CiteLadder research
Original studies based on data we collect ourselves. Our first report reviews robots.txt files from the Tranco Q2VV4, 2026-11-06 list, with more studies on how AI assistants answer buyer questions to follow.
AI crawlers
How the 10,000 most popular websites treat AI crawlers in robots.txt, based on 6130 sites with a readable policy, where 15.5% block GPTBot.
Website files for AI
How many of the 10,000 most popular websites publish an llms.txt file, based on 6943 sites checked, where 12.4% have a valid one.
Prompt Builder models
How the choice of Claude model and effort level changes the buyer questions Prompt Builder drafts, compared across four models at low, medium and high effort.
Claude recommendations
Which brands Claude Sonnet 5.5 names for 400 buyer prompts across 20 categories, based on 1,200 answers per condition, and what its web searches look for and return.
- Does wording change which brands Claude recommends? 20 questions, 10 rewrites eachCiteLadder asked Claude 20 buyer questions in 10 wording styles, three runs each. See which rewrites moved the brand list and which were noise.
- What Claude searches for, and what it finds, when buyers ask for recommendationsWe asked Claude Sonnet 5.5 400 buyer questions three ways. It ran 5261 searches when told to search and cited inline in 2 answers. Here is what it found.
- Which brands Claude recommends from memory: 1,200 answers across 20 categoriesCiteLadder asked Claude Sonnet 5.5 400 buyer questions across 20 categories, three times each, with no tools. See which brands it named and how consistent it was.