“Does ChatGPT keep what I write to it?” The question comes up in every organisation discovering generative AI. The answer depends less on the tool than on how it is used: with a free personal account, an individual subscription or a business plan managed by the organisation, the rules are not the same. This article sets out what OpenAI publishes on its own pages, plan by plan, and then what remains the organisation’s responsibility.
What happens when you send a message
ChatGPT is an online service. The text you type, the files you attach and the answers are processed on OpenAI’s servers, then saved in the account history. Nothing surprising there: that is how every hosted service works. What matters is what happens next: how long this content is kept, who can access it and whether it can be used to improve the models.
The diagram below shows the journey of a message containing a name, an IBAN and a piece of health information. Without a prior check, this data is sent as it is. With a local check, it can be replaced with tags before it leaves the computer.
The message is sent as it is: the name, IBAN and health information reach the vendor, which processes and retains them under the rules of the plan being used.
Personal accounts: training and settings
For services aimed at individuals (free ChatGPT and individual subscriptions), OpenAI states that it may use the content submitted to improve the performance of its models.1 The same page also advises against entering sensitive information that you would not want reviewed or used.1
Users stay in control here. When the “Improve the model for everyone” setting is turned off, new conversations are not used to train OpenAI’s models.2 “Temporary chats” are not used to improve the models either, but may be kept for up to 30 days for safety reasons.2 A detail that is often overlooked: when you give feedback on an answer (thumbs up or down), the entire associated conversation may be used for training.2
This approach is not unique to OpenAI. Google states that human reviewers read some Gemini app conversations and asks users not to enter confidential information.10 Anthropic, for its part, has asked users of its consumer Claude plans to choose whether they allow their data to be used for training, with retention extended to five years if they agree.11
ChatGPT Business and Enterprise: what OpenAI provides
Business plans rest on a different framework. For ChatGPT Business (the former Team plan, renamed on 29 August 20256), ChatGPT Enterprise, ChatGPT Edu and the API platform, OpenAI states that it does not train its models on customer data by default.3 The vendor also states that the customer owns its inputs and outputs, where allowed by law, that data is encrypted at rest (AES-256) and in transit (TLS 1.2 or higher), and that a SOC 2 audit has been carried out.3
For the GDPR, one point is decisive: OpenAI can sign a data processing addendum (DPA) with its customers for ChatGPT Business, ChatGPT Enterprise and the API.3 Some eligible Enterprise, Edu and API customers can also choose to store their content at rest in Europe.7
- Use to improve the models
- Possible, can be disabledContent may be used to improve the models. The “Improve the model for everyone” setting lets users opt out for new conversations.
- Ownership of content
- Consumer termsGoverned by the terms of use and privacy policy of the consumer services.
- Data processing agreement (DPA, GDPR)
- Not availableThe account belongs to the individual, not the organisation: no processing agreement is signed with the employer.
- Administration by the organisation
- NoneThe organisation cannot see these accounts, and can neither configure nor close them.
How long are conversations kept?
A deleted conversation is not erased instantly. According to OpenAI, it is scheduled for permanent deletion from its systems within 30 days, unless it has already been disassociated from the account, or a security or legal obligation requires it to be kept for longer.4 The same 30-day rule appears in the Enterprise commitments.3
The “legal obligation” is not theoretical. In the lawsuit brought by The New York Times, OpenAI had to retain consumer and API content for several months. The vendor states that this obligation ended on 26 September 2025, that it securely retains a limited set of data from April to September 2025, and that ChatGPT Enterprise and Edu were not affected.5
Confidential and sensitive data: the real issue
For an organisation, the main risk is not the tool itself but what gets pasted into it. A busy employee copies a customer email to rephrase it, a payroll spreadsheet to summarise it, a contract to extract its clauses. Each of these actions passes personal data to a third party.
The GDPR requires the data processed to be “adequate, relevant and limited to what is necessary” (data minimisation principle, Article 5).9 Certain categories, such as health data, enjoy enhanced protection (Article 9).9 And it is the organisation, as controller, that must be able to demonstrate compliance with these principles (Article 24).9 In France, the CNIL (the French data protection authority) makes the point for generative AI: the user organisation is the one held responsible if its staff misuse AI.8
Very often, the personal data is not needed for the task. Rephrasing a message does not require the customer’s real name or IBAN: a tag is enough, and the answer remains usable.
Can you rephrase this message for Mr James Whitfield[NAME_1]?
“Further to your call of 3 October, we confirm the refund to IBAN GB29 NWBK 6016 1331 9268 19[IBAN_1].”
He can be reached on 07700 900123[PHONE_1] or at j.whitfield@example.com[EMAIL_1].
Good habits for organisations
France’s data protection authority, the CNIL, recommends governing the use of generative AI through an internal policy or charter that clearly distinguishes permitted and prohibited uses.8 In practice:
- Prefer a business account. A plan managed by the organisation, with a DPA, provides a contractual framework that a personal account cannot. For non-confidential uses, the CNIL also considers a consumer service used with dedicated work email addresses, turning off data reuse where possible.8
- Check the settings. On an individual account, turn off “Improve the model for everyone” and use temporary chats for one-off topics.2
- Minimise before sending. Remove or replace names, contact details, bank details, health data and trade secrets.
- Train your teams. Explain why, rather than simply prohibiting: that is what makes a rule stick.
- Use tools to support minimisation. A check on the computer can detect sensitive data at the moment of sending and offer to anonymise it. That is the role of Tacite-IA, presented below, among other possible measures.
Frequently asked questions
Does ChatGPT use my conversations for training?
For personal accounts, OpenAI states that it may use content to improve its models, unless the user turns off the “Improve the model for everyone” setting. For ChatGPT Business, Enterprise, Edu and the API, OpenAI states that it does not train its models on customer data by default.
How long does ChatGPT keep a deleted conversation?
According to OpenAI, a deleted conversation is scheduled for permanent deletion within 30 days, unless a security or legal obligation requires it to be kept for longer.
Can you enter personal data into ChatGPT?
Technically yes, but the organisation remains responsible for that data under the GDPR. The data minimisation principle means sending only what is necessary: a name or an IBAN can almost always be replaced with a tag without affecting the answer.
Are temporary chats really deleted?
OpenAI states that temporary chats are not used to improve the models and may be kept for up to 30 days for safety reasons.
Sources
- How OpenAI handles data in consumer services. OpenAI Help Center. Accessed on 5 October 2026.
- Data controls in ChatGPT. OpenAI Help Center. Accessed on 5 October 2026.
- Enterprise privacy at OpenAI. OpenAI, page updated on 8 January 2026. Accessed on 5 October 2026.
- Chat and file retention in ChatGPT. OpenAI Help Center. Accessed on 5 October 2026.
- How we’re responding to The New York Times’ data demands in order to protect user privacy. OpenAI, update of 22 October 2025. Accessed on 5 October 2026.
- ChatGPT Business - Overview. OpenAI Help Center. Accessed on 5 October 2026.
- Business data privacy, security, and compliance. OpenAI. Accessed on 5 October 2026.
- Les questions-réponses de la CNIL sur l’utilisation d’un système d’IA générative. CNIL (French data protection authority), 18 July 2024. In French. Accessed on 5 October 2026.
- Regulation (EU) 2016/679 on the protection of natural persons with regard to the processing of personal data (General Data Protection Regulation). EUR-Lex. Accessed on 5 October 2026.
- Gemini Apps Privacy Hub. Google. Accessed on 5 October 2026.
- Updates to Consumer Terms and Privacy Policy. Anthropic, 28 August 2025. Accessed on 5 October 2026.
ChatGPT is a trademark of OpenAI, Copilot of Microsoft, Claude of Anthropic, Gemini of Google. Tacite-IA is not affiliated with any of these vendors. This article is for information only and does not constitute legal advice.