GEO Lab #01 · C-Q2-R2-A3 · Original response
This page presents one selected response from GEO Lab #01. It adds a smaller reading and citation entry point; it does not replace the full archive or collect a new answer.
Recorded conditions
Original prompt
What credible experiments have measured whether GEO tactics increase AI brand mentions or recommendations? Include sources when available.
The agreed prompt is reproduced exactly. Any recorded difference in the actual submitted text or search UI is retained in the raw record and full appendix.
Original answer
What credible experiments show about GEO and AI brand visibility As of October 30, 2026, the evidence supports a qualified conclusion: GEO tactics can influence what AI systems cite and recommend under controlled conditions, but reliable, lasting increases in a real brand’s organic visibility across AI search engines remain difficult to establish. The key is to distinguish three outcomes: Brand mentions: Does ChatGPT, Gemini, Claude, or Perplexity name a brand? Citations: Does the answer link to a brand’s website or a third-party page mentioning it? Recommendations: Does the AI actually recommend the brand or product to the user? These metrics are related, but an improvement in one does not establish an improvement in the others. 1. The strongest evidence, ranked by experimental rigor 1. Princeton University — GEO: Generative Engine Optimization (2024) Peer-reviewed research Finding: GEO methods improved visibility by up to 40% in the researchers’ experimental setup. Princeton University +1 The researchers introduced GEO-bench, a benchmark of queries and relevant web sources, and compared optimized content with unoptimized content. Methods included adding relevant statistics, quotations, authoritative citations, and other content modifications. Important limitation: The experiment operated within a controlled retrieval-and-generation pipeline. It demonstrates that changing content can improve how a retrieved source is represented in an AI answer. It does not establish that the same edits will make a brand more likely to appear in organic ChatGPT or Google AI search results. Source: Aggarwal et al., KDD 2024. Paper and publication details 2. Competitive GEO in AI Answer Engines (May 2026) Large controlled experiment Finding: Topical relevance and the position of a source in the provided context were the strongest drivers of being cited first. Explicit pricing and recent timestamps also helped consistently. arXiv +1 The researchers ran 252,000 trials across six LLMs, using paired comparisons in which candidate documents differed by one of 18 content factors. Brand anonymization and counterbalanced source order helped isolate the effects of content from brand and position biases. Why it matters: This is unusually strong evidence for specific citation-selection mechanisms. It also suggests that superficial formatting changes alone are unlikely to be a reliable GEO strategy. Limitation: The researchers supplied the competing documents directly to the model. The study primarily measures citation selection after retrieval, not whether publishing a page causes an AI search engine to discover it. Source: What Gets Cited: Competitive GEO in AI Answer Engines — preprint. 3. What Drives AI Citations? (October 2026) Recent empirical research Finding: Pages whose titles and paragraphs match users’ questions are cited more often. The authors also found that retrieval behavior explains substantially more variation than page-level characteristics alone. arXiv The study examined two 90-day collections covering 158,099 answers and 468,037 citations across commercial queries. It combined observational analysis with API-based replay experiments. Its central insight is that citation visibility depends heavily on which pages the engine retrieves for its own reformulated queries. Brand familiarity also matters: models sometimes name brands without a supporting page being cited. Why it matters: It separates two different optimization problems: getting a page retrieved and getting a brand mentioned. Improving on-page content may help with the first, but cannot guarantee the second. Source: What Drives AI Citations? Content, Retrieval and Model Knowledge in GEO — preprint. 4. SafeGEO — AI product recommendations (2026) Controlled recommendation study Finding: GEO-style manipulations increased the rate at which targeted, flawed products entered recommendation sets by up to 83.2 percentage points in the tested scenarios. The researchers constructed 600 recommendation cases and tested 22 manipulation variants. Why it matters: It shows that content manipulation can affect actual recommendations, a more consequential outcome than citation frequency alone. Limitation: This is a study of recommendation vulnerabilities under controlled conditions, including potentially misleading content. It does not establish that ordinary, truthful optimization will produce comparable gains in commercial AI search. Source: SafeGEO: Understanding Generative Engine Optimization Risks in Recommendation Agents — preprint. 2. What real-world brand studies tell us The controlled studies above isolate specific mechanisms. These additional studies examine brands and live AI search systems, but their findings are mostly correlational or observational. Study Sample and finding What it establishes Ahrefs, May 2025 75,000 brands. Brand web mentions correlated with Google AI Overview visibility at approximately 𝑟 = 0.664 r=0.664. Ahrefs +1 Brand mentions are associated with AI visibility; causality remains unproven. Ahrefs, December 2025 75,000 brands. YouTube mentions had a reported correlation of approximately 𝑟 = 0.737 r=0.737 with AI visibility across the studied platforms. Ahrefs YouTube presence is a promising signal to investigate, not a proven causal intervention. GEO at Scale, June 2026 100,000+ AI responses across 100+ brands. Initial visibility differed sharply by brand maturity: 73% for major household names, 44% for established mid-market brands, and 11% for niche brands. 数字对象标识符 Brand stature is an important baseline factor; the study does not prove that any particular GEO tactic causes growth. Search Engine Land, September 2026 Two practical experiments tracked 15 commercial-intent keywords across multiple AI platforms, logging 775 citation events. The first campaign saw listicles and PR placements account for most citations. Search Engine Land Third-party placements may be valuable, but the uncontrolled campaign design cannot isolate their causal effect. The Ahrefs results are useful because of their scale. However, correlation coefficients cannot tell us whether creating more brand mentions will increase AI visibility. Already-famous brands may receive more mentions and more AI recommendations for the same underlying reason: existing brand recognition. The June 2026 GEO at Scale paper is similarly useful as a baseline. Its authors explicitly propose further protocols to test causal interventions rather than claiming that the descriptive findings prove which tactics work. 数字对象标识符 3. Which GEO tactics have the best evidence? Improve topical relevance and answerability Strongest direct evidence Match page titles and substantive paragraphs to real user questions. Make claims specific, relevant, and easy to verify. Controlled citation experiments support relevance as a meaningful factor. arXiv +1 Provide concrete facts, pricing, and current information Controlled evidence Explicit prices and recent timestamps improved citation selection in the competitive GEO experiment. The size and transferability of those gains in live organic search remain uncertain. arXiv Earn third-party mentions, reviews, and listicle placements Promising, not yet causal Real-world studies show associations between brand mentions and AI visibility, and campaign reports show third-party listicles can attract citations. Robust evidence that deliberately acquiring such placements causes sustained cross-platform visibility gains is still limited. Ahrefs +1 Add statistics, quotations, and authoritative citations Context-dependent These techniques improved visibility in the original GEO benchmark. Their effectiveness depends on the domain and experimental setup; they should not be treated as universal ranking factors. Princeton University +1 Assume that more pages, FAQ blocks, schema markup, or formatting automatically increase mentions Insufficient causal evidence. These changes may help with accessibility, retrieval, or interpretation in particular contexts, but their independent effect on brand mentions and recommendations has not been established consistently. 4. Why the evidence is harder to interpret than it looks Three methodological issues explain why many GEO success stories overstate what their data proves. 1. Retrieval versus generation. If a page is never retrieved, improving its formatting may have no effect. A test that gives a page directly to an LLM isolates generation behavior but does not measure organic discoverability. 2. Stochastic outputs. The same question can yield different answers, citations, and recommendations across runs. A study published in April 2026 specifically recommends repeated measurements rather than judging visibility from a single response. arXiv 3. Confounding and competition. A brand might gain visibility because it became more famous, earned more media coverage, or improved its traditional search presence. And even if one brand gains recommendations, competitors may lose them. Controlled recommendation research highlights this competitive effect. alphaXiv +1 A particularly relevant recent finding comes from an Axios report published October 28, 2026, covering a Muck Rack and Generative Pulse study. Across repeated ChatGPT queries on the same day, only 41% of cited domains overlapped. This illustrates why single-run comparisons are unreliable. Axios 5. How to test GEO credibly yourself For an independent builder testing whether a tactic improves a brand's AI visibility [personal context omitted], I would use the following protocol. A small, defensible GEO experiment Define one intervention. For example, publish a substantive third-party article about your work, or rewrite a page to answer a specific user question. Avoid changing several things at once. Pre-register 20–50 relevant prompts. Include branded queries, non-branded discovery queries, and recommendation queries. Keep their wording fixed for the primary analysis. Establish a baseline. Run each prompt multiple times across the selected AI engines and record brand mentions, recommendation inclusion, citations, and cited URLs. Use controls. Compare treated pages or topics with similar untreated pages or topics. Where practical, stagger publication dates to help separate intervention effects from general time trends. Measure over time. Repeat the observations for at least several weeks. Record engine version, date, location settings, and whether web search was enabled. Report uncertainty. Calculate changes in mention and recommendation rates, not just raw counts. Include variation across runs, platforms, and prompt categories. The primary metric should be the change in the proportion of eligible prompts that mention or recommend the brand, compared with the baseline and a suitable control. Citation rate should be reported separately. Bottom line The most defensible conclusions from the available evidence are: Content relevance and source position can influence citation selection in controlled tests. Brand mentions, third-party coverage, and YouTube presence correlate with AI visibility, but the causal effect of increasing these signals remains uncertain. Recommendation manipulation can change model outputs, although that does not establish that legitimate marketing tactics will yield equivalent gains. No single tactic has yet been shown convincingly to produce stable, sustained improvements in organic brand recommendations across major AI platforms. For further reading, start with the Princeton GEO paper, then the competitive citation experiment and the October 2026 retrieval study. Together, they provide a stronger foundation than vendor playbooks based solely on correlation or isolated success stories.
Recorded sources
These are the source roles recorded in the public dataset. A retrieval or source-panel item is not automatically an answer-body citation. A URL supplied in the prompt and repeated by an answer does not prove content access. No destination has been repaired or independently validated here.
Recorded sources, with original attribution roles (16)
- Paper and publication detailsanswer_link · doi.org
- What Gets Cited: Competitive GEO in AI Answer Enginesanswer_link · arxiv.org
- What Drives AI Citations? Content, Retrieval and Model Knowledge in GEOanswer_link · arxiv.org
- SafeGEO: Understanding Generative Engine Optimization Risks in Recommendation Agentsanswer_link · arxiv.org
- Princeton University GEO: Generative Engine Optimization 2024年8月24日citation_panel_match · collaborate.princeton.edu
- 数字对象标识符 GEO: Generative Engine Optimization | Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining 2024年8月24日citation_panel_match · doi.org
- Paperity What Gets Cited: Competitive GEO in AI Answer Engines (pdf) 2026年5月25日citation_panel_match · paperity.org
- Ahrefs An Analysis of AI Overview Brand Visibility Factors (75K Brands Studied) 2025年5月26日citation_panel_match · ahrefs.com
- Rhino Rank GEO link building: which sources AI answers cite, study by study 2025年7月30日citation_panel_match · www.rhinorank.io
- Ahrefs Top Brand Visibility Factors in ChatGPT, AI Mode, and AI Overviews (75k Brands Studied) 2025年12月12日citation_panel_match · ahrefs.com
- 数字对象标识符 [2606.20065] Generative Engine Optimization at Scale: Measuring Brand Visibility Across AI Search Engines 2026年6月18日citation_panel_match · doi.org
- Search Engine Land Two GEO experiments challenge conventional AI visibility advice 2026年9月14日citation_panel_match · searchengineland.com
- arXiv Don't Measure Once: Measuring Visibility in AI Search (GEO) 2026年4月9日citation_panel_match · arxiv.org
- alphaXiv Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems 2026年6月16日citation_panel_match · www.alphaxiv.org
- alphaXiv SafeGEO: Understanding Generative Engine Optimization Risks in Recommendation Agents 2026年9月4日citation_panel_match · www.alphaxiv.org
- Axios GEO starts with the question, not the answer 昨天 — A new Muck Rack and Generative Pulse study suggests companies should begin generative engine optimization (GEO) by understanding what their audiences ask, rather than simply trying to secure AI citat…citation_panel_match · www.axios.com
Retrieval-only / source-panel inventory (34)
- www.axios.com GEO starts with the question, not the answer 昨天 — A new Muck Rack and Generative Pulse study suggests companies should begin generative engine optimization (GEO) by understanding what their audiences ask, rather than simply trying to secure AI citations. Researchers tested 1,020 mostly general consumer prompts across ChatGPT, Claude, Gemini and Perplexity, finding that cited sources varied substantially by question and platform. The results challenge the idea that news coverage is always the most reliable path to AI visibility. Citation patterns were also inconsistent over time: when researchers repeated the same questions on ChatGPT on the same day, only 41% of cited domains overlapped, making isolated tests a poor measure of success. The study combined prompt testing with broader monitoring of mostly brand-related queries, covering more than 40 million responses. Its central takeaway is that GEO is unpredictable and continually changing. Communications teams should prioritize the information their audiences seek and assess visibility across relevant questions and platforms, rather than investing in content solely because they hope a particular AI system will cite it.retrieved_source · www.axios.com
- arxiv.org What Drives AI Citations? Content, Retrieval and Model Knowledge in Generative Engine Optimization (GEO) 2026年10月11日retrieved_source · arxiv.org
- searchengineland.com Two GEO experiments challenge conventional AI visibility advice 2026年9月14日retrieved_source · searchengineland.com
- www.alphaxiv.org SafeGEO: Understanding Generative Engine Optimization Risks in Recommendation Agents | alphaXiv 2026年9月4日retrieved_source · www.alphaxiv.org
- www.frictionai.co The Personalization Gap: Account History and Brand Recommendations in Generative AI 2026年8月29日retrieved_source · www.frictionai.co
- arxiv.org GEO-Flag: Detecting and Measuring GEO-Optimized Web Content 2026年8月18日retrieved_source · arxiv.org
- somethinginc.com AI citation evidence review: what the studies show 2026年8月14日retrieved_source · somethinginc.com
- arxiv.org Optimizing Visibility in Generative Engines: A Critical Survey of Generative Engine Optimization (2023-2026) 2026年7月16日retrieved_source · arxiv.org
- www.semrush.com Semrush Releases Expanded 2026 AI Visibility Index, Analyzing 126 Million AI Search Prompts 2026年6月26日retrieved_source · www.semrush.com
- doi.org [2606.20065] Generative Engine Optimization at Scale: Measuring Brand Visibility Across AI Search Engines 2026年6月18日retrieved_source · doi.org
- academ.us [2606.20065] Generative Engine Optimization at Scale: Measuring Brand Visibility Across AI Search Engines - Academus scientific article reader 2026年6月18日retrieved_source · academ.us
- www.alphaxiv.org Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems | alphaXiv 2026年6月16日retrieved_source · www.alphaxiv.org
- arxiv.org SafeGEO: Understanding Generative Engine Optimization Risks in Recommendation Agents 2026年6月9日retrieved_source · arxiv.org
- wegamans.net SafeGEO: Understanding Generative Engine Optimization Risks in Recommendation Agents 2026年6月8日retrieved_source · wegamans.net
- www.xxkt.org.cn 普林斯顿大学2024年KDD会议论文GEO英文版-小小课堂网 2026年6月5日retrieved_source · www.xxkt.org.cn
- arxiv.org What Gets Cited: Competitive GEO in AI Answer Engines 2026年5月25日retrieved_source · arxiv.org
- mlsanity.com What Gets Cited: Competitive GEO in AI Answer Engines · mlsanity 2026年5月25日retrieved_source · mlsanity.com
- paperity.org What Gets Cited: Competitive GEO in AI Answer Engines (pdf) | Paperity 2026年5月25日retrieved_source · paperity.org
- aixiv.science 1. Introduction 2026年4月28日retrieved_source · aixiv.science
- arxiv.org Don't Measure Once: Measuring Visibility in AI Search (GEO) 2026年4月9日retrieved_source · arxiv.org
- ahrefs.com Top Brand Visibility Factors in ChatGPT, AI Mode, and AI Overviews (75k Brands Studied) 2025年12月12日retrieved_source · ahrefs.com
- www.semrush.com 3 Months of AI Visibility Index Data: Here's What We've Learned About AI Search 2025年11月4日retrieved_source · www.semrush.com
- arxiv.org Generative Engine Optimization: How to Dominate AI Search 2025年9月11日retrieved_source · arxiv.org
- doi.org [2509.08919] Generative Engine Optimization: How to Dominate AI Search 2025年9月10日retrieved_source · doi.org
- www.rhinorank.io GEO link building: which sources AI answers cite, study by study | Rhino Rank 2025年7月30日retrieved_source · www.rhinorank.io
- ahrefs.com Google Seems More Biased Towards Big Brands Than ChatGPT and Perplexity 2025年7月7日retrieved_source · ahrefs.com
- ahrefs.com An Analysis of AI Overview Brand Visibility Factors (75K Brands Studied) 2025年5月26日retrieved_source · ahrefs.com
- ahrefs.com An Analysis of AI Overview Brand Visibility Factors (75K Brands Studied) 2025年5月26日retrieved_source · ahrefs.com
- collaborate.princeton.edu GEO: Generative Engine Optimization - Princeton University 2024年8月24日retrieved_source · collaborate.princeton.edu
- doi.org GEO: Generative Engine Optimization | Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining 2024年8月24日retrieved_source · doi.org
- app.cochat.ai GEO: Generative Engine Optimization · CoChat 2024年1月1日retrieved_source · app.cochat.ai
- www.fifthring.com GEO: Generative Engine Optimization KDD ’24, August 25–29, 2024, Barcelona, Spainretrieved_source · www.fifthring.com
- geosource.ai Research | GeoSource.ai - GEOSourceretrieved_source · geosource.ai
- www.researchgate.net (PDF) Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systemsretrieved_source · www.researchgate.net
Evidence and qualifications
This is the existing redacted public record. It has not been rewritten for this page. Earlier privacy edits and collection qualifications continue to apply. Editorial text is English; a few original captures retain non-English UI badges or source-panel date labels. An answer’s empirical claims are platform output, not verified findings of this archive.
Capture method
DOM-backed share-dialog preview; no public share created in this capture
Capture provenance
{
"original_status": "completed",
"selected_capture_id": "C-Q2-R2-A3-recovery-01",
"method": "DOM-backed share-dialog preview; no public share created in this capture",
"captured_at": "2026-10-30T05:00:31.096Z",
"same_conversation": true,
"review_notes": "Recovered the same conversation already submitted by Cursor; no new question was sent. The original page showed today 0:06 and the reply menu today 0:08 (UI date labels translated here); submission time has minute precision. The answer has a complete ending. The evidence limits for model settings and generation metadata are documented.",
"selected_capture_file": "data/records/C-Q2-R2-A3.json"
}
Search evidence
The original question has a Web Search marker. View sources contains 14 cited-source entries and 34 additional retrieval results.
Record notes
Recovered the same conversation already submitted by Cursor; no new question was sent. The original page showed today 0:06 and the reply menu today 0:08 (UI date labels translated here); submission time has minute precision. The answer has a complete ending. The evidence limits for model settings and generation metadata are documented.
Prompt SHA-256: 84f18194f3ea8dd08151f46577cb5614af08d775b76bd516de9d149a7fc01118
Answer SHA-256: c62f9ee2b14f792f9a01878be75add68ac40a5b8bf420f88efe4851763ca663a