Google adds a separate switch to stop Gemini training on your chats
Google plans to let Gemini users opt out of model training while keeping access to their earlier chats, according to ZDNET and Ars Technica.
According to ZDNET, Google is preparing to let Gemini users turn off model training without losing access to their earlier conversations with the company's AI assistant. The change is described as coming soon, and the outlet's report did not give a launch date.
Ars Technica reports that the update splits two settings that were previously tied together in Gemini's activity controls. Under the new arrangement, chat history and AI training will each have a separate toggle. A person could keep past chats available while declining to have new conversations used to train models.
Both outlets present the change as a way to make the choice clearer for users. ZDNET's coverage centers on how easy the opt-out will be, while Ars Technica highlights the separation of the two toggles as the key feature. Neither report, as published, explains how the training setting will apply to conversations that already exist, or what data Google may keep for other purposes.
The reports also do not say which regions will get the change first or whether it will apply to every Gemini product. Users who want to adjust their settings should check the activity controls in their account once the option appears.
Background
Large language models learn from enormous collections of text, and companies that run chatbots often want more examples of real conversations to improve their answers. Many consumer AI services have collected chat logs by default and have offered some form of setting that asks the company not to use those logs for training.
The difficulty has often been that a single switch did two jobs. It controlled whether conversations were saved and listed in a history, and it also controlled whether they could feed future model development. People who wanted searchable past chats sometimes had to accept that their words might be used for training. Privacy-minded users who turned training off could lose the history they relied on.
Separating those functions is a common goal in privacy design. It lets a product keep a record for the user's own convenience while limiting secondary uses of that record. Regulators in several places, including the European Union, have pressed technology companies to explain clearly what personal data is used to train AI systems and to give people meaningful control over it. Consent that is buried or bundled tends to draw scrutiny, while clearly labeled and separate choices are generally easier to defend.
A settings change is also not the same as deletion. A company may stop using future conversations for training while still retaining data for safety review, legal compliance or service operation, depending on its policies. Users generally need to read the privacy documentation to learn how long data is kept and when it may be accessed. Careful habits, such as avoiding sharing sensitive personal information with chatbots, remain useful whatever toggles are switched on.
Why it matters
For people who use Gemini regularly, the change addresses a longstanding tension. Chat history is useful for picking up old threads, finding earlier answers and tracking long projects. Being able to keep that record without also contributing to model development gives users a choice that was harder to express before. It also suggests that a major AI provider sees separate consent as a reasonable expectation rather than an optional extra.
The move matters beyond one product. Competing chatbot services have faced similar questions about whether conversations are used for training. A clearer control in a widely used tool may shape what users expect from other services as well. If a company can show that history and training are genuinely separate, it sets a reference point that others may be asked to match.
Still, the value of a toggle depends on the details behind it, including default settings, how clearly the labels are written, and whether someone who turns off training is told what that does and does not cover. A control that is easy to find and understand carries more weight than one that technically exists but is hard to locate.
What to watch
Watch for the exact rollout date and for which accounts, countries and Gemini apps receive the new toggles first. Pay close attention to the default setting for training, since defaults often determine what most people choose. Check whether Google updates its privacy notice to explain retention periods and any exceptions for safety review or legal requests. Also see whether rival AI services adopt similar separate controls, which would indicate whether this becomes an industry expectation. Finally, note whether the wording of the toggles matches what the training setting actually does in practice over the following months.
Viewpoints
Commentary from our columns: three angles on this story.
The Skeptic open questions and weak points
Good news on a settings page is still a settings page, and the most important questions are the ones the reports leave open. The first is timing. The change is described as coming soon, which in product terms can mean weeks or many months. A promise about a future toggle is not the same as a toggle that exists today. The second is what "training" covers. Does it include human review of conversations, or evaluation data drawn from chats? A user who turns off one switch deserves a precise answer. The third is retention. Keeping past chats available is a different promise from deleting them, and nothing reported so far says how long data stays on Google's systems once training is disabled. Finally, there is the default. If training starts out switched on, most people will never change it. Clear separation is progress, but it should be judged on its defaults, its wording and its fine print. Ask for specifics before calling this easier in any meaningful sense.
The User what changes for people who use it
If you use Gemini for school work, planning or writing, the practical change is simple. You may be able to keep your chat history while deciding separately whether your conversations help train future models. That is a more humane arrangement than forcing a choice between convenience and control. For many people, the most useful question is not how to turn something off but what they actually want Gemini to remember. Old chats can be a real timesaver when you return to a project, and that value does not have to disappear because you care about privacy. Treat the two toggles as two separate decisions. One is about your own convenience. The other is about whether your words become raw material for a company's product development. Once the option appears in your account, take a few minutes to read the labels, decide what fits your comfort level and revisit the choice if your habits change. Avoid pasting sensitive details such as medical records, financial numbers or passwords into any chatbot, whatever the settings say. A setting is a helpful tool, not a guarantee.
The Rulebook privacy, safety and regulation
Separating chat history from model training is the kind of design choice privacy advocates have long asked for. When a service bundles those functions, consent can become a formality, because a user who wants a useful record is effectively pushed into allowing secondary uses. Splitting the toggles makes the choice easier to understand, which is what meaningful consent requires. The harder question is what the labels promise in law and in practice. A training switch should mean that future conversations are excluded from development, with clear statements about exceptions for safety monitoring, abuse prevention, legal requests and basic service operation. Users should be able to find retention periods without digging through long documents. Regulators in several regions have made transparency about AI training data a priority, and interfaces that describe these settings in plain language will be easier to defend under that scrutiny. The real test is whether the explanation matches the behavior over time. Watch the privacy notice as closely as the toggle itself.