Why AI-generated terms and conditions do not automatically protect your business

It has never been easier to create a legal document.

Open an AI tool and type:

“Write comprehensive terms and conditions for my Dutch webshop.”

Within seconds, a professional-looking document appears on the screen.

Article numbers. Liability. Returns. Payment. Force majeure. Intellectual property.

It looks like legal work.

But that does not mean it works legally.

A document does not become good because it sounds legal. It becomes good because it aligns with your business, your sales process, and the law applicable to them.

And that is where AI-generated terms and conditions frequently go wrong.

Terms and conditions are not a universal disclaimer

Many entrepreneurs treat terms and conditions as a protective layer.

The stronger the wording, the better protected the business would be.

That is not how consumer law works.

When selling to consumers, you cannot simply write away statutory rights just because an AI model has created an impressive liability clause.

The ACM, for instance, emphasizes that terms and conditions must be provided before or at the conclusion of the contract and that consumers must have the opportunity to read and save them. Unfair terms can be declared void.

So even perfect content is of little help when the terms are only provided after the order has been placed.

The legal effect is not just in the text.

It is also in how and when the text becomes part of the agreement.

The biggest danger of AI: being convincingly wrong

AI has a trait that is particularly critical in legal work.

It can formulate an incorrect answer extremely convincingly.

Ask a generative system for terms and conditions, and it can produce clauses based on other jurisdictions, outdated regulations, or a combination of legal concepts that do not apply to your business.

A Dutch B2C webshop might, for example, receive terms that state:

“All sales are final.”

“No refunds.”

“We accept no liability under any circumstances.”

That sounds strict.

But strict is not the same as enforceable.

The ACM explicitly highlights a complete exclusion of liability towards consumers as an example of something that cannot simply be included in reasonable terms and conditions.

The problem is therefore not that AI cannot write legal language.

The problem is precisely that AI can write legal language very well without necessarily understanding the commercial reality behind it.

AI does not know your business until you explain the entire business

Suppose you sell a subscription.

Your terms must then align with how the subscription starts, how long it lasts, when payment takes place, how it is cancelled, when prices can change, and what happens upon renewal.

But if your website says one thing and your terms say another, you still have a problem.

The same applies to delivery times, returns, warranties, and complaints.

Furthermore, the ACM states that important information about the purchase itself cannot simply be hidden in the terms and conditions. Consumers must, prior to purchase, receive information on, among other things, the main characteristics, price, additional costs, delivery, and the cooling-off period.

That is why a good lawyer does not just work with Word documents.

They look at the operations.

Your website and terms must form one system

A legally compliant webshop consists of multiple layers.

The product page makes certain commitments.

The checkout makes certain commitments.

Your advertisement makes certain commitments.

Your return policy makes certain commitments.

Your customer service makes certain commitments.

And your terms and conditions must match these.

If the terms state that delivery takes five working days, while your advertisements promise “delivered tomorrow”?

Then you do not have a consistent legal system.

If your return policy states that sale items cannot be returned, while consumers are legally entitled to a cooling-off period for them?

Then a disclaimer in Article 14 of the terms and conditions will not solve that.

AI can generate a document.

But only your business can ensure that the document matches reality.

AI in customer service can even change your contract terms in practice

This is where a relatively new risk arises.

More and more webshops are using an AI chatbot to answer customer queries.

A customer asks:

“Can I return this product within thirty days?”

The chatbot replies:

“Yes, of course. You can return all products free of charge within thirty days.”

But your official return policy says fourteen days and return costs are to be borne by the customer.

What happens next when the consumer relies on the answer from your customer service?

That is precisely the kind of situation businesses must prevent beforehand.

An AI chatbot should not be able to improvise commercial or legal commitments outside of established policy.

Moreover, the AI Act makes transparency around directly interactive AI more topical. Since 2 August 2026, the transparency rules of Article 50 apply to certain AI systems, and clarity must be provided for directly interactive AI that someone is communicating with an AI system, in accordance with the applicable division of roles between provider and deployer.

A webshop deploying an AI assistant would therefore do well to configure this not only technically, but also legally.

Terms and conditions must now also address AI

For some businesses, it is wise to take the use of AI into account in contractual documentation.

Not by adding a random “Artificial Intelligence” clause, but by looking at what is actually relevant.

In B2B service provision, for instance, it can be important whether customer data may be processed in AI systems, who is responsible for AI output, whether confidential information may be entered into external models, and what degree of human oversight is used.

For software suppliers, it may also be relevant whether a service contains AI functionality, what limitations apply to it, and to what extent outputs must be verified.

In creative assignments, questions of intellectual property come into play once again.

The right clause depends on the business model.

That is precisely why automatically generated terms are usually just a starting point.

Do not forget your employees

Perhaps your business officially has no AI processes at all.

But is your marketing employee using ChatGPT?

Is your customer service having emails rewritten by AI?

Is your buyer having supplier contracts summarised?

Is a developer using AI to generate code?

Then the business may, in practice, already be using AI.

Since 2 February 2025, the AI Act has imposed obligations regarding AI literacy on providers and professional users of AI systems. Organisations must take measures to ensure that employees and other persons working with AI on their behalf are sufficiently aware of the functioning and context of those systems.

This makes an internal AI policy increasingly relevant.

Not because everyone needs to become a programmer.

But because employees must understand when AI output needs to be checked and what information must never simply be pasted into an external system.

Pasting an NDA into ChatGPT can also be a compliance issue

This is often underestimated.

An employee receives a confidential agreement from a supplier and asks a public AI tool:

“Summarise this contract and identify all risks.”

Operationally convenient.

Legally potentially problematic.

Not only because of personal data, but also because of confidentiality, contractual non-disclosure obligations, and the terms under which the chosen AI service processes data.

A business that uses AI seriously should therefore lay down which systems are permitted, for which types of information, and under what conditions.

The problem is therefore not AI

AI is excellent at speeding up legal work.

It can suggest structure.

Identify differences between documents.

Compare versions.

Categorise clauses.

Take stock of internal processes.

But AI must not replace the thinking process that determines why a clause is in the agreement in the first place.

The best terms and conditions are not a collection of heavy legal sentences.

They are a legal translation of a carefully structured business model.

Conclusion

You can have terms and conditions generated within thirty seconds today.

That is impressive.

But the relevant question is not how quickly you can produce a document.

The relevant question is:

Does the document protect your business when a dispute actually arises?

For that, the terms must align with your business, be correctly declared applicable, comply with mandatory law, and match what the website and organisation do in practice.

AI can help with this.

But only when someone knows what needs to be checked.

Ecompliance

Do you have terms and conditions that were once copied, generated by AI, or have not been updated for years?

Ecompliance can review not just the document, but indeed the entire legal customer journey: from advertisement and product page to checkout, terms, return process, and AI customer service.