Google separates Gemini AI training from chat history settings

Ars Technica reports that Google is splitting one Gemini privacy setting into two. Users will be able to stop their conversations being used for AI.

Ars Technica reports that Google is splitting one Gemini privacy setting into two. Users will be able to stop their conversations being used for AI training while still keeping their chat history.

Key takeaways

  • Ars Technica reports that Gemini’s activity settings will have separate toggles for AI training and for chat history.
  • Until now, Gemini users who wanted to keep their past conversations generally had to accept that those conversations could also be used to improve Google’s AI.
  • The change gives users a choice they did not clearly have before: keep their history without contributing it to model training.
  • The source does not say exactly when the new toggles will reach every user, region or account type.
  • Separating the two settings follows a wider shift in the AI industry towards giving users more specific privacy controls.

What is changing in Gemini’s privacy settings?

Gemini is Google’s AI assistant. It is available as a standalone app, on the web and inside other Google products. Like most chatbots, it can keep a record of past conversations, so users can scroll back, pick up an earlier thread or find an answer they were given before.

Ars Technica reports that the Gemini activity settings will now have two separate controls. One decides whether conversations are kept as chat history. The other decides whether they can be used to train Google’s AI models. Once these are separate, a person can keep their history and opt out of training, or make some other combination of choices.

The source does not give a full technical description of what “training” covers here. It is also unclear whether the change affects other kinds of data handling, such as human review of conversations. Readers should check the wording in their own settings page once the update reaches them.

Why is this in the news now?

The story is current because Ars Technica has reported the change. It matters because it addresses a long-standing complaint about how Gemini’s settings were designed.

Before this change, one activity setting broadly controlled both whether conversations were saved and whether they could feed into product improvement. That left privacy-conscious users with an awkward trade-off. If they wanted to keep their data out of training, they generally also lost the convenience of a saved chat history. Critics of this kind of design argue that it pushes users to share more data than they would choose to if each decision were separate.

The source does not say what prompted Google to make the change now. Any account of the company’s internal reasons would be speculation.

What background does a newcomer need?

Large language models such as the ones behind Gemini are built by training on very large amounts of text. After a model is first released, companies often keep improving it. One source of material for that work is the conversations people have with the product. Those conversations show where the model gives wrong, unhelpful or unsafe answers, which makes them useful to developers.

They can also be sensitive. People use chatbots to draft work documents, ask health questions, work through personal problems and paste in code or business information. When that material is used for training, it may be processed, stored or reviewed in ways the user did not expect when they typed it.

Chat history is a separate matter. It is a convenience for the user, a log they can return to. Nothing about storing a conversation for the user to reread requires it to be used for training. Combining the two decisions in one setting was a design choice, not a technical requirement, and that is why separating them has drawn attention.

Other AI providers have dealt with the same tension in different ways. Some offer temporary or incognito-style chats that are not saved. Some have separate data controls for training. Others draw a line between consumer and business accounts. Each provider’s approach is different and changes over time, so the details are best checked against each company’s current policy.

Who is affected, and how?

The people most directly affected are individual Gemini users with personal Google accounts. Those who had turned off activity tracking to keep their data out of training, and accepted losing their history as the price, may now be able to get their history back without changing their position on training.

Users who never changed their settings are affected in a quieter way. The update is a chance to look at what they have agreed to, because a default setting is not the same as an informed choice. The source does not say whether existing preferences carry over automatically or what the defaults will be for each toggle. Those details are not known from the reporting.

Organisations that use Gemini through workplace or education accounts may be covered by separate contractual terms on data use. The source does not say whether or how this change applies to them. Administrators should check their own agreements rather than assume the consumer settings apply.

Parents and guardians who manage accounts for younger users may also want to check how the new toggles appear on supervised accounts. The reporting does not cover this.

Where do informed people disagree?

Most observers welcome more specific controls, but views differ on how far they go. One view is that splitting the toggles is a real improvement. It removes a trade-off that discouraged people from protecting their data and gives them a clear, understandable choice.

A more sceptical view is that opt-out controls of any kind put the burden on the user. Under this argument, training on personal conversations should require active consent (opt-in), not continue until someone finds the right setting and turns it off. Whether the new training toggle is on or off by default therefore matters a great deal, and the source does not settle that question.

A third debate is about what opting out actually guarantees. Privacy advocates often point out that a training opt-out may not stop all processing. Data can still be kept for a period for safety, abuse prevention or legal reasons. How Gemini handles conversations after a user opts out of training is not described in the source material.

There is also a view from the industry side. Developers argue that real-world conversations help make models safer and more accurate. If many users opt out, they say, it could become harder to improve the product for everyone. This is a genuine tension, not one with an obvious answer.

What are the practical implications for users?

For most people, the practical step is simple. Once the update is available, open the Gemini activity settings and look at the two controls separately. Decide whether you want a saved history, and then decide whether you are comfortable with your conversations being used for training. Under the new arrangement, these are two independent questions.

Users should also remember that changing a setting may only apply from that point forward. Whether conversations already used for training can be removed from that process is not addressed in the reporting.

Some sensible habits apply whatever the settings say. Avoid pasting passwords, financial details or confidential business documents into any AI chatbot. Treat the history feature as a convenience, not as secure storage. Recheck privacy settings after major app updates, because menus and defaults can change.

What should readers watch for next?

Several things are still unclear. One is the rollout: when the separate toggles reach all users, and whether they appear in every country and on every platform. Another is the default for new users. A training toggle that starts switched off would mean something quite different from one that starts switched on.

It will also be worth watching whether Google publishes clearer documentation on what happens to data after a user opts out of training. Regulators, especially in jurisdictions with strong data protection laws, may look closely at how consent and defaults are presented. Competing AI providers may also come under pressure to offer equally specific controls if they do not already.

Frequently asked questions

How do I stop Gemini using my chats for AI training?

According to Ars Technica, Gemini’s activity settings will include a separate toggle for AI training. Once the update reaches your account, you should be able to switch off training without switching off chat history. The exact menu labels and location may vary by device and region, so check the activity or privacy section of the Gemini app or website and read the description next to each toggle.

Can I keep my Gemini chat history if I opt out of training?

That is the main point of the reported change. Ars Technica says chat history and AI training will have separate controls in Gemini’s activity settings. This should let users keep a record of past conversations while declining to have them used to train Google’s models. The reporting does not confirm when the change reaches every user, so availability may differ for a while.

Why were Gemini history and training linked before?

Previously, one Gemini activity setting broadly controlled both whether conversations were saved and whether they could be used to improve Google’s AI. Google has not publicly explained in the available reporting why it originally combined the two. There is no technical need to link them, because saving a chat for the user to reread is a different process from using it to train a model.

Does opting out of training delete my existing Gemini data?

The available reporting does not say. Changing a privacy setting often applies only to future activity, and conversations that were already processed may be handled under the earlier terms. Users who want past conversations removed should look for a separate option to delete activity, and should read Google’s privacy documentation on how long data is kept and for what purposes.

Does this change apply to Gemini work and school accounts?

The source does not say. Workplace and education versions of Google’s products are often covered by separate agreements on how customer data may be used, which can differ from the terms for personal accounts. Administrators and users of managed accounts should check their organisation’s agreement and admin console instead of assuming the consumer settings described in the reporting apply to them.

Is it safe to share personal information with AI chatbots?

Even with training switched off, it is sensible to be careful. Conversations may still be stored, processed or kept for a period for safety or legal reasons, depending on the provider’s policy. Avoid entering passwords, banking details, medical records or confidential work material into any chatbot. Treat privacy toggles as one layer of protection, not a guarantee that data is never kept.

Sources and further reading

  • Ars Technica: technology news reporting on the separation of AI training and chat history toggles in Gemini’s activity settings.
  • Google privacy and help documentation: official descriptions of Gemini activity settings and data handling.
  • Data protection authorities: regulatory guidance on consent, defaults and the use of personal data in AI systems.
  • Digital rights organisations: analysis of opt-in and opt-out models for AI training on user content.

Surfaced from the rss:arstechnica signal “AI privacy settings change”. AI-assisted draft, editorially reviewed.

Visited 1 times, 1 visit(s) today
share this recipe:
Facebook
X
WhatsApp
Telegram
Email
Reddit