information AI could use to answer questions.
- note.com -- individual experience reports
- YouTube -- tutorials showing specific steps
- PR Times -- official service announcements
Common thread: specific experience, data, and procedures.
Volume wasn't the problem. Perspective was. AI doesn't want to hear about your company. It wants the best answer to the question. Articles that only talk about yourself give AI no reason to cite them.
Finding 4: High-dispersion queries are your best opportunity
We measured how evenly AI distributes citations across domains for each query. Higher dispersion means no established authority -- and maximum opportunity for entry.
| Query | Dispersion | Monthly vol | Score |
|---|---|---|---|
| companies specializing in AI visibility | 0.37 | 3,600 | 24.3 |
| AEO対策で実績のある会社は | 0.47 | 3,600 | 11.5 |
| AI検索で社名が出ない原因と対策 | 0.33 | 3,600 | 16.4 |
| AEO vs SEO for B2B SaaS | 0.33 | 3,600 | 10.9 |
| ChatGPT citation tactics for AEO | 0.30 | 3,600 | 14.7 |
Score = Dispersion x log(monthly vol) x domain strength. Google Keyword Planner, June 2026.
With 3,600 monthly searches for "AEO"-related queries and a dispersion of 0.47 (spread across 14 sites), every major topic has no established default. Now is the optimal time to claim your position.
The 5 conditions for getting cited by AI
Inductively derived from 439 observations.
You have first-party data
Not citations from other sites or general knowledge -- data you collected through your own experiments, research, or measurement. AI prioritizes content it can identify as unique to one source.
You answer the question directly
When someone asks "how to start AEO," don't open with "AEO stands for Answer Engine..." -- start with the steps. AI selects the page that most directly answers the query.
You write from an industry perspective
Not "we did this" but "this is what's happening in the industry." Your improvement diary can become an industry research report with a simple shift in framing. Same data, different perspective.
Your existence is verifiable across multiple sources
Your website alone isn't enough. Press releases, media mentions, social discussions create external signals that AI uses to validate "can I confirm this source elsewhere?"
You target queries with no established default
High-dispersion queries (where AI citations are spread across many domains) signal open territory. It's more effective to claim uncontested queries than to compete for ones already dominated.
What you can do today
- Map your Question Space -- Systematically collect the questions AI gets about your industry and find out who is being cited.
- Find high-dispersion queries -- Where citations are spread across many domains, there is no default. That's your entry point.
- Reframe "our story" as "industry research" -- Your case studies and improvement logs can become industry insights with a perspective shift. Same data, new framing.
- Include first-party data -- Numbers, experiment results, measurements. Information that doesn't exist anywhere else is the single strongest reason for AI to cite you.
- Create external signals via press releases -- Your website alone isn't enough. Use press distribution and media to create sources AI can cross-reference.
See your Question Space in 30 seconds
Enter your URL and see how AI currently perceives your site.
Start free diagnosisAbout the experiment data
Data in this article comes from Sighted's proprietary GPR (Generative PR) experiment framework. We targeted ChatGPT (GPT-4o) as of June 2026, running 109 questions with 3-5 trials each for 439 total observations. AI search responses are non-deterministic; results may vary by timing and model version. Search volumes reference Google Keyword Planner data from June 2026.