Paying £20 a month does not stop AI training on your data

This is the question we get asked more than any other, and it is the one where the internet is least reliable.

While checking the facts for this article we found four different sites making flatly contradictory claims about the same provider's policy, on the same day. Some of them looked authoritative. So everything below comes from the providers' own documentation and nothing else, with links, and we would encourage you to click them rather than take our word for it.

The rule almost everybody gets wrong

Most people assume the split is free versus paid — that the free tier is the product and paying makes you the customer.

It is not. The split is consumer versus business, and the £20-a-month plans sit on the consumer side.

Claude Pro and ChatGPT Plus are consumer products. Upgrading from free to £20, or from £20 to £200, moves you up the consumer ladder. It does not move you across the line. What moves you across the line is a team, business or enterprise plan, or using the API.

What each provider actually says

Anthropic, on commercial plans: "By default, we will not use your inputs or outputs from our commercial products (e.g. Claude for Work, Anthropic API, Claude Gov, etc.) to train our models." That is their own privacy documentation, and the default is the important word — it is off unless you turn it on.

Anthropic, on consumer plans (Free, Pro, Max): training is a setting you chose, or had chosen for you by not answering. Under the consumer terms updated in 2025, if you allow your data to be used for model improvement, retention extends to five years. If you decline, it stays at thirty days. Existing users were asked to pick by 8 October 2025. If you clicked through that dialog without reading it — and a lot of people did — you may not know which way you went.

OpenAI, on business products: data from ChatGPT Business, Enterprise, Edu and the API Platform is not used to train models unless you explicitly opt in. API data has worked this way since 1 March 2023.

OpenAI, on consumer plans (Free, Go, Plus, Pro): data may be used for training depending on whether you have opted out.

Two different companies, same shape. Business default is off. Consumer default is a setting, and the setting is not obvious.

Two things worth doing this week

Go and look at your setting. Not your plan — your setting. In both products it lives in privacy settings and takes under a minute. Most people we ask have never opened it, and a fair number find it set the way they would not have chosen. Note that changing it going forward does not retrospectively pull back conversations already used, though both providers exclude deleted chats and closed accounts from future training.

Work out whose account your business data is actually going into. This is the bigger exposure and almost nobody thinks about it.

The bit nobody is looking at

The Office for National Statistics published something in its survey of AI in UK businesses that ought to have got more attention than it did:

55% of employees report using AI personally. Only 35% of businesses report using AI at all.

That gap is not a rounding error. It is people using their own accounts to do their job — pasting a customer email into a free ChatGPT tab to get a better reply written, running a spreadsheet of figures through it to spot a pattern, drafting a quote. On personal accounts, with consumer defaults, invisibly, and with the best of intentions.

For a five-person business, this is the real question. You can be scrupulous about your own settings and still have your customer list, your pricing and your contract wording sitting in three personal accounts you have never seen and cannot audit.

And under UK GDPR the position is uncomfortable, because it does not care whose subscription it was. If that pasted email contained a customer's personal data, you are still the controller of it. The ICO's guidance on AI and data protection is worth a read on this, and its central point is not exotic: the same rules you already follow apply, and using a new tool does not create an exemption.

What we would actually suggest

Not "ban it." Banning it produces exactly what you had before, only now nobody tells you about it.

The useful version is duller. Decide which account the work happens in, and pay for the tier where the default is off. Say plainly which categories of thing may go in and which may not — customer personal data, anything under NDA, anything you would not want read aloud. Write that down once, in a page, rather than saying it at a meeting where three people were not listening.

That is not really an AI problem. It is the same conversation businesses had about Dropbox, and about personal phones before that, and the answer landed in the same place: give people a sanctioned way to do the thing they are going to do anyway.

The difference this time is that the sanctioned way is genuinely better. A business plan is not just a licence with the training switch flipped — it is where the account controls, the shared context and the connections into your actual systems live. Which is the point at which it stops being a chat window someone pastes into and starts being something that works on your data properly.

Sources

• Is my data used for model training? — Anthropic Privacy Center
• Updates to our Consumer Terms and Privacy Policy — Anthropic
• How your data is used to improve model performance — OpenAI
• Artificial intelligence in UK businesses: 2023 to 2026 — ONS
• Guidance on AI and data protection — ICO

Checked against the above on 8 September 2026. These policies change more often than most, so check the links rather than trusting the date on this page.