How to Use AI Chatbots Anonymously, and Where It Stops

Additional resource for the lesson “Chatbots and AI Assistants: What (Not) to Share” — Online Anonymity course

You can use AI chatbots anonymously only up to a point, and knowing where that point is matters more than any single setting. A conversation becomes linked to you through the account you sign in with, the history it keeps, the things the assistant remembers about you, and the details you type yourself. This resource helps you see those links in one account and decide which conversations need fewer of them.

A. Why this matters

In this course, privacy is deciding what others know about you; anonymity is being able to act without what you do being linked back to who you are. With an AI chatbot, the two questions are easy to mix up. Another resource in the Online Privacy course asks what you should type. This one asks a narrower question: once you have typed it, what connects that conversation to you?

That link is often useful. Being signed in lets you find last week’s draft, and a remembered preference saves you explaining yourself again. The point is not to cut every link, but to notice them, so that a conversation you would rather keep apart from your name does not end up sitting next to everything else.

Anonymity here has limits. Even without an account, the service still receives your connection and whatever you write. What you can change is how many of your conversations are joined together, and whether they are joined to a named profile.

What the sources say

The European Commission, explaining EU data protection rules, reminds us that personal data includes “different pieces of information, which together can lead to the identification of a particular person”. Its examples include an IP address, the number that identifies your connection, alongside a name and an email address.

A 2025 report on large language models, prepared for the European Data Protection Board (EDPB) by an external expert, describes long-term memory in AI agents as storing “user preferences, past interactions, and learned insights”. It also warns that “users may unknowingly or inadvertently input sensitive personal data” into their prompts. The EDPB notes that the report does not necessarily reflect its official position.

France’s data protection authority, the CNIL, published a poster with its South Korean counterpart, the PIPC, listing six questions to ask before, during and after using generative AI. In French, it suggests using AI without creating an account where that is possible, choosing temporary mode, and always logging out, especially on a shared device.

The providers describe the same links from their side. Google explains that when you use its Gemini assistant signed out, “Google processes your data”, such as your IP address and browser type, but that “your data is not associated with your Google account.” OpenAI says that saved memories in ChatGPT “are stored separately from chat history”.

B. Key concepts: what it takes to use AI chatbots anonymously

Five links between a conversation and you, from the strongest to the one you write yourself.

The account: a name on every conversation

An account is the profile you sign in with: an email address, often a real name, sometimes a phone number or a payment method. When you are signed in, each conversation is stored under that profile. If you signed in “with” another account, such as Google or Apple, that account is now linked to the chatbot too.

Why it matters to you: The account is the link that turns a series of questions into a record about a named person. It is also the one you choose most directly.

History: the conversations joined together

Chat history is the list of past conversations saved to your account so that you can come back to them. Each one on its own may say little. Side by side, they show your work, your plans and your questions over months, and they are there wherever you sign in to that account.

Why it matters to you: History is what lets a single, harmless question become part of a pattern. Deleting old chats you no longer need shortens the pattern; it does not touch what you type next.

Memory: what the assistant keeps about you

Some assistants can remember details between conversations. OpenAI explains that ChatGPT has saved memories, details you ask it to keep or that it saves as useful context, and a setting called Reference chat history that lets it draw on past conversations. OpenAI adds that “ChatGPT can remember a name that you provide”, and, when Reference chat history is on, may also use the account name shown in the product. Google’s Gemini keeps saved information under Personal context.

Why it matters to you: Memory carries a detail from one conversation into the next, even when you did not repeat it. OpenAI also warns that “deleting a chat alone does not necessarily delete a separate saved memory created from that chat”, so the two need checking separately.

Using a chatbot without an account

Some services can be used signed out. OpenAI announced in April 2024 that ChatGPT could be used without signing up, with an account needed to save and review your chat history; check whether this is still offered where you are. Google says that “you can use some features in the Gemini web app” without signing in; saving your activity requires a Google Account.

Why it matters to you: Without an account, your conversations are not stored under your name, which removes the strongest link. It does not make you anonymous to the provider: Google, for example, still processes your IP address and browser type when you are signed out.

Temporary chats and the details in your prompt

A temporary chat is a conversation that is not kept in your history. OpenAI says that ChatGPT’s temporary chats “do not create or update memories”, and that you choose before you start whether one uses your existing memories. Google describes Gemini’s Temporary Chat as a place for “info that you don’t want Gemini to remember”, but notes that these chats are still “retained with your account” for a short period. And whatever the mode, the words you type travel with the conversation: a job title, a town and a recent event can describe one person without a name.

Why it matters to you: A temporary chat keeps one conversation out of your history and memory. It cannot loosen the link you write into the prompt yourself, which is why the details matter in every mode.

C. A practical example: Luca’s travel reviews

Luca is 36, lives in Milan and travels often for work. He uses an AI chatbot most days, signed in “with” his personal Google account, on his laptop and on his phone. It helps him plan trips, make sense of long train timetables and write polite emails to hotels.

He also writes, now and then, under a nickname on a travel forum: candid reviews of hotels and a few blunt opinions about airport services. Lately he has been asking the chatbot to polish those posts. One evening, reading a thread in which members guessed who might be behind some of the nicknames, he wondered what his chatbot account would show about him to anyone who could see it.

What changed when Luca looked

He spent fifteen minutes on the exercise in section D. Nothing he found was alarming, but several links were stronger than he had assumed:

• His account showed his full name, and because he had signed in with Google, the chatbot and his Google account were linked.
• His history held the forum drafts in the same list as his trip plans and hotel emails, months of them, visible on both his laptop and his phone.
• The memory settings showed that the assistant had saved that he lives in Milan, and the name of the company he travels for.
• Two of his drafts gave the exact dates of a recent trip. With the hotel emails beside them, anyone reading the history would need no guesswork.

He made three decisions. He deleted the forum drafts from his history and removed the saved memory about his employer. From now on he drafts nickname posts in a temporary chat that does not use his memories, and leaves exact dates out. And on the shared computer at the co-working space he sometimes uses, he now signs out of the chatbot before he leaves. He kept everything that helps him plan his travels, signed in as before. Not fewer conversations, but two separate kinds.

D. Try it yourself: review one chatbot account for its links (15 minutes)

Four short steps with the chatbot you use most. You do not have to change anything; the aim is to see the links first. If a setting name is not where this resource says, search the service’s help pages for the words in quotation marks.

Step 1 — Note how you are signed in (3 minutes)

• Open the account or profile section and write down what the service knows you by: an email address, a name, a phone number, a payment method.
• Note whether you signed in directly or “with” another account, such as Google or Apple.
• List the devices where you are signed in, including any shared ones at home or at work.

Step 2 — Check history and memory (5 minutes)

• Scroll through your chat history and note how far back it goes.
• Find the memory settings. In ChatGPT, OpenAI’s help page puts them under Settings › Personalization › Memory, with saved memories and Reference chat history. In Gemini, Google’s help page puts saved information under Settings › Personal context.
• Read what the assistant has saved about you. Note any detail that describes your work, your town or the people around you.

Step 3 — Spot the identifying combinations (4 minutes)

• Choose three conversations from your history, ideally of different kinds.
• In each one, underline the details that could point to you together, even without a name: a job, a place, a date, a rare situation.
• Ask yourself, for each one: would I be comfortable if this conversation sat next to my name?

Step 4 — Decide which conversations need fewer links (3 minutes)

Pick one option for the kind of conversation you most want to keep apart:

• Keep it as it is: the link is useful and you are comfortable with it, now that you have seen it.
• Loosen it: use a temporary chat that does not use your memories, delete the related chats and saved memories, and leave identifying details out of the prompt.
• Separate it: use the chatbot signed out for that kind of question, where the service allows it, knowing that the provider still receives your connection and what you type.

One link you decided about is worth more than every setting changed in a hurry.

Already covered in the Online Privacy course

AI Chatbot Privacy: Where Your Conversations End Up
It follows a conversation through storage, human review, training and connected apps; this resource looks at what ties that conversation back to you.

What Not to Share with AI Chatbots: Keep, Generalise, Share
It gives a rule for deciding what goes into a request; this resource adds the links that remain around a request, such as account, history and memory.

Opt Out of AI Training: Check Your Chatbot Settings
It covers whether your chats help train future models; this resource covers whether they can be connected to your identity.

E. Videos, articles and further resources

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

Institutional sources

European Data Protection Board (Support Pool of Experts) — AI Privacy Risks & Mitigations: Large Language Models (LLMs)
A 2025 expert report for data protection authorities on how LLM systems handle personal data, including memory across sessions; not an official EDPB position.
https://www.edpb.europa.eu/our-work-tools/our-documents/support-pool-experts-projects/ai-privacy-risks-mitigations-large_en

CNIL and PIPC — IA générative & vie privée
A one-page poster from the French and South Korean data protection authorities, in French: six questions to ask before, during and after using generative AI.
https://www.cnil.fr/sites/default/files/2026-05/affiche_ia_generative_et_vie_privee_fr.pdf

European Commission — Data protection explained
What counts as personal data under EU rules, including pieces of information that identify someone only together.
https://commission.europa.eu/law/law-topic/data-protection/data-protection-explained_en

Platform documentation

OpenAI Help Center — Memory in ChatGPT
Saved memories, Reference chat history, temporary chats and why deleting a chat and deleting a memory are separate steps.
https://help-lb.openai.com/en/articles/8590148

OpenAI — Start using ChatGPT instantly
The 2024 announcement that ChatGPT could be used without an account, and what an account adds.
https://openai.com/index/start-using-chatgpt-instantly/

Gemini Apps Help — What you need to sign in to Gemini Apps
Which features need a Google Account, and the note that some work signed out.
https://support.google.com/gemini/answer/13278668?hl=en-GB

Gemini Apps Help — Save info and reference past chats in Gemini Apps
Where Gemini keeps what it remembers about you, and how to edit or delete it.
https://support.google.com/gemini/answer/16413516?hl=en-GB

Gemini Apps Help — Gemini Apps Privacy Hub
What happens to your data when you use Gemini signed out, and how temporary chats are kept.
https://support.google.com/gemini/answer/13594961?hl=en-GB

Links checked in November 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: seeing what links your chatbot conversations to you before deciding which links to keep.

Skills

• Finding the account details, history and memory settings of the chatbot you use.
• Telling a deleted chat apart from a deleted memory.
• Starting a temporary chat, with or without memory, when it suits the question.

Awareness

• Understanding that account, history, memory and prompt details are separate links.
• Knowing that using a chatbot signed out removes the account link but not the provider’s view of your connection.
• Recognising that a few ordinary details in a prompt can point to one person without a name.

Secure Behaviour

• Keeping conversations you want apart in temporary chats or signed-out sessions, where the service allows it.
• Reviewing saved memories now and then, and removing what you no longer want carried over.
• Signing out of chatbots on shared devices.

G. Questions to sit with

• Which of your chatbot conversations would you most like not to sit next to your name, and why?
• Is there something the assistant remembers about you that you would not have chosen to tell it?
• Where does being signed in genuinely help you, and where are you simply used to it?

H. What to do now

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

1. The account and how you sign in

• R31 — Limit social login: signing in to a chatbot “with” a Google, Apple or social media account links the two. A separate sign-in keeps the chatbot apart from the rest of your accounts.
• Sign out of chatbots on shared computers and on devices other people use.

Minimum commitment: this week, check which account your main chatbot is signed in with, and on which devices.

2. History, memory and the words you type

• Open the memory settings of your chatbot (Personalization in ChatGPT, Personal context in Gemini) and remove anything you would rather it did not carry into other conversations.
• For a conversation you want to keep apart, use a temporary chat that does not use your memories, or the service signed out where that is offered.
• In any mode, leave out the details that together point to you: a place, a job, a date.

Minimum commitment: delete one old conversation, and any saved memory that came from it, that you would rather not keep linked to your account.

In short

• To use AI chatbots anonymously, look at four links: the account, the history, the memory and the details in your prompts.
• The account is the strongest link; signing out or using a temporary chat removes some links but not the provider’s view of your connection.
• Deleting a chat and deleting a memory are separate steps on some services.
• Anonymity with a chatbot stops at what you type: a few ordinary details can describe one person.

Related resources in this course

• Reverse Image Search Yourself: Who Can Find You from a Photo
• Stylometry: How Your Writing Style Can Identify You
• Protect Your Identity from Deepfakes: Limit, Notice, Act
• How Websites Recognise You: Cookies, Ad IDs and Logins

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