GEO Lab #01 · G-Q2-R1 · 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

Record ID
G-Q2-R1
Platform
Grok
Question / condition
Q2
Repeat
1
Model label
Not disclosed (exact model version)
Visible mode
Expert
Reasoning
See model/mode evidence in the raw record
Submitted / first recorded time
2026-10-29T09:36:13.543Z
Completeness
Selected complete public answer; capture and independent verification limits are preserved below.
Search
observed

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 show that certain GEO tactics can increase a brand’s share of visibility or citations once a source is already retrieved into an AI model’s context, but there is little strong evidence of stable, cross-platform causal effects on organic discoverability (getting retrieved in the first place) or on brand mentions/recommendations at scale. A 2026 critical survey of 45 studies reached this conclusion.⁠Geoaura

Foundational Controlled Study: Princeton GEO Paper

The most-cited rigorous experiment is “GEO: Generative Engine Optimization” by Aggarwal et al. (Princeton, Georgia Tech, and others; arXiv:2311.09735, presented at KDD 2024).⁠Collaborate.princeton

Method: GEO-bench with ~10,000 real queries. Content modifications were applied to sources already placed in a generative engine’s context (primarily an open-source GPT-3.5-style system, with some validation on Perplexity). Visibility was measured mainly via position-adjusted word count (share of the answer text attributable to the source, weighted by position).
Results (relative lifts on the main metric, approximate):
Expert quotations: +41%
Statistics addition: +33%
Fluency optimization: +29%
Citing sources: +28%
Quotation addition: +21%
Keyword stuffing: −8% (harmful)
Lower-ranked sites saw larger relative gains (up to ~115% in some cases for fifth-ranked pages under certain tactics). The widely quoted “up to 40%” figure refers to this conditional share-of-answer improvement, not a guaranteed increase in mention/citation probability or traffic.
Key caveat: Effects are conditional on the source already being retrieved. Later analyses and replications note that some volume-controlled tests of the same tactics showed little or no pooled effect when relevance is controlled, and the gains are better understood as redistributive within fixed context rather than pure discovery lifts.⁠Geoaura
Controlled Tests of Specific Tactics (Mostly Null for Schema)
Ahrefs schema experiment (2026): Difference-in-differences analysis of 1,885 pages that added JSON-LD schema vs. ~4,000 matched controls. Adding schema produced no meaningful citation lift (Google AI Mode +2.4%, ChatGPT +2.2% — both within noise; Google AI Overviews −4.6%). Observational correlations (cited pages more often have schema) did not hold as causal effects for already-visible pages.⁠Ahrefs
Synthetic-brand RCT on schema (Bouet, 2026): Randomized controlled trial with eight invented low-authority running-shoe brands (four with Schema.org markup, four controls). On Gemini with Google Search grounding, schema brands showed essentially identical mention rates to controls (~8.8%). The study ruled out large industry-claimed gains and attributed prior correlations to confounding by authority. Low-authority pages were rarely retrieved, so markup never reached the model.⁠Rucforsk.ruc
Competitive / Factorial Experiments on Citation Preference

A 2026 controlled study (“What Gets Cited,” arXiv:2605.25517) ran 252,000 trials across six LLMs. It injected exactly two competing sources into context (differing in one factor at a time), with brand anonymization and order counterbalancing.⁠arXiv

Strongest drivers of being cited first: topical relevance and list position (near-deterministic).
Consistent helps: explicit price information and recent timestamps.
Smaller gains: completeness and trust cues.
Negligible: pure formatting edits.

This reinforces that content factors matter mainly in competition after retrieval.

Critical Survey Synthesis

Olivier Martinez’s 2026 critical survey (arXiv:2607.14035) reviewed 45 studies (2023–2026). It found solid evidence that tactics can change how an already-retrieved page is cited or used, but no reviewed technique showed a stable, longitudinal, cross-platform causal effect on organic discoverability or downstream behavior. Relevance and position are the most reproducible levers; generic heuristics transfer poorly and can erode under competition. The Princeton gains are valid in their fixed-context setting but do not establish broader discovery effects.⁠arXiv

Practical / Quasi-Experimental Work
Search Engine Land structured experiments (2026): Two before/after tests (existing brand over months; cold-start SaaS agency over 30 days) tracking commercial keywords across multiple platforms. Placing listicles and other content on sources already commonly cited by AI produced clear increases in appearances and citations (hundreds of citation events; cold-start yielded 298 appearances). Earned third-party placements outperformed owned content; concentration on a few sources was extreme. These are not randomized but are among the more transparent real-world measurements of off-site tactics.⁠Searchengineland
Large observational baselines (e.g., Kumar et al. arXiv:2606.20065 on 100k+ responses; Seer Interactive’s 231k+ response GEO Olympics study) document brand-stature ladders, the dominance of third-party/earned sources for recommendations, and signal architecture effects (entity authority + third-party validation + community discussion), but they are primarily correlational rather than interventional experiments.
Overall Takeaways from the Evidence
Tactic Category	Evidence Strength for Increasing Mentions/Citations	Notes
On-page evidence (stats, expert quotes, citations, fluency)	Moderate (conditional on retrieval)	Princeton and related; mainly share-of-answer once in context
Schema / structured data	Weak / null	Ahrefs DiD + synthetic RCT
Off-site placements / listicles on already-cited sources	Practical positive signals	Quasi-experiments; third-party mentions strongly correlated in large studies
Pure formatting or keyword stuffing	Weak or negative	Multiple studies

Bottom line: GEO tactics that make content more extractable, verifiable, and relevant can improve how a brand is used if the model already retrieves the material. Getting retrieved (and becoming a known entity that models mention or query for) appears driven more by existing authority, third-party mentions, and retrieval match than by isolated on-page “GEO tricks.” Rigorous causal evidence for large, generalizable lifts in brand recommendations remains limited.

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 (8)

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

Rendered browser answer DOM + screenshot

Capture provenance

{
"original_status": "completed",
"selected_capture_id": "G-Q2-R1-legacy-01",
"method": "Rendered browser answer DOM + screenshot",
"captured_at": "2026-10-29T09:38:29.582Z",
"same_conversation": true,
"review_notes": "Original export retained; no recovery selection.",
"selected_capture_file": "data/records/G-Q2-R1.json"
}

Search evidence

The UI reported 3 searches performed, followed by 2 searches performed.

Record notes

A fresh web conversation received the original question once, with no follow-up. Existing account-personalization settings were not changed.

Prompt SHA-256: 84f18194f3ea8dd08151f46577cb5614af08d775b76bd516de9d149a7fc01118
Answer SHA-256: 8771a16165ebf20d3c20e0233f17e1f64a067e9088f9a2948355ec832800984c