Initiating a dialogue with ChatGPT likely does not conjure expectations of those exchanges being searchable via Google—but for several users, that scenario became a reality. As reported by TechCrunch, by searching with the term “site:https://chatgpt.com/share,” one could unearth actual transcripts of discussions conducted with OpenAI’s chatbot. Conversations that were presumed to be private were accessed as easily as online recipes or tech tips.

It’s not surprising that many of these uncovered dialogues featured sensitive or potentially embarrassing exchanges. For instance, TechCrunch highlighted an instance where a user solicited help from ChatGPT for refining a resume tailored to a specific job application—details that TechCrunch could ascertain based on the conversation’s content. Another individual posed queries to ChatGPT that, according to TechCrunch, resembled discussions from an “incel forum,” although the specifics were not disclosed.

Before anxiety sets in, there are several important clarifications. For starters, OpenAI has discontinued the functionality that allowed chats to be public on search engines. Currently, searches yield no results for ChatGPT discussions, meaning any new chats initiated carry no risk of being exposed in this manner. Notably, the conversations that were revealed were only discoverable because users had deliberately opted into that feature. To make a chat publicly searchable, one needed to click the “share” button, select a “create link” option, acknowledge a warning indicating that personal details and additional messages would remain private, and then toggle a switch to permit its visibility on search engines.

Wayback Machine Archives

After the story broke, researchers from Digital Digging revealed that some of these shared conversations had even been archived via the Wayback Machine. They identified approximately 110,000 ChatGPT threads retrievable through this platform. Despite OpenAI’s determination to eliminate these chats from public search results, many remained accessible to those in the know. Digital Digging’s findings were published on August 1, and at that moment, OpenAI had not yet submitted a removal request to the Wayback Machine regarding the chatgpt.com/share links. However, by August 4, the Wayback Machine indicated that this domain would be excluded, suggesting that OpenAI eventually sought to have these records removed.

Why was this feature introduced initially? OpenAI expressed to TechCrunch, “We’ve been testing ways to facilitate sharing valuable conversations while ensuring user control, and we recently ended an experiment that allowed chats to appear in search engine results with explicit user consent.”

This statement may seem vague, yet it’s not difficult to infer potential advantages for OpenAI. Increased visibility of ChatGPT could prove beneficial for the organization. As the online landscape shifts towards both AI-based solutions and discussion-based answers, such as AI Overviews and Reddit, it appears OpenAI may see an opportunity to leverage this trend. If someone finds ChatGPT’s response particularly helpful, they might decide to share it with search engines, subsequently allowing other users to benefit. Consequently, a similar query could fetch that ChatGPT exchange among the top search results, alongside relevant AI Overviews or pertinent Reddit discussions.

OpenAI isn’t alone in exploring the public aspect of AI interfaces. Recently, it was revealed that Meta AI has also introduced a feature enabling users to publicly share their queries and output—not for search purposes, but to the accessible Meta AI feed. It seems AI organizations are increasingly keen on broadcasting AI-generated outputs, whether they involve chatbot dialogues or artworks produced by AI.

While it’s reassuring to know that your recent ChatGPT discussions won’t appear in Google’s search results, it’s vital not to assume that your interactions with any chatbot are inherently private. In fact, there’s a strong likelihood that the organization behind your chosen bot is leveraging your conversations for training purposes, or that human moderators may have access to review them.

Depending on the specific chatbot, various privacy settings could potentially safeguard users. For instance, ChatGPT offers an option titled “Improve the model for everyone,” which governs whether ChatGPT can utilize user conversations to enhance its model. However, disabling this option does not prevent the company from retaining chat logs. Even transient conversations, which do not show up in user history, can remain on the ChatGPT servers for up to 30 days.

Given these considerations, it is wise to refrain from using chatbots for sensitive or personal matters. Recently, OpenAI’s Sam Altman emphasized this point during a discussion with Theo Von, remarking on the tendency of younger users to treat ChatGPT like a therapist or life coach. Altman cautioned, “When speaking to a professional—whether a therapist, lawyer, or doctor—those conversations are protected by legal privilege…We haven’t established this for ChatGPT yet.”