ChatGPT’s Strangest Leak: Cringey Chat Logs Found in Google Analytics
▼ Summary
– Extremely personal ChatGPT conversations have been leaking into Google Search Console, where developers monitor search traffic rather than private chats.
– Starting in September, Google Search Console began showing unusually long queries containing user prompts from private chatbot conversations about sensitive topics.
– SEO consultant Jason Packer and web optimization expert Slobodan Manić investigated and found evidence suggesting OpenAI directly scrapes Google Search with user prompts.
– OpenAI acknowledged awareness of the issue and confirmed resolving a glitch affecting how a small number of search queries were routed.
– Packer expressed satisfaction with the quick resolution but noted OpenAI’s response left uncertainty about whether scraping occurred and if the problem was fully resolved.
For months, a surprising privacy breach allowed highly personal ChatGPT conversations to surface within Google Search Console, a platform typically reserved for analyzing website traffic and search performance. This unusual incident meant that private user prompts, often detailing sensitive relationship advice or confidential business strategies, became visible in a tool never intended for such exposure. The situation highlights ongoing concerns about data security in the age of advanced artificial intelligence.
Starting around September, website administrators began noticing strange entries in their GSC performance reports. Instead of the usual short keyword searches, these entries displayed lengthy conversational prompts, sometimes exceeding 300 characters, that clearly originated from ChatGPT users. These logs contained only the user’s side of the dialogue, revealing how individuals were seeking help for personal and professional matters under the assumption their discussions would remain confidential.
Jason Packer, an SEO and web development expert from Quantable, initially brought this concerning development to public attention through a detailed blog post. Curious about the underlying cause, he collaborated with web optimization specialist Slobodan Manić to investigate further. Their joint testing reportedly uncovered what they consider definitive evidence that OpenAI was directly harvesting Google Search data using actual user prompts. This practice would represent a significant privacy compromise, potentially allowing the AI company to gather engagement-boosting search information that Google doesn’t normally make available to external parties.
When approached for clarification, OpenAI declined to confirm or deny the specific theory proposed by Packer and Manić. The company also didn’t address several outstanding questions that could help determine how widespread the problem might have been. An OpenAI representative did acknowledge awareness of the situation and stated they had fixed a technical glitch that temporarily affected how a limited number of search queries were processed.
Packer expressed satisfaction with OpenAI’s prompt resolution of the immediate problem but noted the company’s response left important questions unanswered. Without explicit confirmation about whether Google scraping occurred, doubts remain about whether the underlying issue has been fully addressed, leaving room for continued uncertainty about user privacy protections.
(Source: Ars Technica)
