Published research · Open data · CC-BY-4.0

The AI Visibility Study.

A cross-platform audit of 2,729 businesses measuring who generative AI systems actually mention when buyers ask for recommendations — and which off-page signals predict it. Published May 2026 as an open working paper with a citable DOI.

2,729
Businesses audited
266,844
Observations
5
Generative AI systems
69.5%
Mentioned by none of them
What the data shows

Four findings that change how visibility works.

FINDING · 01

Seven in ten businesses are invisible to AI.

69.5% of audited businesses were not mentioned by any of the five systems tested — ChatGPT, Claude, Perplexity, Google Gemini, and Google AI Overviews. When a buyer asks an AI assistant for a recommendation, most businesses simply do not exist in the answer.

FINDING · 02

Each AI system has its own visibility logic.

Off-page signals show model-specific effects: what correlates with visibility in one system does not necessarily correlate in another. Aggregated analyses obscure these patterns — which means one-size-fits-all "AI SEO" advice is built on averages that describe no single system.

FINDING · 03

Domain Authority is not the predictor it was assumed to be.

Domain Authority — long treated as the primary predictor of search visibility — showed limited predictive power in multivariate analysis. The signals that made a business rank are not the same signals that make a business get recommended.

FINDING · 04

Community and entity signals move the needle where DA doesn't.

Some systems showed small but robust positive partial correlations with off-page presence on platforms like Reddit, Wikipedia, and professional directories — evidence that entity-level and community signals are part of how AI systems decide who to name.

Methodology

Audited, not surveyed.

The study audited 2,729 businesses across multiple markets, generating 266,844 observations of whether — and where — each business appeared in the answers of five generative AI systems. Off-page presence signals were then tested against visibility outcomes per system, rather than in aggregate.

Systems testedChatGPT · Claude · Perplexity · Google Gemini · Google AI Overviews
Sample2,729 businesses, multiple markets
Observations266,844
PublishedMay 8, 2026 · Zenodo working paper · v1
LicenseCreative Commons Attribution 4.0 (CC-BY-4.0)
DOI10.5281/zenodo.20076380
Cite this study

House, J. (2026). Off-Page Signals Have Model-Specific Effects on Generative AI Search Visibility: Evidence from a Cross-Platform Audit of 2,729 Businesses Across Five Generative AI Systems. Zenodo. https://doi.org/10.5281/zenodo.20076380

The paper and data are published under CC-BY-4.0 — free to use, quote, and build on with attribution. For interviews, commentary, or the dataset behind a specific figure, email joel@joelhouse.com.

Where the research runs in production

The findings became the operating playbook for MentionLayer — the GEO platform that earns the citations, reviews, and entity signals this study shows AI systems respond to. Read the thinking in The Answer Era Is Here.

Follow the research.

New findings, methods, and the next waves of the audit — sent to a small list as they land.

Unsubscribe anytime · Never shared