GEO Lab #04 · S4-GROK-Q2-R2 · Original response

This page presents one selected response from GEO Lab #04. It adds a smaller reading and citation entry point; it does not replace the full archive or collect a new answer.

The OpenAI arm uses the Responses API with web search and does not represent consumer ChatGPT Search. Grok records identify their actual consumer-interface mode and available model label.

Recorded conditions

Record ID
S4-GROK-Q2-R2
Platform
Grok
Question / condition
Q2
Repeat
2
Model label
Exact backend version undisclosed
Visible mode
Expert
Reasoning
Not independently selectable
Submitted / first recorded time
2026-11-03T09:14:58.699Z
Completeness
Recorded body complete
Search
Actual search activity observed; exact controls and evidence remain in the raw record.

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

Credible experiments on GEO (Generative Engine Optimization) tactics and AI brand mentions/recommendations are limited but growing. The strongest evidence comes from controlled academic/lab studies on content features (when a source is already in context or retrieved) and a smaller number of industry tests with some controls. Pure observational correlations (e.g., brand mentions vs. visibility) are more common but weaker for causality.

1. Foundational Controlled Study: Aggarwal et al. (KDD 2024)

The seminal paper “GEO: Generative Engine Optimization” (Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, Deshpande; arXiv:2311.09735, accepted at KDD 2024) introduced the term and the first systematic benchmark.⁠arXiv

Setup: GEO-bench (~10,000 diverse queries across domains, with associated web sources). They tested a simulated generative engine (retrieval + LLM) and a real one (Perplexity.ai). Content was rewritten with 9 tactics via an LLM-based black-box optimizer.
Tactics tested: Quotation Addition, Statistics Addition, Cite Sources, Fluency Optimization, Authoritative tone, Easy-to-Understand, Unique Words, Technical Terms, and Keyword Stuffing (as a baseline).
Results: Content-focused rewrites improved visibility metrics (e.g., position-adjusted word count and subjective impression scores) by up to ~40%. Quotation Addition was strongest (~+41–43% on the main metric from a ~19 baseline), followed by Statistics Addition and Cite Sources (~+25–33%). Keyword stuffing showed little or no benefit (or slight declines). Gains were larger for lower-ranked sources. Domain effects varied. Real-world Perplexity tests showed similar directional lifts (quotes/stats/citations helping, stuffing not).⁠Jrkdm

This is the cleanest interventional evidence that specific on-page content enhancements (evidence-rich elements like quotes, stats, and citations) can increase a source’s share of generated answers once it is retrieved/in context. It does not directly measure organic discovery or brand-level recommendation rates in the wild.

2. Competitive Citation Drivers: Vishwakarma et al. (2026)

“What Gets Cited: Competitive GEO in AI Answer Engines” (arXiv:2605.25517) used a highly controlled two-document RAG testbed.⁠arXiv

Setup: Exactly two candidate sources injected into model context across six LLMs; 252,000 trials with brand anonymization and counterbalanced order. Sources differed on one factor at a time (factorial design over 18 content factors).
Key findings: Topical relevance and list/position were the strongest drivers of being cited first. Explicit price information and a recent timestamp helped consistently. Completeness and trust cues gave smaller gains. Formatting-only edits had little impact.

This isolates what makes one already-retrieved source preferred over another for citation.

3. Null Result on Schema Markup: Synthetic-Brand RCT

A 2026 Roskilde University (RUC) Master’s thesis by Christopher Mads Hammerum Bouet ran a randomized controlled trial with eight synthetic low-authority running-shoe brands.⁠Rucforsk.ruc

Setup: Four brands received Schema.org markup; four identical controls did not. Mentions measured across prompts on Gemini 2.5 Flash with Google Search grounding.
Result: No detectable effect on mention rate (effect size essentially zero; 95% CI roughly −2.7 to +2.5 percentage points). The authors concluded schema never reached the model because low-authority pages were not retrieved. This challenges industry claims of large schema lifts and suggests correlations are often confounded by authority/retrieval.

Related observational work (e.g., Chen & Koudas, University of Toronto preprint arXiv:2610.15033) finds that within-site, page features like FAQ blocks, schema, summary boxes, or speed explain very little citation variation (~2.8%) compared with whether the engine’s own search retrieves the page (~32%). Brand parametric knowledge also drives many mentions without any supporting citation.⁠arXiv

4. Industry Experiments with Partial Controls
OtterlyAI off-page cold-start (2026): Created a fictional zero-authority GEO agency (7 pages) and generated ~16 external mentions via directories, listicle placements, Reddit, and LinkedIn. Within 14 days it reached ~10% share of voice and rank ~#7 among competitors on tightly matched prompts (e.g., location + year-specific). Broader prompts showed near-zero results. Platforms differed (stronger on some, weaker on others). Directional evidence that targeted third-party mentions can bootstrap early visibility.⁠Otterly
OtterlyAI publication-date update (2026): Updated only the visible “last updated” date on 10 articles vs. 2 controls. Over 30 days, citations rose ~31.5% on treated articles while controls declined. Authors described it as directional; external commentary noted pre-existing upward trends and mixed engine results.⁠Airanklab
Search Engine Land experiments (2026): Two structured tests (one multi-month for an existing brand, one 30-day cold-start for a SaaS agency) tracking 15 commercial-intent keywords across 4–6 platforms. Listicle placements on sources already cited by AIs, PR/guest posts, and related outreach produced measurable increases in appearances/citations (hundreds of events logged). One listicle-focused source drove the bulk of citations. Results varied by platform and keyword; authors noted that conventional advice did not always hold and that sources mentioning the brand mattered heavily.⁠Searchengineland
Supporting Observational Evidence

Large correlational analyses (Ahrefs ~75k brands) show unlinked brand web mentions correlating strongly with AI visibility (~0.65–0.71), often stronger than backlinks or Domain Rating. Seer Interactive’s GEO Olympics study (~231k responses) linked stronger “signal architecture” (entity authority + third-party validation + community discussion) to ~7.8× more mentions. Kumar et al. and others document a clear brand-stature ladder (household names appear far more often than niche ones). These are not interventional experiments.

Important Limitations Across the Field

AI responses are highly stochastic—many prompt–engine pairs flip between mentioning a brand or not across repeated runs. Mentions and citations are distinct (brands are often named from training knowledge without citing a page). Retrieval is a major bottleneck; on-page tactics mainly matter after a page is found. Effects are platform- and query-specific. Many industry case studies lack strong controls or report absolute numbers that can be influenced by external trends. A 2026 critical survey of dozens of GEO studies concluded that evidence for stable, cross-platform causal effects on organic discoverability remains thin, while effects on already-retrieved content are clearer.

Bottom line: The most credible measured effects come from content quality enhancements (quotes, statistics, citations, relevance, freshness signals) in controlled retrieval settings (Princeton and competitive RAG studies) and from targeted third-party mentions in early industry tests. Schema alone has failed controlled tests on low-authority brands. Stronger longitudinal, multi-platform RCTs with proper controls are still relatively scarce.

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 answer-body links (8)

Retrieval-only / source-panel inventory (5)

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

Assistant read full native rendered answer; matching prompt checked; DOM body and expanded citation groups captured.

Body review

{
"reviewer": "assistant",
"complete_body_read": true,
"coherent_ending": true,
"inline_url_coverage_checked": true,
"external_claims_verified": false,
"independent_human_double_coding": false
}

Existing public-text privacy policy

{
"body_changed": false,
"policy": "Only actual URL tracking, signed/session URL parameters and necessary identifying information are removed; original wording, including analytics parameter examples outside URLs, remains.",
"annotation_positions": "Original position indexes omitted because URL/privacy edits can change character locations."
}

Prompt SHA-256: 84f18194f3ea8dd08151f46577cb5614af08d775b76bd516de9d149a7fc01118
Answer SHA-256: 22f905cf904cd747994d2c2971d804f8d7d8ebef389fe400dbdb9481c6054945