Does ChatGPT train on my data by default?

On the consumer free and Plus tiers, yes - unless you turn it off. OpenAI uses conversations from these tiers to improve its models through a process called reinforcement learning from human feedback (RLHF) and related techniques. You can opt out in Settings > Data Controls > Improve the model for everyone. On the Team, Enterprise, and API tiers, your data is not used for training by default - no opt-out required.

If I turn off chat history, does that stop training?

Turning off chat history removes conversations from your sidebar and prevents new conversations from being used for training. It's essentially the same as the training opt-out toggle - they're linked settings. OpenAI still retains conversations for up to 30 days for safety and abuse monitoring, even with history off. After 30 days they are deleted. So: turning off history does prevent training, but doesn't prevent short-term retention for safety purposes.

Can I delete my existing conversations to stop them being used for training?

You can delete conversations from your history, which removes them from the interface. OpenAI's policy is that deleted conversations are retained for up to 30 days for safety monitoring before being deleted from their systems. Whether conversations you've already had - before enabling the opt-out - have been used in any training runs is not something OpenAI discloses. The opt-out is forward-looking: it stops future conversations from being used, but doesn't retroactively remove past data from any models already trained on it.

Does ChatGPT Plus stop training on my data?

No. ChatGPT Plus costs £16/mo and gives you faster access, GPT-4o, image generation, and custom GPTs - but it provides no additional data protection compared to the free tier. Your conversations can still be used for training on Plus unless you opt out in Settings > Data Controls. The meaningful data protection jump is from Plus to Team, which includes a Data Processing Agreement and a commitment not to train on your data.

Does OpenAI use my conversations to train GPT-5 or future models?

OpenAI's policy is that conversations from opt-in (non-opted-out) free and Plus tier users may be used to improve OpenAI models generally. This isn't limited to the current model - it could include future versions. OpenAI does not commit to which specific model runs any given conversation feeds into. If you have opted out, or are on a Team/Enterprise/API plan, your conversations are not used for any model training.

Is the API different from chatgpt.com for training?

Yes - meaningfully different. API usage is governed by OpenAI's API terms, which include a commitment not to use your API inputs and outputs to train OpenAI models by default. You also get access to OpenAI's Data Processing Agreement through the API terms. This makes the API a better choice than the consumer interface for processing sensitive content, though it requires technical integration rather than a browser interface.

What does 'training' actually involve - does a human read my conversations?

Not necessarily. AI training involves processing large volumes of data algorithmically - the model learns from patterns across many conversations, not from a human reading each one. However, OpenAI does employ human reviewers and contractors who review a sample of conversations for quality, safety, and to generate training labels. These reviewers are subject to confidentiality obligations. The risk isn't that your specific conversation will be read by a named individual - it's that the content could contribute to model training and that a small sample may be reviewed by contractors.

Data & privacy

Does ChatGPT Train on My Data?

Per-plan breakdown for 2026. Which tiers train on your data, which don't, and the one setting that changes everything on the free tier.

Book a discovery call

Data & privacy

Does ChatGPT Train on My Data?

Per-plan breakdown for 2026. Which tiers train on your data, which don't, and the one setting that changes everything on the free tier.

Book a discovery call Client data safety guide

5ChatGPT plan tiers - each with a different training policy you need to know

1toggle in Settings > Data Controls that stops training on the free and Plus tiers

30days OpenAI retains conversations even after you delete them, for safety monitoring

2026verified against OpenAI's current Privacy Policy and Terms of Use, May this year

Quick answer

Free and Plus: yes, unless you opt out. Team, Enterprise, and API: no, by default and by contract. The opt-out on the free tier is a single toggle in Settings > Data Controls - it takes 30 seconds and takes effect immediately.

The question "does ChatGPT train on my data" has a different answer depending on which version of ChatGPT you're using. Most small businesses use the free web interface or ChatGPT Plus - both of which train on your conversations by default, though you can turn this off. The paid tiers above Plus provide a contractual commitment that your data won't be used for training at all.

This page also explains what "training" actually involves - because the risk isn't quite what most people imagine, and being specific about it helps you make a better decision about which tier to use.

Per-plan breakdown

Which ChatGPT plans train on your data in 2026

Five plan tiers, five different policies. The key distinction is between consumer plans (Free, Plus) and business plans (Team, Enterprise, API). Note that ChatGPT Plus provides no training-related improvement over the free tier - it's a common misconception.

Plan	Price	Trains on data?	DPA	Human review	Notes
Free	£0	Opt-out only	No	Possible	Trains by default. Toggle off in Settings > Data Controls > Improve the model for everyone.
Plus	£16/mo	Opt-out only	No	Possible	Same training policy as Free. Plus upgrades speed and model access - not data protections.
Team	~£20/user/mo	No	Yes	Limited, confidential	Not used for training by default. DPA included. Retention for safety monitoring still applies.
Enterprise	Custom	No	Yes	Contractually restricted	No training, extended retention controls, SSO, audit logs. Human review access is contractually restricted.
API	Usage-based	No	Yes	Limited, confidential	API terms commit to no training by default. DPA available. Requires technical integration.

Source: OpenAI Privacy Policy, Terms of Use, and API Data Usage Policy - verified May 2026. "Human review" refers to contractor or staff access for quality, safety, or training-label purposes.

These plans train on your data by default

Free tier - opt out to stop
ChatGPT Plus - opt out to stop

These plans never train on your data

ChatGPT Team (~£20/user/mo)
ChatGPT Enterprise
OpenAI API

How to turn training off

Opting out on the free and Plus tiers - step by step

This applies to the consumer free and Plus tiers only. If you're on Team, Enterprise, or using the API, training is already off by default - you don't need to do anything.

Open Settings

Click your name or profile icon in the bottom-left corner of ChatGPT. Select 'Settings' from the menu.

Go to Data Controls

In the Settings panel, click 'Data Controls'. This is where all training and history settings live.

Toggle off 'Improve the model for everyone'

Switch this toggle to the off position. This stops your future conversations from being used to train OpenAI models. The toggle takes effect immediately.

Optionally turn off chat history

Turning off 'Save chat history and training' also stops training AND removes conversations from your sidebar. Use this if you want both. Note: this is essentially the same opt-out through a different toggle - they are linked.

Verify the change is saved

Close and reopen Settings. Confirm the toggle remains off. There's no confirmation email - the change is saved automatically.

Important: this is forward-looking only

Opting out stops future conversations from being used for training. It doesn't retroactively remove past conversations from any model training runs that may have already taken place before you enabled the opt-out. OpenAI also retains conversations for 30 days for safety monitoring regardless of this setting.

What "training" actually means

What OpenAI is actually doing with your conversations

"Training on your data" sounds alarming but the mechanism matters. Understanding what's actually happening helps you assess the real risk for your specific situation - and avoid both over-reaction and under-reaction.

RLHF (Reinforcement Learning from Human Feedback)

Human raters score ChatGPT's responses, and the model is adjusted to produce more of what scores well

Your conversations might be shown to a human contractor who rates whether the response was good or bad. That rating helps shape future model behaviour. This is why the 'human review' column matters.

Supervised fine-tuning

Examples of ideal conversations are used to teach the model the right format and style

Actual conversations (or edited versions of them) can serve as training examples. If your conversation is selected, the content of what you wrote could directly influence how the model is trained to respond to similar inputs.

Pre-training data

Vast amounts of text scraped from the internet and other sources are used to build the model's foundational knowledge

ChatGPT conversations are not typically used for pre-training (that uses web-scale data). The training risk from your conversations is at the fine-tuning and RLHF stage - which is more targeted but uses real user interactions.

Safety and abuse monitoring

All conversations are reviewed algorithmically (and sometimes by humans) for harmful content

This happens regardless of your training opt-out. The 30-day retention period exists for this purpose. It's separate from the training question - opting out of training doesn't opt you out of safety monitoring.

The practical takeaway

The realistic training risk for most conversations is low - OpenAI processes enormous volumes of data and the chance of any specific conversation having a meaningful influence on model behaviour is small. The risk that matters for businesses is the 30-day safety retention window and the possibility of human contractor review. That risk doesn't disappear with the training opt-out. It only fully disappears on Team, Enterprise, or API plans.

Need help choosing the right setup?

We'll help you choose the right AI tier for your business and get it set up properly

Whether you need to upgrade your ChatGPT plan, draft an internal AI policy, or understand your obligations under UK GDPR, our consultancy sessions cover the practical and legal side in plain language.

Book a discovery call Client data safety guide

Common questions

FAQs

Get in touch