OpenAI’s response leaves users with “lingering questions”
After ChatGPT prompts were found surfacing in Google’s search index in August, OpenAI clarified that users had clicked a box making those prompts public, which OpenAI defended as “sufficiently clear.” The AI firm later scrambled to remove the chats from Google’s SERPs after it became obvious that users felt misled into sharing private chats publicly.
Packer told Ars that a major difference between those leaks and the GSC leaks is that users harmed by the prior scandal, at least on some level, “had to actively share” their leaked chats. In the more recent case, “nobody clicked share” or had a reasonable way to prevent their chats from being exposed.
“Did OpenAI go so fast that they didn’t consider the privacy implications of this, or did they just not care?” Packer posited in his blog.
Perhaps most troubling to some users—whose identities are not linked in chats unless their prompts perhaps share identifying information—there does not seem to be any way to remove the leaked chats from GSC, unlike the prior scandal.
Packer and Manić are left with “lingering questions” about how far OpenAI’s fix will go to stop the issue.
Manić was hoping OpenAI might confirm if prompts entered on https://chatgpt.com that trigger Google Search were also affected. But OpenAI did not follow up on that question, or a broader question about how big the leak was. To Manić, a major concern was that OpenAI’s scraping may be “contributing to ‘crocodile mouth’ in Google Search Console,” a troubling trend SEO researchers have flagged that causes impressions to spike but clicks to dip.
OpenAI also declined to clarify Packer’s biggest question. He’s left wondering if the company’s “fix” simply ended OpenAI’s “routing of search queries, such that raw prompts are no longer being sent to Google Search, or are they no longer scraping Google Search at all for data?
“We still don’t know if it’s that one particular page that has this bug or whether this is really widespread,” Packer told Ars. “In either case, it’s serious and just sort of shows how little regard OpenAI has for moving carefully when it comes to privacy.”
