Is It Safe to Talk to ChatGPT? UK Guide

Millions of people in the UK now use ChatGPT for everyday tasks, from writing emails to looking up recipes. If you're wondering whether it's safe to use, you're asking exactly the right question before you start.

We went through OpenAI's settings pages and privacy documentation to check each of the controls described here. Most of the fears people have are understandable but manageable. A few specific things (particularly around voice mode) are worth knowing about before you use them.

Is It Safe to Talk to ChatGPT?

For most everyday uses, yes. Asking ChatGPT to help you write a letter, explain a news story, suggest a recipe, or answer a general question carries very little risk. The data collected is similar to what most websites already collect when you browse.

The safety picture changes depending on how you use it:

• Text conversations: Generally safe. You have control over training dataThe examples used to teach an AI model. and can delete history
• Voice mode (speaking aloud): More to consider. Audio is retained for 30 days. Voice contains biometric data. See the voice section below
• Sharing sensitive personal details: Never advisable. Treat each chat like a postcard, not a sealed letter

The most important rule: ChatGPT is not confidential in the way a conversation with your doctor or solicitor is. But for the kind of questions most people want to ask, that distinction rarely matters.

What Does ChatGPT Do With Your Conversations?

When you send a message to ChatGPT, OpenAI stores it. Here is what that means in plain terms.

Your conversation history is saved to your account so you can refer back to previous chats. You can see this in the left-hand sidebar when you log in. You can delete individual conversations or your entire history at any time.

Training data: By default, OpenAI may use your conversations to improve future versions of ChatGPT. This is a separate toggle from deleting history. You can switch it off in Settings by going to Data Controls and turning off "Improve the model for everyone." Switching it off does not change how ChatGPT responds to you. It simply means your conversations are not used for training.

On the question of who can read your chats: OpenAI staff and contractors can theoretically access conversations. In practice, they review anonymised snippets for safety and quality purposes. The most significant security incident involving chat histories was a bug in March 2023, where some users briefly saw other users' chat titles. OpenAI fixed it within hours.

How long is data kept? Your conversation history is kept as long as your account exists. If you delete your account, conversations are removed within 30 days. You can also set history to stop saving new conversations at any time, in which case those conversations are deleted within 30 days automatically.

What Should You Never Say to ChatGPT?

Some information genuinely should not go into any AISoftware that can perform tasks that normally require human thinking, like understanding language or spotting patterns. chat, regardless of how trustworthy the service is. Our full guide to what not to share with ChatGPT covers this in detail, but the short version:

• Bank account numbers, sort codes, or card details: ChatGPT cannot help you with transactions; sharing these serves no purpose and creates risk
• Passwords or PINs: never paste these into any online service
• Your full name combined with your home address: either alone is often fine; combined with date of birth they become an identity-theft risk
• Medical diagnoses or test results: ChatGPT can offer general health information but you should not share your personal medical records with it
• Anything about an ongoing legal matter: conversations with ChatGPT do not carry legal privilege. US courts have ruled that AI chatbotA program you can talk to in text or voice. conversations can be discoverable in legal proceedings, and UK courts could reach similar conclusions
• Confidential work documents: company information, client data, or anything marked confidential should not be entered into a free consumer AI tool

For general questions (explaining a term in a letter you received, getting ideas for a birthday present, asking about a medication side effect in general terms), ChatGPT is fine. The distinction is between general information (safe) and your specific personal data (avoid).

Is ChatGPT Voice Mode Safe?

Voice mode (where you speak to ChatGPT and it speaks back) has a different privacy profile from text. If you plan to use it, these are the key points.

Audio is kept for 30 days. When you speak to ChatGPT in voice mode, OpenAI retains the audio recording for 30 days. This is longer than text conversations, which you can delete immediately. After 30 days the audio is automatically deleted, but transcripts may be kept longer under standard data policy unless you have opted out of training.

Your voice contains biometric information. The pitch and tone of your voice are as unique as a fingerprint. Even anonymised voice data can in theory re-identify a person. This is a concern researchers have raised: when OpenAI launched its Advanced Voice Mode, EU regulators temporarily blocked it over AI Act concerns before allowing access for paying users in late 2024. The UK is not subject to EU rules post-Brexit, but the underlying privacy consideration is the same.

GPT-Live (launched July 2026). In July 2026, OpenAI launched GPT-Live, a new full-duplex voice mode that allows natural, overlapping conversation rather than the turn-based format of older voice features. Separately, OpenAI's safety research conducted jointly with MIT Media Lab found that people who used conversational AI heavily reported feeling lonelier over time and less likely to seek human connection. OpenAI has committed to monitoring emotional wellbeing as part of GPT-Live's ongoing safety programme. GPT-Live includes safety interruption features: if the conversation moves toward harmful content, the system can pause, issue a safety message, redirect to a resource, or end the call entirely.

Practical advice for voice mode: Keep voice conversations topic-focused rather than personal. Using it to get a recipe read aloud or to dictate a shopping list is very different from using it as an emotional companion. The latter pattern is where the research shows risk of dependency developing.

You can now check whether a voice reply is genuinely from ChatGPT. Since 31 July 2026, audio generated through GPT-Live carries SynthID, an invisible watermark built with Google DeepMind that sits inside the sound itself, not the file name or metadata. You cannot hear it and it does not change how the voice sounds. OpenAI has also opened a public verification tool, so a family member (or you) can upload a recording and check whether it genuinely came from ChatGPT. This matters most if you ever receive a suspicious voice message and want to rule out an OpenAI-generated fake before you worry further; it is a provenance check, not a scam detector, so treat a "not verified" result as inconclusive rather than proof of fraud.

Voice mode has parental safeguards too. A parent or guardian can switch off ChatGPT Voice entirely for a teenager's account through OpenAI's parental controls. If the system detects a child steering a voice conversation towards self-harm, it notifies the parent. Useful to know if you are helping a grandchild set up their own account, even though the safety points above are aimed mainly at your own use.

There is no opt-out of audio retention for voice mode the way there is for text training. If this concerns you, use the text interface instead, where you have more control.

Your Rights as a UK User

UK users have legal rights over their data under the Data Protection Act 2018, which implements GDPR in UK law. These rights apply to OpenAI regardless of where the company is based.

• Right to access: You can request a copy of all personal data OpenAI holds about you. OpenAI must provide this within 30 calendar days
• Right to deletion: You can ask OpenAI to delete your personal data: conversations, account records, and voice data
• Right to correction: If OpenAI holds inaccurate information about you, you can ask for it to be corrected
• Right to object: You can object to your data being used for training. The toggle in settings exercises this right, but you can also make a formal request
• Right to portability: You can request your data in a format that allows you to transfer it to another service

To make any of these requests formally, use OpenAI's personal data request page. OpenAI must respond within 30 days. If you are not satisfied with the response, you can complain to the Information Commissioner's Office (ICO), the UK's independent data protection authority.

See our guide on ChatGPT safety and ChatGPT privacy settings for more on these rights in practice.

Five Practical Safety Steps

These five steps take less than ten minutes to set up and cover the main risks.

1. Use a strong, unique password

If you haven't already, set a password for ChatGPT that you don't use anywhere else. If your account were ever compromised, a unique password limits the damage to that account only. A phrase like "BlueTea2026!Garden" is both memorable and strong. See our guide to creating your ChatGPT account safely for step-by-step instructions on this.

2. Turn off training data

In ChatGPT, go to Settings, then Data Controls, and toggle off "Improve the model for everyone." This stops your conversations being used to train future versions. It does not affect how ChatGPT responds to you.

3. Delete conversation history regularly

In the left-hand sidebar, hover over any conversation and click the three dots to delete it. You can also go to Settings, then Data Controls, and delete all conversations at once. Our guide to deleting your ChatGPT conversation history walks through this with screenshots.

4. Keep personal details out of your messages

Before you type anything sensitive, ask yourself: would I be comfortable if a stranger read this? If the answer is no, rephrase the question in general terms. "What are the rules about council tax for pensioners?" is fine. "My name is [name], I live at [address], and I owe [amount]" is not necessary and adds risk.

5. For voice mode: keep it topic-focused

If you use voice mode, use it for practical tasks rather than personal conversations. Getting a recipe read aloud, hearing a news summary, asking what a word means, or practising a foreign language phrase are all perfectly reasonable uses. Long personal conversations (about loneliness, worries, or relationships) are where research suggests emotional dependency can develop. Use ChatGPT for tasks; use people for connection.

If you'd like step-by-step guidance through all of this, our WellWired Academy takes you through AI safely, at your own pace, in plain English.

Getting the Best Out of Your ChatGPT Conversations

Once you feel confident about safety, good results from ChatGPT come from asking clearly. You don't need special technical language. Plain English works well.

Tell ChatGPT what you want, who it's for, and what kind of result you'd like. That combination produces far better answers than a vague question.

Vague: "Write about pensions."
Better: "Write a short explanation of pensions for a beginner in the UK. Keep it friendly and under 200 words."

Vague: "Help me plan a trip."
Better: "Plan a three-day trip to York for two adults, no car, moderate budget."

These promptsThe message or instruction you give an AI. you can copy and paste to get started:

• "Explain [topic] in simple terms for a complete beginner."
• "Write a polite email to [person] about [issue]. Keep it short."
• "Give me three options and explain the differences."
• "Summarise this in five bullet points: [paste text]."
• "Rewrite this so it sounds warmer: [paste draft]."
• "Create a checklist for [task] with six to eight steps."

If the first answer isn't quite right, tell ChatGPT what to change. "Make it shorter," "use simpler words," or "add an example" all work well. You don't need to start the conversation over. For more ideas, see our ChatGPT prompts for beginners guide.

One important reminder: ChatGPT can be wrong. For health topics, always verify with the NHS website. For financial or legal matters, use ChatGPT only for background understanding, then consult a professional. AI is a useful starting point, not a final authority.

When you're ready to go deeper, our AI safety hub brings together everything WellWired has written about using AI safely as a UK reader.

Frequently Asked Questions