When you start a conversation with ChatGPT, you probably don’t expect that chat to end up discoverable in a Google search—but that’s exactly what happened for some users. As TechCrunch highlights, if you filtered your search engine results by “site:https://chatgpt.com/share,” you could find the transcripts for real conversations people were having with OpenAI’s bot—chats you’d think remain private to the account they’re associated with were as easy to find as a recipe or tech hack.
As one might assume from conversations that weren’t meant to be shared publicly, some of these chats contained embarrassing or questionable discourse. TechCrunch said it found a user asking ChatGPT for help rewriting a resume for a specific job application, a job which TechCrunch was able to deduce based on the conversation. Another user asked ChatGPT questions that, according to TechCrunch, “sound like they came out of an incel forum,” though the outlet didn’t elaborate on the contents of the chat.
OpenAI’s experimental feature
Before you panic, there are a couple of caveats to this particular situation. First, OpenAI has since removed the ability to make chats public to search engines, and, from what I can tell, any new searches return zero results for ChatGPT conversations. Any chats you start now do not come with the risk of exposure—at least, not in this capacity. To that point, the exposed chats in question were only discoverable on Google because the users had explicitly opted into that feature. You would have needed to click the “share” button on a chat, choose a “create link” option, skim past an alert letting you know your name, chat instructions, and messages you add after the fact remain private, then hit a toggle to make that chat discoverable in search.
Some chats were archived by the Wayback Machine
Following the initial story, sleuths like those from Digital Digging discovered that some of these shared conversations even ended up archived on the Wayback Machine. By their count, 110,000 ChatGPT threads were accessible through this tool. Despite OpenAI’s efforts to erase the chats from search, many were still available to skim for anyone in the know and interested. Digital Digging’s article dropped on Aug. 1, and, at the time, OpenAI had not issued a take down request to the Wayback Machine concerning the chatgpt.com/share URLs. However, as of Aug. 4, the Wayback Machine says this domain is excluded, so it appears OpenAI did eventually make the request.
Why even make this a feature in the first place? OpenAI had this to say to TechCrunch: “We’ve been testing ways to make it easier to share helpful conversations, while keeping users in control, and we recently ended an experiment to have chats appear in search engine results if you explicitly opted in when sharing.”
That’s not overly clear, though it’s not hard to assume the benefit to OpenAI. The more exposure ChatGPT has, the better it is for the company. And as the internet increasingly moves toward both AI (think AI Overviews) and forum-based answers (think Reddit), I could see OpenAI thinking they have an opportunity to capitalize on the market here. If a user asks ChatGPT a question they think was answered well, perhaps they share it with search engines, so that other users benefit as well. Now, when someone googles that same question, maybe that ChatGPT conversation floats to the top of the search results, right next to the AI Overview or relevant Reddit threads.
OpenAI also isn’t the only company to experiment with public AI conversations. Back in June, we learned that Meta AI also had a function that would allow users to post their questions and generations—not to search, mind you, but to the public Meta AI feed. It seems AI companies are increasingly interested in publicizing AI-generated content, whether that’s a conversation you had with a chatbot, or an AI artwork that bot produced.
Chatbots are not private
You can now rest easy knowing your ChatGPT conversations won’t end up on the front page of Google. However, don’t assume that your chats with any bot are generally private. In fact, there’s a good chance the company that owns your bot is using your conversations to train their models, or that human reviews will even be able to see your chats.
Depending on the bot, there are some privacy settings you can enable to protect yourself. ChatGPT’s “Improve the model for everyone” setting controls whether or not ChatGPT can take your conversations to train their model—though disabling it won’t stop the company from storing your chats. Even temporary chats, which don’t appear in your history, remain on ChatGPT servers for up to 30 days.
As such, you really shouldn’t use chatbots for anything sensitive or personal. OpenAI’s Sam Altman offered a good reminder of this last month: During an interview with Theo Von, Altman discussed how so many of their users, especially young people, use ChatGPT as a therapist or life coach. Altman said: “Right now, if you talk to a therapist or a lawyer or a doctor about those problems, there’s legal privilege for it … We haven’t figured that out yet for when you talk to ChatGPT.”
