Live pull from refillhq-db · analytics_events · May 10 – Jun 4, 2026. Tracked events = store-page views, searches, and "Get directions" clicks.
Two things happened at once, and only one of them is real.
1. Most of the raw spike is bots. AI crawlers ramped up hard in late May — ClaudeBot, Amazonbot, and GPTBot are now indexing the site, on top of SemrushBot, Googlebot, and a scraper spoofing an old Chrome. On May 29 & 31 that was 80–87% bots.
2. But there's a genuine human bump too. Real visitors jumped to 337 on Jun 1 and 282 on Jun 3 vs a ~50–80/day baseline. That traffic is almost entirely "direct / no referrer" — the fingerprint of dark social (links shared inside the Facebook/Instagram apps, group chats, DMs, email). The fbclid tags in the data confirm Facebook is the driver — someone shared RefillMap and it spread.
The tall raw bars on May 11, 16, 29–31 and Jun 1 are mostly the red (bot) layer. The teal (human) layer is what actually matters — and it clearly steps up around Jun 1–3.
AI-training crawlers (ClaudeBot, Amazonbot, GPTBot) are the new arrivals — a sign the site is getting indexed as a reference source. Harmless, but they inflate raw counts. Filter them out of any dashboard number you report.
"Direct / dark social" dominates because shares inside the Facebook & Instagram apps strip the referrer — they look "direct" but they're social. The recorded Facebook number is an undercount: people who land on the homepage from FB and don't click into a store page aren't tracked at all.
Right now only store pages, searches, and directions clicks get logged. The homepage, /browse, city/state, and article pages are not tracked at all. So the true source mix is blind to anyone who lands and bounces — which is most social traffic. If you want a real "where are people coming from" chart in the admin, the fix is to log a lightweight pageview (with referrer) on every route, plus strip fbclid/utm into a clean source column. I can wire that into the RefillMap admin analytics tab whenever you want.