Crawler observation limits
A product that 'sees GPTBot on your blog' from a script tag is describing CDN logs it does not have. Motif reports tracker and collector hits, classifies a versioned token list, and stops.
The request that never comes
GPTBot, ClaudeBot, PerplexityBot, and CCBot typically request HTML. They do not run /m.js. Motif never hears those fetches. Empty Insights crawler state means no observed tracker hits, not 'nobody crawled you.'
The request Motif can see
If a known token hits GET /m.js or POST /v1/events, Motif upserts crawler_daily by website, UTC day, exact agent, family, class, surface, and classifier version. Hits are not visitors. They cannot win.
Self-reported identity
User-Agent is attacker-controlled. OpenAI, Anthropic, Perplexity, Amazon, and Common Crawl publish IP lists. Motif does not verify those lists in v1. Directories that mark a hit 'verified' from the token alone are overselling.
Policy tokens
Google-Extended and Applebot-Extended are robots.txt product tokens, not HTTP User-Agents. Motif's classifier omits them on purpose. Listing them as bots beside GPTBot confuses operators.