Who Owns the Work Your AI Produces?

If you use an AI tool to write a proposal, design a logo, or draft a report for a client, a reasonable question follows: who actually owns that output? The answer is not as clean as most vendors suggest, and the gap between "you own it" and "you can protect it" matters more than most business owners realise.

This article covers the practical questions. What do major AI vendor terms typically say? Why does copyright protection sit separately from ownership assignment? What does this mean when you bill a client for AI-assisted work? And what should your own contracts say?

What the Major Vendors Typically Say

The short version: most large AI vendors assign output rights to the user. OpenAI, for example, states in its terms of service that users own the output generated from their inputs. Google's Gemini terms take a similar position. Adobe Firefly states that commercially generated content belongs to the user. Microsoft Copilot's terms largely follow the same pattern.

But "typically" is doing heavy work in that sentence. Vendor terms change. They vary between personal and commercial plans. They differ depending on whether you have an enterprise agreement or a consumer account. And they include carve-outs that matter.

Common carve-outs across vendor terms include:

• The vendor retains a licence to use your inputs and outputs for model training unless you opt out (and opting out is not always the default).
• Outputs that are similar or identical to outputs given to other users are not exclusively yours. If ten businesses ask the same prompt, they may all receive similar text, and none of them has a stronger claim than the others.
• The vendor does not guarantee that output is free of third-party IP claims. You use it at your own risk.
• Some tools restrict commercial use at certain subscription tiers.

The checklist at the end of this article covers exactly what to look for in any vendor's terms.

This is general information. Your vendor's specific terms govern your specific situation, and those terms can be updated without much notice.

"You Own It" and "It Is Protectable" Are Different Claims

Here is where most business owners hit a problem.

A vendor saying "you own the output" is a contractual statement. It tells you what the vendor is assigning to you. It does not tell you whether copyright law in your country will actually protect that output against someone else copying it.

Copyright, in most legal systems, requires human authorship. The United States Copyright Office has declined to register works that are purely AI-generated, on the grounds that there is no human author. The UK, Australia, and the EU are all working through similar questions, with different draft positions and case law emerging at different speeds.

What this means in practice:

A piece of writing you produce entirely by typing a one-line prompt and accepting the output verbatim sits in a grey zone. A piece of writing where you develop a detailed brief, edit the draft substantially, select and arrange sections, and add original ideas has more human creative input. Courts and IP offices are more likely to recognise the second as protectable, though even this is unsettled law.

For businesses billing clients, this creates a real exposure. You may own an assignment from your vendor. You may not be able to stop a competitor from using very similar output if they run a similar prompt. And your client may ask you directly whether the work you are delivering is original and protectable.

None of this means AI-assisted work is useless or unusable. It means you need to be clear, with yourself and with clients, about what you are actually delivering.

Again: this is general information, not legal advice. IP law varies significantly by jurisdiction. Get specific advice for your situation.

What It Means for Work You Bill to Clients

If you use AI tools in client work, the practical risks fall into a few areas.

Representation of originality. Some client contracts include a warranty that the work delivered is original and does not infringe third-party rights. If you warrant that and the work is AI-generated with thin human input, you may be in breach before anyone has even complained about copying.

Ownership transfer. A client contract often says something like "all intellectual property in the deliverables transfers to the client on payment." If you cannot establish that you own strong IP in the first place, you cannot reliably transfer it. The client ends up with something that looks like IP but may not hold up.

Identical output risk. As noted above, another user with a similar prompt could receive output very close to what you delivered. This is not theoretical. It is a structural feature of how large language models work. A client who later finds near-identical content elsewhere has a legitimate grievance, even if no one acted in bad faith.

Disclosure. In some industries and some countries, there are emerging requirements or strong norms around disclosing AI involvement in deliverables. Failing to disclose when a client would reasonably want to know is a reputational and contractual risk.

The straightforward response is transparency. Be clear about your process. Do not warrant more than you can deliver. Adjust your contract language to reflect reality.

What to Put in Your Own Terms

You do not need complex legal language to handle this reasonably. You need to be accurate about what you are providing.

Consider these adjustments to your standard terms:

Define what you are delivering. Instead of warranting original copyright ownership, describe the deliverable as "professional services output produced using a combination of human expertise and AI-assisted tools." This is accurate and does not overstate.

Limit your IP warranty. State that you assign to the client whatever rights you hold in the output, but do not warrant that the output is registerable as copyright or that it is entirely unique. This protects you while still giving the client a clean transfer of whatever rights exist.

Include an AI disclosure clause. State clearly that your services may involve AI tools and name the categories of tools you use (text generation, image generation, coding assistance, and so on). This removes any argument that you concealed the process.

Reference the vendor terms. Note that the output is also subject to the terms of the underlying AI platform, and that those terms may include their own limitations on exclusivity and protection.

Recommend the client takes independent advice if the IP status of the deliverable is material to their business. You are not their lawyer.

This is general guidance on contract drafting, not legal advice. Have a qualified lawyer review your client terms before you rely on them.

Checklist: What to Check in Any AI Vendor's Terms

Before you use a vendor's output in commercial work, find answers to these specific points in their documentation:

• Does the vendor assign output rights to users on your plan type (personal, pro, enterprise)?
• Does the vendor retain a licence to use your inputs or outputs for training? Can you opt out?
• Does the vendor warrant that output is free from third-party IP claims, or do they disclaim that?
• Are there restrictions on commercial use at your subscription tier?
• Does the vendor acknowledge that similar outputs may be delivered to multiple users?
• When were the terms last updated, and do you have a process to check for changes?

This is a starting checklist, not a complete legal review. If the answer to any of these points creates a problem for your client work, get specific advice before you proceed.

A Practical Starting Point

Getting your AI use onto a solid footing does not require a legal retainer before you can do anything. It requires clear internal rules about how AI tools are used, what gets disclosed, and what your contracts actually say.

Plainstart's free AI Usage Policy gives you a plain-language starting point: a ready-to-adapt policy document that covers acceptable use, data handling, and output standards for your team. It is free, it is plain, and it is written for small and medium businesses, not legal departments.

Download the free AI Usage Policy at Plainstart.

AI adoption, done properly.