GEO in 2025: what two large studies actually found

A chart-led synthesis of two 2025 studies on AI search citations, source selection, content style, semantic relevance, and what the evidence does not prove.

By dotSuper Research DeskPublished Sep 1, 2026Reviewed Sep 1, 202612 min read
Search & discoveryTwo preprints, 5,060 prompts and paired queries, 98,477 websites, and current Google guidanceUpdated Sep 1, 2026

/ THE SHORT ANSWER

The studies suggest that generative search visibility is shaped by source authority, engine-specific retrieval behavior, semantic relevance, and the readability of source text. They do not reveal a universal GEO formula. The practical response is to publish original evidence, preserve strong SEO foundations, make each page easy to understand, and measure visibility across engines rather than assuming one ranking system.

Key takeaways
  • 01In the sampled brand-ranking prompts, earned sources supplied most citations across all four AI engines, but the exact mix varied materially by engine.
  • 02In the Google AI Overview study, a one-standard-deviation decrease in source perplexity was associated with a nine percentage point increase in predicted citation probability.
  • 03Engine overlap was low enough that a single-platform visibility check can produce a false sense of coverage.
  • 04Google now describes GEO as part of SEO and explicitly rejects special files, excessive chunking, and scaled pages created only to capture query variants.

/ dotSuper point of view

GEO should be run as an evidence and distribution discipline, not a collection of AI-search hacks.

The evidence base at a glance

The first study evaluated how web-enabled Claude, ChatGPT, Gemini, and Perplexity cited sources across consumer ranking prompts, local service queries, languages, and paraphrases. Its core ranking dataset contained 1,000 prompts across ten categories.

The second study compared Google AI Overview citations with conventional Google results for 4,060 queries. It analyzed 98,477 unique websites, then ran controlled retrieval experiments and a 147-participant randomized trial.

RESEARCH SCALE

Two studies, four useful lenses.

The samples answer different questions. Read the figures as complementary evidence, not as one combined experiment.
Ranking prompts1,000

GEO engine study across ten consumer categories

Paired queries4,060

Google organic and AI Overview observations

Unique websites98,477

Website-level sample in the content study

Citation shift+9pp

Predicted probability after 1 SD lower perplexity

A percentage-point change is not the same as a relative percentage increase. The study reports a move from 47% to 56%.

View the chart data
Core study scale and headline effect
MeasureValueStudy context
Consumer ranking prompts1,000Four web-enabled engines plus Google comparison
Paired Google queries4,060AI Overview and first-page organic results
Unique websites98,477Website-level content analysis
Predicted citation probability47% to 56%One standard deviation lower perplexity

Where the papers converge

Both papers point toward retrieval compatibility. A page must be discoverable, relevant to the question, and easy for a system to use as support. The GEO study emphasizes third-party authority, scannable comparisons, explicit claims, and language-aware distribution. The content study finds that lower-perplexity source text and stronger query-to-source similarity are associated with citation selection.

That does not mean every page should be rewritten to sound generic. The content study tested AI polishing inside a controlled retrieval pipeline, while Google advises publishers to prioritize unique, non-commodity, people-first material. The sensible synthesis is clear writing around distinctive evidence.

What each study contributes
QuestionStudy signalPractical interpretation
Who gets cited?Earned sources dominate many sampled ranking promptsBuild evidence other credible sites have a reason to reference
What text gets selected?Lower perplexity and higher semantic similarity correlate with citationAnswer the intended question clearly and keep the supporting evidence close
Does one engine represent all AI search?Source mixes and domains vary by engineMeasure multiple engines and keep results separated
Do special GEO hacks replace SEO?NoMaintain crawlability, indexing, useful content, and accurate structured data

What this changes for dotSuper

dotSuper should use its Feeds library as an evidence system rather than a page-volume system. Each research cluster should begin with one original question, a direct answer, visible methodology, downloadable or inspectable numbers, limitations, and a clear route to the relevant product.

The immediate opportunity is to build a repeatable visibility baseline across Google generative AI reports, ChatGPT, Claude, Gemini, and Perplexity. The baseline should distinguish citation presence, cited URL, source type, answer accuracy, and assisted conversion. A single blended score would hide the engine differences shown in the research.

  • Create original benchmarks from dotSuper delivery work using anonymized and permissioned data.
  • Build one strong topic hub for each product, then connect narrow supporting pages only when each page answers a genuinely different question.
  • Place the direct answer, evidence, definitions, and caveats early in the page.
  • Track Google generative AI impressions in Search Console and maintain a separate monthly multi-engine citation test.
  • Pursue authentic earned distribution through partners, events, associations, customer evidence, and expert contributions.

The five-part dotSuper research series

This report is the overview. Four focused field notes separate the mechanisms so readers and retrieval systems can reach the precise evidence they need.

GEO research series
PageQuestion answeredPrimary metric
AI search source mix by engineWhich kinds of sources does each engine cite?Earned, brand, and social share
Engine overlap and query volatilityHow stable is visibility across engines and prompts?Jaccard overlap and local-result overlap
Content readability and citation probabilityWhat page-level characteristics correlate with citation?Predicted probability and treatment effects
dotSuper GEO measurement playbookHow should this become an operating system?Citation coverage, assisted demand, and freshness

What this page cannot conclude

  • 01Both papers are preprints. Their findings should be treated as evidence to test, not as a settled ranking formula.
  • 02The GEO engine study collected data in August 2025. Models, retrieval systems, interfaces, and citation behavior can change quickly.
  • 03Observed citation patterns do not prove that changing one page element will cause an engine to cite that page.
  • 04Google states that there is no special schema, file, or content format required for its generative AI features. Core SEO and useful, original content remain the foundation.

Sources

  1. 01Generative Engine Optimization: How to Dominate AI SearcharXiv · accessed Sep 1, 2026
  2. 02When Content is Goliath and Algorithm is David: The Style and Semantic Effects of Generative Search EnginearXiv · accessed Sep 1, 2026
  3. 03Optimizing your website for generative AI features on Google SearchGoogle Search Central · accessed Sep 1, 2026
  4. 04Introducing Search Generative AI performance reports in Search ConsoleGoogle Search Central · accessed Sep 1, 2026
  5. 05Google Search's guidance on using generative AI content on your websiteGoogle Search Central · accessed Sep 1, 2026
BUILD A MEASURABLE DISCOVERY SYSTEMGEO in 2025: what two large studies actually found

/ APPLY THE THINKING

Turn evidence into discoverable demand.

dotSuper connects original research, search foundations, AI visibility monitoring, content operations, and qualified lead routing as one accountable system.

Question for the working sessionWhat can these two studies reliably tell a business about visibility in generative search?

/ Topic-led working session · GEO in 2025: what two large studies actually found

Turn this question\ninto a useful first move.

Bring how this question currently shows up in your business: “What can these two studies reliably tell a business about visibility in generative search?” We’ll test the page’s evidence against your context and define the smallest useful next move.

Live availability from ceo@dotsuper.net Your time zone · Local time
  1. 01Bring the contextWhere this issue shows up in the work.
  2. 02Test the relevanceUse the evidence against your reality.
  3. 03Choose the next moveOne accountable action, clearly owned.
Live availability
  1. Date
  2. Time
  3. Booked

Syncing live times