Many people assume a chat with an AI assistant stays between them and the screen, but several services may use conversations to improve their models unless you say no. A recent privacy ranking found that an AI chatbot training opt-out exists on most major platforms, often as a single toggle, yet it only protects future chats.
In this article, we will explore which platforms make opting out easy, which do not, what the switch cannot undo, and what to keep out of chats.
The Study Behind the Ranking
The facts here come from the Gen AI and LLM Data Privacy Ranking 2026, published in mid-August by Incogni, a company that sells privacy services. That matters: it is a vendor study, not an independent audit or a regulator’s finding.
The researchers looked at 13 platforms: ChatGPT, Claude, Gemini, Grok, Vibe (formerly Le Chat), Perplexity, Qwen, DeepSeek, Z.ai, Kimi, Meta AI, Pi and Copilot. They scored 11 weighted criteria across three areas: what happens to user data, transparency, and data collection and sharing. A lower overall score means a less invasive platform. The data was collected between June 15 and July 6, and mobile apps were assessed separately, covering 12 apps.
The page did not give exact per-platform scores, so only rough figures can be repeated. Vibe came out lowest at about 12 points, with ChatGPT also around 12. Most platforms sat near 14, while Copilot, Meta AI and Kimi scored about 16 or more. These are approximate, and the gaps between neighbors are small enough that you should not pick a service on rank alone.
AI Chatbot Training Opt-Out by Platform Group
The part of the study that is most useful in daily life is how easy it is to refuse training. The sheet sorts platforms into four groups, which the table below reproduces in plain terms.
| Group | Platforms | What the study found |
| Single toggle, no trade-off | ChatGPT, Claude, DeepSeek, Vibe, Grok, Copilot, Perplexity, Pi | One setting turns off training with no loss of features |
| Toggle with a trade-off | Gemini | Opting out also stops chats appearing in your chat history |
| Right exists, process unclear | Kimi (Moonshot AI), Qwen, Meta AI | Kimi asks you to email support and verify identity, Qwen uses a request, and Meta never explains how to object |
| No functioning opt-out found | Z.ai | The researchers found no working way to opt out |
Reading the Table Sensibly
Eight of the 13 platforms fall in the easiest group, so for most users the barrier is knowing the switch exists, not finding a way around a policy. Gemini’s trade-off is worth weighing: if you rely on your history, you must choose between privacy of future chats and convenience. In the unclear group, assume your chats may be used and type accordingly.
Limits of Opting Out

The most important finding is easy to miss. According to the study, no platform lets users remove data that has already been used in training. Opting out affects future use only. A chat you typed last year, before you changed the setting, is not pulled back.
Opting out also does not mean the provider stops holding your chats. Storage for your history, safety review and legal obligations are separate questions from training, and a training toggle usually says nothing about them. That is general guidance, so read the provider’s privacy page for what happens to stored conversations.
Finally, the study says no provider disclosed exactly which datasets it trained on. So even a perfect opt-out cannot tell you what is already in a model.
Steps for Your AI Chatbot Training Opt-Out
Menu names change often, so what follows is general guidance, not a sourced walkthrough. Check each provider’s own help page for the current wording. First, open the app or website and find the account menu, then look for a section named privacy, data controls or something close to it.
Second, look for wording about improving the model with your content, and switch it off. Third, if your platform offers a temporary or incognito chat mode, use it for sensitive topics, since it is designed to keep those chats out of your regular history. Fourth, delete old chats that contain anything personal. Fifth, repeat the process on each device and each app, because the study assessed mobile apps separately and settings may not be identical.
If you use a platform in the unclear group, follow the provider’s stated route, which might be a request form or an email to support. Also note that the study used California-style policies where available, so rights in the EU or elsewhere may differ.
Things to Keep Out of Chats
Opt-outs reduce risk but do not remove it, so the safest habit is deciding what never goes in. This is general guidance, not legal advice. Do not type passwords, recovery codes or bank details. Avoid pasting medical records, legal documents or tax information that identifies you. Be careful with other people’s private details, such as a colleague’s performance review or a friend’s health news. Work files that your employer treats as confidential should stay out unless your organization has approved the tool.
Conclusion
An AI chatbot training opt-out is easy on most big platforms, but it works like a door closing behind you, not a rewind button. The study comes from a privacy-service vendor, gave no exact scores, and may be out of date, so confirm every setting yourself. Make the switch, use temporary chats for sensitive topics, and keep secrets out of any chatbot, whatever the policy says.
Frequently Asked Questions
How Do I Turn off Training on My Chats?
Open the privacy or data controls in the app or website and look for the option about using your content to improve the model. Menu names change, so check the provider’s help page. For platforms with unclear processes you may need to send a request.
Does Opting Out Delete My Old Chats From Training Data?
No, According to the study, no platform lets users remove data already used in training. Opting out affects future use only.
Which Chatbots Have a Single Toggle With No Trade-Off?
The study lists ChatGPT, Claude, DeepSeek, Vibe, Grok, Copilot, Perplexity and Pi. Policies may have changed since the data was collected in June and July, so verify in the app.
Can I Opt Out of Training on Gemini Without Losing History?
According to the study, opting out on Gemini also stops chats appearing in your chat history. That makes it a toggle with a trade-off. Check the current settings, as this may have changed.
Is the Incogni Ranking Reliable?
It is a useful starting point, but it is a vendor study with approximate scores, and the researchers noted that vague policies made matching uncertain. Policies may have changed after the collection window, and the Copilot Android score was copied from iOS.
