Opt Out of AI Training: Check Your Chatbot Settings

Additional resource for the lesson “Model Training and Opt-Out Settings” — Online Privacy course

When you chat with an AI assistant, your conversations may be used to help train its future versions. Many services let you opt out of AI training with a single setting, offer chats that are not kept in your history, and, in the UK and EU, the law gives you a right to object. This resource shows you where to look and how to decide calmly.

A. Why this matters

This resource asks one narrow question: does what you type help to shape the next version of the model? That is privacy in the sense this course uses: what is known about you, by whom, and what you choose to let them know.

A conversation used for training does not become public. But it is used for a purpose beyond answering you, and that is a purpose you are entitled to know about and, often, to decline.

Leaving training on is not a mistake. Some people are glad to help improve a tool they use. What matters is that it is your choice, made knowingly.

What the sources say

The UK’s Information Commissioner’s Office (ICO), the independent authority for data protection, explains that “You can ask organisations to stop using your personal information.” This is the right to object. It applies, among other cases, when an organisation relies on what the law calls “legitimate interests”. The ICO adds that the organisation does not always have to agree: it may refuse if it “can demonstrate they have a strong reason to continue using your information”.

The ICO has also spoken about AI training directly. When LinkedIn announced that it would resume using UK users’ information to train generative AI, the ICO said that organisations should put safeguards in place before they start, “including providing a clear and simple route for users to object to the processing.” It added that LinkedIn would need to make sure its approach kept complying with the law.

In the EU, the European Commission lists the right to object among the rights people have under the GDPR. In December 2024 the European Data Protection Board (EDPB), which brings together the EU’s data protection authorities, published an opinion on AI models. It describes a three-step test for deciding whether “legitimate interest” can be used for AI, and gives examples of measures that make it easier for people to exercise their rights.

B. Key concepts

Five ideas that make the settings easier to read.

Model training

Training is how an AI model learns patterns from large amounts of text and other content, so that it can produce new answers. Anthropic, the company behind Claude, explains that “Models do not store text like a database” and do not pull from the original training data once trained.

Why it matters to you: Training is not the same as publishing your chat. It is a further use of what you wrote, and that is why services ask, or let you refuse.

Opt-out setting

A switch in your account that decides whether your conversations may be used to improve the company’s models. Each service names it differently. In the three assistants covered here, it sits in the settings or on an activity page.

Why it matters to you: When you opt out of AI training, the setting usually covers future conversations, so the choice is worth making early.

Chat history and training are separate

On some services, turning off training does not remove your chats from your history. OpenAI, for example, states that conversations “can still appear in your chat history” after you turn its training setting off.

Why it matters to you: Keeping a history is useful for you. Training is useful for the company. Knowing they are separate lets you keep the first and decline the second.

Temporary chats

A conversation that is not saved to your history and is not used for training. ChatGPT calls it Temporary Chat, Gemini uses temporary chats, and Claude calls it an incognito chat. Each service still keeps them for a short period, stated on its help pages.

Why it matters to you: A temporary chat is a good fit for one-off questions you would rather not keep, without changing your general settings.

Feedback can be an exception

The thumbs-up and thumbs-down buttons send the company a rating, and several services treat feedback differently. OpenAI says that if you give feedback after opting out, “the entire conversation associated with that feedback may be used to train OpenAI models.”

Why it matters to you: Even after opting out, one tap on a rating button can share that whole conversation. Rate only chats you are happy to share.

C. A practical example: Elena reads the small switch

Elena is 35 and works in marketing, where she handles client information. At work she uses the assistant her company provides, under its rules. At home she uses two assistants on personal accounts to plan trips, rewrite a letter to her landlord and test ideas for a short story.

A colleague mentions that personal accounts may use chats for training unless you change a setting. Elena is not alarmed, but she has never looked. She spends a quarter of an hour on it:

• In the first assistant she finds the setting under Settings › Data controls, called “Improve the model for everyone”. It is on. She decides to turn it off, and notices that her history stays where it was.
• In the second she finds the “Keep activity” setting. She keeps it on for now, because she likes being able to return to her trip plans, and decides to use temporary chats for anything personal.
• She remembers that she has rated a few answers with a thumbs-up. She decides to rate only chats she is happy to share.

What changed when Elena looked

• Her settings now reflect a choice instead of a default she never saw.
• She made two different choices on two services, for her own reasons. Both are reasonable.
• She drew a line between home and work: the work tool follows her employer’s agreement, and questions about it go to her company.

Her day-to-day use did not change. What changed is that she knows.

D. Try it yourself: check your AI training settings (15 minutes)

Five short steps for each assistant you use with a personal account. You do not have to change anything.

Step 1 — List your assistants (2 minutes)

• Write down every AI assistant you have used in the past month with your own account: on your phone, in your browser, inside other apps.
• Mark which are personal and which are provided by work or school. Leave work and school settings to the organisation, and ask it if you have questions.

Step 2 — Find the training setting (5 minutes)

For each personal account, open Settings and look for words such as “Data”, “Privacy”, “Activity” or “Improve”. The official help pages, at the time of checking, say:

• ChatGPT: open Settings, then Data controls. The setting is called “Improve the model for everyone”. When you are signed in, OpenAI says it “applies to your account across devices.”
• Gemini: the setting is called “Keep activity” and is managed in Gemini Apps activity. When it is on, Google says your activity helps improve its services, “including training generative AI models”.
• Claude: open Settings, then Privacy, and find the model improvement setting. Anthropic says that if you turn it off, it “will not use any new chats and coding sessions you have with Claude for future model training.”

If you use a different assistant, search its help centre for “training” or “improve the model”.

Step 3 — Decide, one service at a time (3 minutes)

• For each setting, choose: keep it on, knowingly, or turn it off.
• Remember what turning off does and does not do. Anthropic, for example, explains that data already used in training runs that have started, or in models already trained, stays there.
• Note your decision in one line, with today’s date.

Step 4 — Try a temporary chat (3 minutes)

• Open one temporary or incognito chat and ask something simple.
• Check the help page for how long it is kept. At the time of checking: ChatGPT says temporary chats “may be retained for up to 30 days for safety purposes”; Gemini says it keeps them for 72 hours; Claude says incognito chats are kept for 30 days by default.
• Decide which kinds of question you will ask this way from now on.

Step 5 — Know your other route (2 minutes)

• If a service has no setting, or you are unsure what it covers, remember that you can object directly.
• You can make your request “verbally or in writing at any time”, the ICO explains, and “There are no specific words that you must use.”
• Organisations must respond within one calendar month, which they can extend for complex requests if they tell you.

A decision you wrote down is easier to revisit than one you only remember.

E. Videos, articles and further resources

Institutional sources first, then official platform help pages. All in English.

Institutional sources

Information Commissioner’s Office (UK) — The right to object to the use of your information
When you can ask an organisation to stop using your information, how to ask, and how long it has to reply.
https://ico.org.uk/for-the-public/the-right-to-object-to-the-use-of-your-information/

Information Commissioner’s Office (UK) — Our statement on changes to LinkedIn AI data policy
The regulator’s view that people should have a clear and simple route to object before their data is used to train generative AI.
https://ico.org.uk/about-the-ico/media-centre/news-and-blogs/2024/09/our-statement-on-changes-to-linkedin-ai-data-policy/

European Data Protection Board — EDPB opinion on AI models: GDPR principles support responsible AI
How EU data protection authorities approach personal data in AI models, including legitimate interest.
https://www.edpb.europa.eu/news/edpb-opinion-on-ai-models-gdpr-principles-support-responsible-ai_en

European Commission — Data protection explained
Your rights under the GDPR in brief, including the right to object.
https://commission.europa.eu/law/law-topic/data-protection/data-protection-explained_en

Platform documentation

OpenAI Help Center — Data controls in ChatGPT
Where the training setting is, what it covers, how Temporary Chat works and how feedback is treated.
https://help.openai.com/en/articles/7730893-data-controls-in-chatgpt

Gemini Apps Help — Gemini Apps Privacy Hub
What the “Keep activity” setting does, how long chats are kept, human review and temporary chats.
https://support.google.com/gemini/answer/13594961?hl=en-GB

Anthropic Privacy Center — How do I change my model improvement privacy settings?
Where the setting is in Claude and what turning it off does, and does not, change.
https://privacy.claude.com/en/articles/12109829-how-do-i-change-my-model-improvement-privacy-settings

Anthropic Privacy Center — Is my data used for model training?
When consumer chats may be used to improve Claude, how feedback is handled and why incognito chats are excluded.
https://privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training

Links checked in October 2026. If an address changes, searching for the title on the organisation’s site usually finds it.

F. The Cyber Welfare Framework: Skills, Awareness, Secure Behaviour

This resource sits on the Skills pillar at level FL2 – Beginner, where the work is finding a setting, reading what it really does and using it on purpose.

Skills

• Finding the training setting in each assistant you use.
• Telling apart chat history, training and feedback.
• Starting a temporary or incognito chat when it suits the question.

Awareness

• Understanding that using an assistant and letting it learn from you are separate decisions.
• Knowing that a setting usually changes what happens from now on, not what has already been used.
• Knowing that, in the UK and EU, you can object even where no switch exists.

Secure Behaviour

• Checking training settings when you start using a new assistant.
• Giving feedback only on chats you are happy to share.
• Leaving work and school accounts to the organisation’s rules, and asking it when in doubt.

G. Questions to sit with

• Which of your conversations with an assistant would you be glad to help improve it, and which would you rather keep to yourself?
• Do you prefer one general setting, or deciding chat by chat with temporary chats?
• If a service had no setting at all, would you use your right to object, or choose a different service?

H. What to do now

The recommendations (R) and security measures (MS) from the Cyber Welfare database that apply most directly here. No R or MS code covers AI training settings directly, so the steps below are habits from this course rather than database recommendations.

1. Settings you control

• Check the training setting in every personal assistant account, using the names in step 2 as a guide: “Improve the model for everyone”, “Keep activity”, model improvement in Claude.
• Recheck it after major updates, since settings can be renamed or moved.

Minimum commitment: today, open the settings of the assistant you use most and decide, knowingly, whether training stays on or off.

2. Rights you can use

• If no setting exists, write to the service and say clearly that you object to your personal information being used to train its models.
• Keep a copy of your request and the date, since organisations have one calendar month to respond.
• If you object by phone or in a chat with support, the ICO suggests following it up in writing.

Minimum commitment: note the help page or contact address of your main assistant, so that you know where to send an objection if you ever need one.

In short

• Using an assistant and letting it learn from your chats are two separate decisions.
• To opt out of AI training, look under Settings for “Data”, “Privacy” or “Activity”. The setting usually covers future chats.
• Temporary and incognito chats are not used for training, but they are kept for a short period.
• In the UK and EU you have a right to object, in writing or verbally, with no special words.

Related resources in this course

• AI Chatbot Privacy: Where Your Conversations End Up
• What Not to Share with AI Chatbots: Keep, Generalise, Share
• Family AI Rules: Agreeing How AI Is Used at Home and at School
• AI Search Privacy: What Changes When You Ask Instead of Search

Discover more companion resources from the online courses of the Protect Your Digital Privacy programme.