Blog / Brand Mentions Across Models
Research
Brand Mentions in Perplexity, ChatGPT, and Claude: What's Actually Different
The same brand can have 3 different descriptions across 3 AI models. From our server-side data: observed crawler activity and the limits of interpreting it — and which model matters most depends on your audience.
Three Models, Three Versions of Your Brand
We operate one of the largest AI brand directories — 56,251+ brand profiles, available for retrieval. Our server logs record requests to our site, not the complete information consumed by any platform.
The same brand profile page gets crawled by GPTBot, ClaudeBot, and PerplexityBot within the same week — but the resulting brand representation in each model can be meaningfully different. Here's why.
A single 'AI visibility' score obscures critical differences between models. Your brand might be an Incumbent in Perplexity and a Challenger in Claude. Model-level granularity isn't optional — it's the difference between an accurate diagnosis and a misleading average.
Brand mentions are one of the key signals that build Brand Authority — but not all mentions are equal, and not all models weight them the same way.
Historical request counts as originally reported, not current traffic. The listed OpenAI and Anthropic bot subtotals do not sum to their platform totals; the archive does not explain the remainder.
Platform-by-Platform Breakdown
OpenAI (ChatGPT) 10,816/week
Crawlers: GPTBot (8,159), OAI-SearchBot (1,691), ChatGPT-User (515)
Interpretation limit: Requests observed on our site do not measure the balance of trained knowledge and retrieval in the platform’s answers, or prove that fetched content was used.
Practical implication: Review actual answers and citations. Crawl activity alone does not establish answer accuracy or an update timeline.
Anthropic (Claude) 4,669/week
Crawlers: ClaudeBot (4,235), Claude-User (303), Claude-SearchBot (99)
Interpretation limit: Requests observed on our site do not measure the balance of trained knowledge and retrieval in the platform’s answers, or prove that fetched content was used.
Practical implication: Review actual answers and citations. Crawl activity alone does not establish answer accuracy or an update timeline.
Perplexity 1,699/week
Crawlers: PerplexityBot (1,699)
Interpretation limit: Requests observed on our site do not measure the balance of trained knowledge and retrieval in the platform’s answers, or prove that fetched content was used.
Practical implication: Review actual answers and citations. Crawl activity alone does not establish answer accuracy or an update timeline.
The Parametric vs. Retrieved Distinction
The key to understanding model differences is the distinction between parametric and retrieved knowledge.
Parametric knowledge refers to information represented in model parameters. Server logs do not reveal which fetched pages were incorporated into those parameters.
Retrieved knowledge refers to information supplied through retrieval. A fetch does not prove that every answer uses the latest page, or that trained knowledge plays no role.
We have removed the original table of platform knowledge weights and update speeds. Those percentages and timelines were not measured by these logs.
Which models should you test?
Choose models based on evidence about your audience. Our crawler logs do not establish which products enterprise, technical or consumer buyers prefer.
Keep public information accurate
Maintain clear, factual information on your website and the profiles you control. Check actual answers after changes. Neither structured data nor llms.txt guarantees retrieval, correct answers or a training update.