Google has been quietly indexing publicly shared conversations from its AI chatbot Bard. The results in Google were spotted by Gagan Ghotra, an SEO consultant who shared them on X (formerly Twitter).
He warned users about Google indexing shared Bard conversation URLs, advising them not to share personal information to prevent potential access by others who find these conversations through Google, in a tweet, adding that some of the results are ranking as snippets (top results) on the search engine.
Google’s decision to index these links is, to say the least, absurd. Even if this was a slip by the company, it raises significant concerns about privacy and data security. Many observers and experts have already expressed privacy concerns about generative AI products due to their ability to handle, process, and in some cases use personal data, in training large language models.
The response by Google SearchLiason, an official company account on X, suggests that the pages were indexed as a result of an error, “Bard allows people to share chats, if they choose. We also don’t intend for these shared chats to be indexed by Google Search. We’re working on blocking them from being indexed now.”
It is not clear how the pages got indexed. Even at the time of writing, and after Google SearchLiason’s response was posted, the Google search for public conversations from Bard returns over 300 results in different langauges.
Peter J. Liu, a Research Scientist at Google, who appears to be working on the company’s generative AI products, responded to Gagan’s tweet, remarking, ‘That is only for explicitly [sic] shared conversations. Your conversations are not public by default.”
As Gagan also contended in his response to the Google employee, users are unaware that the conversation will be indexed by Google when they create a public link. “I was thinking of it as a feature to share conversations with friends or colleagues, and it being visible only to people who have the conversation URL.”
In response, the Research Scientist at Google stated, “Good feedback; I think you have a point.”
The share feature on Bard was introduced in July, along with some other major updates. Earlier this month, Google made additional enhancements to this feature as part of a big update, enabling anyone to continue chatting with Bard based on shared conversations.
When a user clicks or taps on the ‘Share’ button on Bard, the chatbot shares a screenshot of the chat, offering users a preview of how the publicly shared chat will appear. It also presents a disclaimer: “Let anyone with the link see what you’ve selected. You can remove the link in your public links. Remember that you have agreed to Google’s Terms of Service and the Generative AI Additional Terms of Service.”
“The ‘Learn More’ button, located towards the end of the warning, directs users to the ‘Share your Bard chat page‘ within Bard Help. This page provides comprehensive details about the implications of creating a public link, the process for removing it, and other related information.
At the beginning of the page, there is a warning labeled as important and heighted in bold : “If you choose to share a chat, anyone with the link can read the chat, reshare it with others, and continue the chat with Bard on their own.”
The page also advises users against sharing confidential information in their Bard conversations or any data they would prefer to keep private, as it may be accessible to reviewers and could be used by Google for improving its products, services, and machine-learning technologies.
But nowhere on the page, Google has shared any information or warning that the public conversations could be indexed by Google. It doesn’t reflect well on Google when the company is now outright requesting users to allow its AI chatbot, which has grown into the third most widely used generative AI product in the world, within months of its launch, to scan their emails, documents, and other types of data to enable enhanced functionality.