Proof Entry
43,309 GPTBot Requests Into One Crawl Trap
On September 9, GPTBot made 43,309 requests across the BellyUp ATL network while the structure exposed 40,404 filter combinations from only 31 venues.
The finding started as a bot-authenticity question. Across the retained log window, 46,727 of 46,829 GPTBot-classified requests came from 41 of 43 source IP addresses inside OpenAI's published GPTBot ranges.
On September 9 alone, the network recorded 43,309 GPTBot requests. Across the six live properties, the requested URL strings collapsed to 38,971 distinct values when the same route on different hosts was counted once.
The cause was faceted navigation. Four combinable filters plus sorting exposed 40,404 crawlable combinations from 31 venue records. The bars property produced 26,863 combinations from twelve venues.
The machine-readable layer received almost no attention during the retained period. Venue JSON was requested twice, AI foundation files twice, and the AI instructions file once, while 46,504 requests went to filter HTML.
That makes this a structure receipt, not an AI visibility victory.
Evidence
- 46,829 GPTBot-classified requests across the retained period - 46,727 requests from published OpenAI GPTBot IP ranges - 41 of 43 source IP addresses verified against those ranges - 43,309 GPTBot requests on September 9, 2026 - 38,971 distinct requested URL strings across the six live properties that day - 40,404 filter combinations exposed across the network - 26,863 combinations generated from twelve venues on the bars property - 31 venue records in the complete machine-readable dataset - 46,504 filter HTML requests across the retained period - 2 venue JSON requests, 2 AI foundation requests, and 1 AI instructions request
Methodology
The receipt uses the live Digital Karma Data Warehouse event and retained raw request records for the BellyUp ATL hosts. GPTBot identity was checked against OpenAI's published IP ranges rather than accepted from the user agent alone. Request targets were grouped by host, URL string, query-string presence, response, source IP, and crawler fingerprint.
The faceted inventory was counted from the actual filter dimensions exposed by the site. Representative nonsense and empty combinations were requested and compared to confirm the system returned full near-duplicate pages.
A server request proves only that the request reached the recorded URL. It does not prove indexing, training use, an AI citation, ranking influence, search visibility, or human demand.
Outcome
The crawl trap was closed or neutralized across the live category properties, the robots contradiction was corrected, the machine catalog now exposes all 31 venue records, and the missing AI instructions layer was added.
Those changes do not turn the receipt into a conclusion. They create a dated intervention that can be measured against the before state.
The next evidence belongs to the live crawler-redistribution experiment.
See More Evidence
More SEO proof from live websites, each one with the numbers attached.