Oddest ChatGPT leaks yet: Cringey chat logs found in Google analytics tool (arstechnica.com)

Researchers and SEO pros found an alarming leak: extremely personal ChatGPT prompts — sometimes over 300 characters and clearly user-authored (e.g., relationship or business advice requests) — were appearing in Google Search Console (GSC), a tool meant to show site search queries. Jason Packer of Quantable and web optimization consultant Slobodan Manić flagged the weird entries, ran tests, and published a detailed theory that the prompts arrived in GSC because OpenAI’s systems were routing or “scraping” Google search data containing user inputs. OpenAI told reporters it was “aware” and has “resolved” a glitch that temporarily affected how a small number of search queries were routed, but declined to confirm whether the investigators’ scraping theory is correct. The incident is significant because it suggests a breakdown in how user prompts and third‑party telemetry are handled: private conversational inputs ended up in a public-facing webmaster analytics tool, exposing sensitive data to site owners and increasing regulatory and trust risks for AI platforms. Key technical notes: leaked items were long free‑text prompts (not short keywords) and appeared only as user inputs; investigators’ tests are presented as the first direct evidence tying those prompts back to OpenAI’s search routing, but the company has not confirmed the mechanism. The episode underscores the need for clearer logging policies, audits of data routing, and stronger guarantees that chat inputs won’t be inadvertently forwarded to external analytics channels.