Seven weeks of a new browser-game site: what 893 plays and our server logs show
Small-sample notes from the first seven weeks of JoyForGame: which games people played, how often AI assistants looked our pages up, and why AI user-agent strings are not trustworthy on their own.
What the data shows
Three of the four most-played games were sandboxes.
Battle Simulator – Sandbox (83 plays), Road Designer (43) and Last Play. Ragdoll Sandbox (21) are all "set it up, then watch it play out" games. Simulation was the largest category at 22% of all plays, ahead of puzzle at 16%.
Play was spread thin across a long tail.
893 plays were spread over 400 different games. The single most-played game accounted for about 9% of all plays.
AI assistants looked our pages up rarely — and almost only through ChatGPT.
In seven weeks, AI assistants fetched one of our game or walkthrough pages while answering a user 105 times. ChatGPT accounted for 103 of them; Perplexity and Claude for one each.
Many requests claiming to be an AI assistant were not.
Of 284 requests identifying themselves as Claude-User, at least 142 were probes for credential files such as .env and .aws paths; only one fetched a real content page. Of 207 identifying as Perplexity-User, at least 91 were probes and one fetched a content page. Counting AI traffic by user-agent string alone overstates it.