Research
Four questions, studied in the open.
The questions this lab is working on: LLM retrieval and citation, AI crawlers, generative engine optimization, and retrieval evaluation.Every experiment ends in a write-up on the blog — method, data, and what didn't work included.
LLM retrieval & citation
How assistants pick their sources: what gets retrieved, what gets cited, and how that differs across models and question types.
- Which page-level signals correlate with being cited in AI answers?
- How do answer engines choose between overlapping sources?
- How stable are citations for the same question over time?
AI crawlers & indexing
What GPTBot, ClaudeBot, PerplexityBot and friends actually fetch, how they respect (or ignore) directives, and how crawling translates into visibility.
- What do AI crawler hit patterns look like on a real site?
- Which robots directives change behavior in practice?
- How long is the path from crawl to citation?
Generative engine optimization
Separating measured effect from cargo-cult advice: which on-page changes demonstrably shift inclusion in AI answers.
- Do structure, schema, or wording changes move AI answer inclusion?
- Can GEO effects be isolated with controlled page experiments?
- Which popular GEO claims survive a fair test?
Retrieval quality & evaluation
The engineering side: embeddings, chunking, and ranking — and small, honest, reproducible benchmarks for judging them.
- How much do chunking strategies really change retrieval quality?
- What does a minimal, reproducible retrieval benchmark look like?
- Where do embedding models disagree, and does it matter?