Research & Data
Original research on AI search behavior, citation patterns, and brand visibility. Insights from millions of AI-generated responses.
Current Research
Current Research
The Anatomy of an AI Citation
How AI models construct citations. Schema markup, FAQ pages, and structured data - which page elements earn citations and which get ignored.
Hidden Prompts in AI Search
We analysed 833,791 domains and found 7,029 sites embedding hidden prompts that instruct AI chatbots to cite, remember, or prioritize their content.
AI Crawls Your Product Pages. It Cites Your Blog.
The first Page Type Performance Matrix: 337K AI citations cross-referenced with 11.4M crawler visits reveal a systematic mismatch.
The llms.txt Effect
We HTTP-scanned 37,894 AI-cited domains from a corpus of 103K. 13.3% have llms.txt. The citation advantage? Zero.
The Model Divergence Report
Same question, different AI, different answers. How 8 major AI models disagree on which brands to recommend - and what it means for your visibility strategy.
When AI Comes to Your Website
A behavioral analysis of AI crawlers in the wild. 575,788 visits across 84 brands reveal how ChatGPT, Claude, and other AI systems discover and index your content.
How AI Translates Your Questions
An analysis of how AI transforms user prompts into search queries. Only 0.17% exact match rate. 33% complete rewrites.
Where AI Gets Its Answers
An analysis of citation patterns across 1.3M+ AI responses. Which sources do AI systems trust? Which domains appear most frequently?
Further Reading
Our research builds on a growing body of academic work exploring how LLMs form recommendations, exhibit brand preferences, and respond to content signals. These independent studies provide the theoretical foundation for what we measure at scale.
Content optimization increases LLM visibility by 30-40%. Statistics and quotations provide the strongest lift.
Princeton / KDD 2024Social proof consistently boosts recommendation rates. Scarcity and exclusivity signals surprisingly reduce visibility.
NTUA Athens / EMNLP 2025LLMs systematically favor global brands over local ones and exhibit significant country-of-origin effects.
University of South Florida / EMNLP 2024Different LLMs recommend distinct products with low overlap. 567K samples across models confirm systematic preferences.
Xi'an Jiaotong University, 2025Simply paraphrasing a prompt can cause up to 100% difference in which brands get mentioned.
Carnegie Mellon / CHI 2025Alignment training causes models to overweight majority preferences, reducing recommendation diversity.
MIT CSAIL / ICLR 2025