ChatGPT has become synonymous with AI chatbots and is widely regarded as the most popular one. However, users should be aware that OpenAI uses private chats to improve its model, known as training. Although OpenAI warns users not to share sensitive information as chats can be reviewed for training purposes, this warning is only given to new users when they join ChatGPT for the first time.
Additionally, OpenAI employs human reviewers from “trusted service providers” to process private, de-identified chats, although this is not explicitly mentioned on ChatGPT’s homepage. In comparison, Google clearly informs users on the homepage that human reviewers are used to train and improve its Gemini chat model.
Although OpenAI provides a “Data controls” option in ChatGPT’s settings to halt model training on all chats, activating this setting also disables the Chat History feature. This linkage feels punitive and may dissuade users from prioritizing privacy.
Access to past conversations is a fundamental feature and should not be contingent upon privacy settings. Both aspects should be able to coexist, as they do elsewhere.
OpenAI provides a privacy portal page where users can opt out of model training while keeping their Chat History intact. Surprisingly, this option is not promoted anywhere in ChatGPT’s interface. Users must navigate to the Data Controls FAQ page, where it is mentioned that they can disable model training and retain their chats.
How can I keep my chat history but disable model training? We do not utilize content from our business offerings, such as ChatGPT Team, ChatGPT Enterprise, and our API Platform, to train our models. For more details on how we handle business data, please refer to our Enterprise Privacy page. For users on a ChatGPT Plus or ChatGPT Free plan within a personal workspace, there are two ways to opt out of training. You can use the privacy portal and click on "do not train on my content," or you can turn off training for your ChatGPT conversations by toggling off the Chat history & training option in Settings. Once you opt out, any new conversations will not be used to train our models.
Continuing my investigation, I accessed OpenAI’s privacy portal page and submitted a request to cease training on my content. If you’re already signed in to your OpenAI account, you’ll be automatically logged in. Otherwise, you’ll need to provide the email address linked to your ChatGPT account, and OpenAI will send you an email. Click on the link in the email to submit your request. This request can be made by both free and ChatGPT Plus users. Here’s an overview of the process:
The request applies to all future chats and not your past conversations. An active request will be generated. Reload the page after a minute or two, and your request will be processed, this was my experience. Additionally, you will receive an email confirming the following:
“We successfully processed your request to not train on content provided to our consumer services. We will no longer use your content to train our models. As a reminder, this request is forward looking and does not apply to content that had previously been disassociated from your account.”
Now, you can continue using ChatGPT with your Chat History turned on (including Chat sharing), while no longer being part of model training. Thank you for considering this option. Adding a link to the privacy portal page inside the Data Controls settings page would be a valuable addition, allowing users to make more informed decisions about their privacy.
This type of dark pattern raises concerns about OpenAI’s commitment to transparency, especially regarding user chat history. It would benefit the company to align their actions with their transparency goals to maintain user trust.
