AI Chatbot Privacy: Where Your Conversations End Up

Additional resource for the lesson “AI Assistants: Where Your Data Ends Up” — Online Privacy course

When you type into an AI assistant, your words do not vanish once the answer appears. AI chatbot privacy is about what happens next: where the conversation is stored, for how long, who might read it, whether it helps train the model, and which other services it reaches. This resource follows one conversation along that path, so that you can decide what you are comfortable with.

A. Why this matters

An AI assistant is useful because you give it context: a draft email, a question about your week, a document to summarise. That context is often personal data, and it travels further than your screen.

Privacy, in this course, means three questions: what is known about you, by whom, and what you choose to let them know. With an AI assistant, the answers sit in two places that are easy to skip: the settings page and the privacy notice. Neither needs reading line by line. None of this means you should stop using these tools, only that you can use them knowingly.

What the sources say

The UK’s National Cyber Security Centre (NCSC) set out the basic picture in 2023. A question you type into a public chatbot “will be visible to the organisation providing” it, and such queries “are stored and will almost certainly be used for developing the LLM service or model at some point”. (LLM means large language model, the technology behind these assistants.) Its advice: leave sensitive information out, and read the terms of use and privacy policy first.

In the EU, data protection authorities ran a joint taskforce on ChatGPT within the European Data Protection Board (EDPB). Its 2024 report says people should be “clearly and demonstrably informed” that what they type may be used for training. It adds that “the responsibility for ensuring compliance with GDPR should not be transferred to data subjects”, meaning the users themselves.

The UK’s Information Commissioner’s Office (ICO), the independent data protection authority, told generative AI developers in December 2024 that it was time to tell people how they use their information, warning that “without better transparency, it will be hard for people to exercise their information rights”.

B. Key concepts

Six ideas about where a conversation goes after you press Enter.

Conversation history and retention

Retention is how long a service keeps your data. Assistants such as ChatGPT and Gemini save your chats to your account so that you can return to them.

OpenAI’s help page says regular and archived ChatGPT chats “remain saved to your account until you delete them”. Google’s Gemini Apps Privacy Hub says activity is auto-deleted after 18 months by default, and that you can choose 3 months, 36 months or no auto-deletion.

Why it matters to you: Deleting a chat is not instant erasure. OpenAI’s Europe privacy policy says deleted personal data is removed from its systems within 30 days, with some legal and security exceptions.

Temporary chats

ChatGPT and Gemini both offer a mode that does not save the conversation to your history.

Google says that temporary chats are not used to train its AI models, and OpenAI says ChatGPT’s are not used to improve its models. Both still keep them for a short period, stated on their help pages; the resource in this course on opting out of AI training gives the current figures.

Why it matters to you: “Temporary” means “not in your history”, not “never stored”. It suits one-off questions, as long as you know that a short retention period still applies.

Human review

Some services have people, not only software, read a sample of conversations to check the quality and safety of answers.

Google is explicit: human reviewers, including trained reviewers from its service providers, read some Gemini chats, and chats are disconnected from your account before they are sent to those service providers. Reviewed chats are kept for up to three years and are not deleted when you delete your activity.

Why it matters to you: It turns a vague worry into a clear question: would I mind a stranger reading this, without my name attached?

Use for training

Training means using examples, which can include your conversations, to improve the model.

In ChatGPT, the setting is called “Improve the model for everyone”. OpenAI says that when it is off, “your new conversations won’t be used to train OpenAI models”. In Gemini, Google says that with “Keep activity” on, your activity is used to develop and improve its services, including training generative AI models.

Why it matters to you: Note the word “new”: the change affects what comes next.

Connected apps (connectors and plug-ins)

A connector, plug-in or connected app links the assistant to another service, such as your email or cloud storage.

OpenAI says that data shared with apps “is handled according to each app’s terms of service and privacy policies”. Google says that “apps retain data that Gemini shares with them subject to their own policies”. For individual ChatGPT plans, OpenAI also says information accessed from apps may be used for training when “Improve the model for everyone” is on.

Why it matters to you: Each connection adds a new place where your data ends up, with its own rules, and it stays in your settings until you disconnect it.

Where the data is processed

Data location is the country where your data is stored and processed, which affects the laws and safeguards that apply.

OpenAI’s Europe privacy policy, for example, says it “processes your Personal Data on servers located outside of the EEA, Switzerland and the UK”, including in the United States, and names the legal safeguards it relies on for those transfers.

Why it matters to you: You do not need to be a lawyer here. The ICO says an organisation must tell you if it transfers your data overseas, including the country involved, so look in the privacy notice.

C. A practical example: Gianni’s assistant

Gianni is 50 and runs a small business. He uses an AI assistant on his personal account to draft replies to customers and tidy up quotes. It saves him time, and he has no intention of giving that up.

Over a few months, his use grew in small steps:

• He pasted a supplier’s price list to ask for a summary.
• He asked for help with a difficult reply, including the customer’s first name and the problem with their order.
• He connected his email account so the assistant could find past messages, for one task he has not repeated since.
• He signed in with an account he already had elsewhere, because it was quicker than a new password.

Each step made sense at the time. But when a supplier asked him, half joking, “Where does all that go?”, Gianni realised he did not know.

What changed when Gianni looked

Gianni spent fifteen minutes on the exercise below. What he found was calmer than he feared:

• His chat history went back to his first question, customer message included.
• The training setting was on. He had not realised it was there.
• The email connection was still active, months after that one task.
• The privacy notice said, in one paragraph, which country his data was processed in.

He made three decisions. He disconnected his email account. He deleted the old chats that held customers’ details. He kept the training setting on for general questions, and chose temporary chats whenever a customer was involved. He also noted a question for later: whether a business plan, which has different default rules, suits his work. Not less help from the assistant, but help on terms he chose.

D. Try it yourself: follow one conversation (15 minutes)

Five short steps with the AI assistant you use most. Setting names differ between services and change over time: if you cannot find one, search the service’s help pages for the words in quotation marks.

Step 1 — Find your history and its retention period (3 minutes)

• Open the assistant’s settings, often under your profile icon, then a section called something like “Data controls” or “Activity”.
• Check how far back your chat history goes and whether the service offers an auto-delete period.
• Note where to start a temporary chat.

Step 2 — Check the training setting (3 minutes)

• Look for the setting that decides whether your conversations are used to train or improve models. In ChatGPT it is “Improve the model for everyone”; in Gemini, it is linked to “Keep activity”.
• Write down whether it is on or off. Do not change anything yet.

Step 3 — Look for human review (2 minutes)

• Search the service’s privacy hub or help pages for “human review” or “reviewers”.
• Note who reviews, whether chats are separated from your account first, and how long reviewed chats are kept.
• If you cannot find an answer, write that down too.

Step 4 — List your connected apps (3 minutes)

• Find the list of connected apps, connectors or plug-ins in the settings.
• Next to each, note when you last needed it.
• Note whether you signed in to the assistant with another account.

Step 5 — Read the privacy notice with four questions (4 minutes)

In the privacy notice, look only for these four answers, which the ICO lists among the things an organisation must tell you:

• How long is my data kept?
• Who is it shared with, by name or by category?
• Is it transferred to another country, and which one?
• How do I use my rights, and how do I complain?

Then make one decision: keep a setting as it is, now that you know it; change one setting; or disconnect one app. Any of the three is a good result.

E. Videos, articles and further resources

Institutional sources first, then the help pages of two widely used assistants, as examples rather than recommendations.

Institutional sources

National Cyber Security Centre (UK) — ChatGPT and large language models: what’s the risk?
What happens to the questions you type into a public chatbot, with simple advice (2023).
https://www.ncsc.gov.uk/blog-post/chatgpt-and-large-language-models-whats-the-risk

Information Commissioner’s Office (UK) — Your right to be informed if your personal data is being used
What an organisation must tell you, including retention and overseas transfers.
https://ico.org.uk/for-the-public/your-right-to-be-informed-if-your-personal-data-is-being-used/

Information Commissioner’s Office (UK) — Generative AI developers, it’s time to tell people how you’re using their information
The UK regulator’s call for transparency about personal data in generative AI.
https://ico.org.uk/about-the-ico/media-centre/news-and-blogs/2024/12/generative-ai-developers-it-s-time-to-tell-people-how-you-re-using-their-information/

European Data Protection Board — Report of the work undertaken by the ChatGPT Taskforce
The 2024 findings of EU data protection authorities on transparency, training and users’ rights.
https://www.edpb.europa.eu/system/files/2024-05/edpb_20240523_report_chatgpt_taskforce_en.pdf

Platform documentation

Gemini Apps Help — Gemini Apps Privacy Hub
Retention periods, human review, temporary chats, connected apps and the “Keep activity” setting.
https://support.google.com/gemini/answer/13594961?hl=en-GB

OpenAI Help Center — Data controls in ChatGPT
The training setting, temporary chats and where to find them.
https://help.openai.com/en/articles/7730893-data-controls-in-chatgpt

OpenAI Help Center — Connected apps in ChatGPT
What connected apps can access, whose rules apply, and how to disconnect them.
https://help.openai.com/en/articles/11487775-connected-apps-in-chatgpt

OpenAI — Europe privacy policy
Deletion, where data is processed and transfer safeguards.
https://openai.com/policies/eu-privacy-policy/

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 Awareness pillar at level FL2 – Beginner: AI chatbot privacy begins with understanding how a familiar tool handles your data.

Skills

• Finding the history, retention and training settings of the assistant you use.
• Locating and removing connected apps you no longer need.
• Reading a privacy notice for four specific answers.

Awareness

• Recognising that a conversation may be stored, reviewed and used for training.
• Understanding that “temporary” and “deleted” still involve short retention periods.
• Seeing each connected app as another place where your data ends up.

Secure Behaviour

• Using a temporary chat for one-off or more personal questions.
• Removing connections once the task they were set up for is done.
• Checking the settings again when the privacy policy changes.

G. Questions to sit with

• If one of your recent conversations were read by a reviewer, without your name attached, which one would you mind most, and why?
• Which of your connected apps did you set up for a single task, and is that task still going on?
• Is a long chat history worth what it says about you?

H. What to do now

The recommendations (R) and security measures (MS) from the Cyber Welfare database that apply most directly here.

1. What is kept, and for how long

No recommendation in the database covers AI assistant settings directly yet, so the advice sits here: choose how long chats are kept, decide knowingly about training, and keep temporary chats in mind.

Minimum commitment: this week, open your assistant’s data settings and write down two answers: how long your chats are kept and whether they are used for training.

2. What is linked

• R31 — Limit social login: signing in to an assistant “with” another account links the two. Connected apps link in the other direction, so review the list in the assistant’s settings and disconnect what you no longer use.

Minimum commitment: remove one connected app or plug-in you set up for a task that is now finished.

In short

• AI chatbot privacy starts with one fact: with the services covered here, what you type is stored by the provider, at least for a while.
• Retention, human review and training are separate questions, each answered in the help pages.
• Every connected app is another place where your data ends up, under that app’s own rules.
• Four questions are enough to read a privacy notice: how long, shared with whom, which country, and your rights.

Related resources in this course

• What Not to Share with AI Chatbots: Keep, Generalise, Share
• Opt Out of AI Training: Check Your Chatbot Settings
• 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.