Copyright and AI-generated Content for Business

Content made by AI is not automatically yours in the way content written by your own staff is. Whether you own it depends on how much of it a person shaped, which country’s law applies and what the tool’s terms say. The practical risk for most businesses is smaller than the headlines suggest, but it is real, and it sits in a few specific places.

This article explains who owns what AI produces, what the rules in the EU, the UK and the US currently say, where the genuine risks are for a business publishing AI-assisted work, and what to put in a one-page rule so your team knows where the lines are. It is general information and not legal advice. For anything high value, such as a logo or a campaign built around a single image, ask a qualified adviser in the country where you trade.

Key takeaways

• Copyright protects human creative work. Output from a tool with little human input is unlikely to be protected in the EU or the US, so you may be unable to stop others copying it.
• What you add counts. Writing, editing, selecting and arranging with real creative choices can be protected, even when AI produced the first draft.
• The tool’s terms and the law answer different questions. Terms decide whether you may use the output. The law decides whether anyone owns it.
• The main exposure is publishing output that copies something else, such as a recognisable character, a person’s likeness or a brand. Ordinary drafting carries far less risk than images and logos.
• The UK still has a special rule for computer-generated works, but it is under review and looks likely to go.
• Write it down. A short rule on which tools are approved, what needs a human check and when to disclose removes most of the risk.

Who owns what AI writes?

Nobody may own it, and that is the awkward starting point. Copyright exists to reward human authorship. Where there is no human author making creative choices, many legal systems treat the result as having no owner, which means it can be reused by anyone.

This does not mean you cannot use what a tool produces. It means that your right to use it comes from the tool’s terms of service, not from copyright law, and that you may be unable to stop a competitor reusing the same text or image. For a routine email or an internal summary, nobody minds. For a logo, a campaign image or a document you intend to sell, it matters a great deal.

Keep three ideas apart, because people mix them up:

1. Ownership under the law. Does a copyright exist, and who holds it?
2. Permission under the terms. Does the provider allow you to use the output commercially?
3. Infringement risk. Does the output copy someone else’s protected work?

Each has a different answer, and good practice covers all three.

Can you copyright AI-generated Content?

Partly, and only through the human contribution. The more a person decides about the expression, meaning the words, the structure, the composition and the choices between options, the stronger the claim. The less a person decides, the weaker it is.

Think of a spectrum. At one end, someone types a one-line request and publishes whatever comes back. A claim to copyright there is weak. At the other end, a writer drafts a report, uses a generative AI tool to suggest alternative phrasing, rejects most of it, rewrites the rest and reorders the sections. The report is a human work that used a tool, and it is protected in the ordinary way.

Most business use falls in between, so the sensible question is not “was AI involved?” but “which parts did a person decide?”. If you cannot answer that for something valuable, you have a gap.

Keep evidence. Save the drafts, the prompts you used and the edits you made. If you ever need to show that a document or image reflects your own choices, a trail of versions is far better than an assertion.

What does the Law say in the EU and the UK?

The EU has no rule specific to AI output. Copyright across the Union rests on the idea that a work must be the author’s own intellectual creation, and a long line of court decisions applies that test. A 2025 study prepared for the EU legislature argues that content produced autonomously by AI, without meaningful human intervention, does not meet it, and that merely entering prompts is not enough. The same study accepts that selecting, iterating, editing and combining AI output with human work can qualify. It is a study and not a law, so treat it as a strong indication of where the reasoning points, not a guarantee.

Two practical consequences follow for an EU business. A plain, untouched AI output probably has no copyright protection. A document where your team made real creative decisions probably does, and the protection covers the human expression, not the machine’s contribution.

The UK is different. Its Copyright, Designs and Patents Act includes a rule for computer-generated works, which treats the author as the person who made the arrangements necessary for the creation of the work. It has been in the statute since 1988 and gives protection for a shorter period than ordinary works. The government reviewed the whole area and published a report in March 2026. Commentators describe it as leaning towards removing the provision while gathering more evidence, so it remains in force for now but should not be relied on as a long-term foundation. If you trade in the UK, assume the rule may disappear and prefer human authorship for anything important.

For data protection, which is a separate legal question from copyright, see what is safe to paste and the EU AI Act and GDPR compared.

What about the United States?

Many AI tools are American and many businesses sell to American customers, so the US position is worth knowing. The US Copyright Office published a report on copyrightability in January 2025. It concluded that, with the technology generally available, prompts alone do not give a user enough control over the output to count as the author of it. It also said that AI can be used as a tool without losing protection, that human selection and arrangement of AI material can be protected, and that a person’s own work remains protected where it is still visible in the result.

In plain terms, the US and the EU are heading to the same place. Protection follows human creative input. The thinner that input, the weaker the claim.

Does AI-generated Content infringe someone else’s copyright?

It can. This is a separate risk from ownership and it is the one most likely to cause trouble.

Generative tools learn patterns from very large collections of existing work. Usually the output is new. Occasionally it is close enough to an existing work, a famous character, a real person’s face, a recognisable brand or a distinctive style to create a problem if you publish it. The tool does not check, and it will not warn you.

Risk varies by type of output:

• Ordinary business text. Emails, summaries, internal notes, first drafts. Low risk, although the output can hallucinate facts or quotations, which is a different problem.
• Marketing copy and articles. Low to moderate. Check for lifted phrases and invented sources before publishing.
• Images. Moderate to high. Avoid prompts that name a living artist or a fictional character, and be careful with anything that includes people.
• Logos and brand assets. High, because you want exclusive rights and the tool cannot give them.
• Code, lyrics and quoted text. Check licences and sources, because output can reproduce existing material.

Responsibility generally rests with whoever publishes. Some providers say they will defend business customers against infringement claims on certain paid plans, subject to conditions. Read those conditions before relying on them. They usually require that you used the product’s safety filters, did not knowingly infringe and made no changes designed to trigger a copy.

This is why checking AI output before it reaches a customer belongs in your routine. A person looking at the result for anything that seems copied is the cheapest protection available.

What do AI vendors’ terms say about Ownership?

Terms differ, which is why you should read the ones for the plan you actually use. As a general pattern, major providers say that, as between you and them, you own the output or they assign you their rights in it, and that they do not claim ownership. That is a contractual statement, not a legal fact about copyright. It tells you the provider will not object. It cannot create a copyright that the law does not recognise.

Check five things in the terms for each tool your team uses:

1. Does your plan allow commercial use of the output?
2. Does the provider claim any rights over it?
3. Is your input used to train the provider’s models, and can you switch that off?
4. Is there any indemnity for infringement claims, and what are its conditions?
5. Do the terms differ between personal and business plans?

The last point matters because personal accounts often carry different terms from business ones. This is part of why banning AI tools backfires: staff use personal accounts on the quiet, and then the business has no idea whose terms apply to the work. That is shadow AI, and it makes every ownership question harder.

Is it Safe to use AI Images in your marketing?

Usually yes, with care. Use the output where its originality does not matter and where nobody would be harmed by a lookalike. A background illustration or a generic office scene is fine. A hero image that defines your brand is not, because you cannot stop others using something very like it.

A few simple rules reduce nearly all of the risk:

• Do not use a living artist’s name or a named character in the request.
• Do not generate images of real people, including staff and customers, unless they have agreed in writing.
• Look closely at the result for stray lettering, logos and signatures, which are signs of copying.
• Keep a record of the tool, the plan, the date and the request.
• For logos, have a designer create or substantially rework the artwork, and register the result as a trade mark.

See the AI image tools compared for how the main options differ on terms and output quality.

Disclosure is a separate question. The EU AI Act includes transparency duties under its Article 50. Providers of systems that generate synthetic content must mark it in a machine-readable way. People who deploy a system to produce realistic synthetic images, audio or video, often called deepfakes, must disclose that the content is artificial. People who publish AI-generated text to inform the public on matters of public interest must say so, unless a person has reviewed it and takes editorial responsibility. The dates and details are set out in what the EU AI Act requires of a small business, and the European Commission’s regulatory framework page keeps the official timetable.

What should your AI Policy say about copyright?

Short and specific. A paragraph or two in your AI use policy is enough if it answers the questions staff will actually ask. A workable version covers these points:

• Approved tools. Name the tools and plans staff may use for work, so the terms are known.
• Where a person must decide. For anything you want to own, such as a logo, a core product document or a campaign concept, a named person must write or substantially rework it, and keep the drafts.
• What needs a check. Anything public, any image and anything quoted gets a human review before it goes out.
• Disclosure. State when you tell customers, readers or partners that AI was used, and who decides.
• Suppliers. Freelancers and outside suppliers declare whether they used AI and assign you the rights they hold.
• Escalation. If someone is unsure, they ask a named person, not the tool.

This fits into wider AI governance without needing a separate project. If you want an outside view of what your team already uses and where the gaps are, see what an AI audit checks.

Good inputs also reduce risk. When you brief an AI tool like a new hire, you supply your own material, your own structure and your own examples, which makes the output more clearly yours and less likely to resemble anything else. Making sure a person stays in the loop is also what turns tool output into a human work in the legal sense.

What to do this week

Start with an inventory. List the AI tools your team uses, which plans they are on, and what each one produces that you publish or sell. That alone usually shows a gap, such as a logo or a set of product images made on a personal account.

Then pick the three outputs that matter most to your business and decide for each who the human author is. Where the answer is “nobody”, assign a person to rework it and keep the drafts. Add the short paragraph above to your policy and send it to the team with a note on why it matters.

If you would like an independent look at which tools your staff use and what leaves the business, we cover that in our audits. Start with what an AI audit checks, or book an audit.

Sources

• US Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability: concludes that prompts alone do not make a user the author of AI output, while human selection, arrangement and modification of AI material can be protected.
• Copyright, Designs and Patents Act 1988, section 9: the UK rule that treats the author of a computer-generated work as the person who made the arrangements necessary for its creation.
• Law firm summary of the UK government’s March 2026 report on copyright and AI: reports that the government wants more evidence and is leaning towards removing the computer-generated works provision, which remains in force.
• 2025 EU legislature study, Generative AI and Copyright: argues that fully autonomous AI output does not meet the EU originality test and that prompts alone are not enough, while human selection and editing can qualify.
• EU AI Act, Article 50: sets the transparency duties for AI-generated and manipulated content, including deepfakes and AI-generated text published to inform the public.
• European Commission, AI regulatory framework: the official overview and timetable for the EU AI Act.