Publications
The method behind the numbers on this site. Preprints are on arXiv, data is on Zenodo under CC BY 4.0, and every DOI resolves to the archived files rather than to us.
Preprints
- arXiv:2607.23893 · July 2026
Who Gets Named: Citation Type Predicts Individual Naming by Grounded Language Models, and a Roster Instrument Captures 0.5% of It
In categories where the buyer picks a person rather than a firm, what a model cites decides who it names. 2,400 buyer-intent answers across four European markets.
- arXiv:2607.13304 · July 2026
Where Does the Noise Come From? A Variance-Components Decomposition of Non-Determinism in LLM Brand Answers
Ask a model the same question twice and the answer moves. This separates how much of that movement is the model, the prompt wording and the run, and says how many times you actually have to ask.
Plain-language write-up: The Dice Roll Method - arXiv:2606.25787 · June 2026
How Large Language Models Source Brand Reputation Across Languages and Markets
167,551 URL-grounded citations across 128 brands, 12 home markets and 13 languages. 85.7% of what AI says about a brand is grounded in sites the brand does not own.
Plain-language write-up: Where AI Gets Its Answers - arXiv:2606.23165 · June 2026
The Language Blind Spot: How Query Language and Brand Recognition Tier Shape AI-Constructed Brand Reputation Across Twelve European Languages
35,640 grounded answers about 66 brands in 12 languages. Reputation is language-bound: monitoring in English alone does not describe what buyers in other markets are told.
- arXiv:2606.23057 · June 2026
Who Owns the AI Recommendation? A Multi-Industry Empirical Map of Brand Category Ownership Across Large Language Models
3,750 responses over 50 brands and 5 industries, asked without naming anyone. Introduces the Category Ownership Index and finds competitive vacuums in 8% of buying questions.
Open datasets
- 10.5281/zenodo.21904654
Does an AI describe your company, or your company's name? 9,600 model answers about invented and real companies
264 real companies across 21 markets, plus invented names matched for length and phonology.
- 10.5281/zenodo.19225834
Supplementary Materials: Measuring Corporate Reputation in the Age of AI
A multi-industry probe of how models describe listed companies.
- 10.5281/zenodo.21612690
Individual Professional Visibility in Grounded LLM Answers: 2,400 Buyer-Intent Responses Across Four European Markets
The data behind "Who Gets Named".
- 10.5281/zenodo.20829524
How LLMs Source Brand Reputation Across Languages and Markets: A Cross-Market Citation Dataset
Citations classified by domain and source type, across 12 markets.
- 10.5281/zenodo.20794390
Cross-Language AI Brand Reputation: A 66-Brand, 12-Language Dataset of LLM-Constructed Reputation
Grounded answers about the same brands asked in twelve languages.
- 10.5281/zenodo.20788142
Category Ownership Map (COI/CVI/DS): A Multi-Industry Dataset of Brand Recommendations
Brand-free category questions and the brands models answered with.
- 10.5281/zenodo.19225835
Supplementary Materials: How Large Language Models Source Brand Reputation Knowledge
The cross-industry precursor to the citation datasets above.