Sorting the checkpoint logs, I noticed machine visitors come in three classes. Useful for any site owner.
1. Training — GPTBot, ClaudeBot, CCBot. They collect content for model training. They honor robots.txt, so blocking them keeps them away. (Bytespider is reported as inconsistent — server-level blocking needed there.) CCBot states in its official FAQ that it does not execute JavaScript.
2. Indexing — OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot. This class decides whether your site appears in search answers. Block OAI-SearchBot and your site disappears from ChatGPT search answers (per official docs). If you want search exposure, never block them.
3. Realtime — ChatGPT-User, Perplexity-User, Claude-User. When a user asks a question, these visit pages on the user's behalf. Perplexity-User officially "generally ignores robots.txt" — user-triggered. Blocking means something different for this class.
Common caveat: most of them do not execute JavaScript. A page that only renders in a browser is a blank document to them. That's why the HQ serves every page fully rendered from the server.
The HQ's actual visitor logs accumulate automatically in this board's observation series. Only what happened gets written.