Can I put client data into ChatGPT? An Irish business owner's answer

It's the question that comes up in every workshop, usually about ten minutes in, usually from someone who has already done it and is now slightly worried. The honest answer has two halves.

Short version: on a free or personal ChatGPT account, no — not client data with identifying information in it. On a business or enterprise tier with a data processing agreement in place, yes, within sensible limits.

The rest of this is why that distinction matters more than anything else you'll read on the subject.

The difference between the tiers is legal, not technical

The interface is the same. The model is broadly the same. What differs is the contract sitting behind it.

Free and Plus (personal) accounts. No data processing agreement with your business. Historically, conversations on consumer tiers may be used to improve models unless you opt out, and the opt-out is a setting most people never touch. No admin visibility, no organisational controls, no meaningful audit trail. For personal use, entirely fine. For your clients' personal data, not appropriate.

Team, Business, and Enterprise accounts. A data processing agreement is available, business data is contractually excluded from model training by default, there are admin controls and retention settings, and EU data residency is available on the higher tiers. This is a normal processor relationship of the kind your business already has with its accounting software and its email provider.

Under GDPR your obligations don't change based on how convenient the tool is. They change based on whether you have an agreement with the processor. That's the whole distinction.

What "client data" actually covers

Broader than people assume. Names and email addresses obviously, but also: an invoice with a company and a contact person on it, a contract with signatories, a set of accounts identifiable to a client, a support email thread, a CV, a photograph, an IP address in a log file. If a living person could be identified from it directly or in combination with something else, it's personal data.

Special category data — health, ethnicity, religious belief, trade union membership, biometrics, criminal offence data — sits in a stricter tier again and shouldn't go into a general-purpose AI tool at all.

The three-question test

Before pasting anything, run this:

1. Do I actually need the identifying details in this prompt? Usually not. "Redraft this email to be firmer" works exactly as well with the name replaced by a placeholder. Strip the identifiers and the question mostly disappears.

2. Am I on the approved business account? If you're not sure, you're on the personal one.

3. Would I be comfortable if the client knew? Not a legal test, but a very reliable instinct. If the honest answer is "they'd be uneasy", that's usually the right signal.

What's genuinely safe on any tier

Anonymised or synthetic examples. Publicly available information. Your own internal process documentation with no personal details. General questions about how to approach something. Drafting from scratch. Summarising a document you've already redacted. Formatting, tone, and structure work.

That covers most of what people actually want AI for, which is the encouraging part of this answer.

Not sure whether your current AI use is compliant? Take the free AI Scorecard — it takes five minutes and gives you a structured read on your data exposure and what to do next.

What to do if it's already happened

Almost certainly it has, and almost certainly nobody meant any harm. The proportionate response:

Find out what was shared and roughly when — ask without blame, or you'll get incomplete answers. Turn off training data use in the settings of any personal accounts still in play. Delete the conversation history where you can. Assess whether the exposure is significant enough to warrant a breach entry in your internal log; in most cases it will be a recorded near-miss rather than a notifiable breach, but that's a judgement to make deliberately rather than by default. Then buy the business tier, so the same thing doesn't happen next Tuesday.

The practical answer for most Irish SMEs

Buy one business-tier AI subscription. Tell your team it's the only place client information goes. Teach the habit of stripping identifiers before pasting. Write down four rules and spend twenty minutes walking through them.

That converts an ongoing, invisible exposure into an ordinary, documented processor relationship — the same kind you already have with every other piece of software in the business.

Our fuller piece on GDPR and AI tools sets out the four-rule policy in detail, and the EU AI Act guide covers the separate — and lighter — set of obligations arriving under that regime.

If you want a structured read on where your business currently stands, the free Scorecard takes five minutes and gives you a written answer.