The Ownership Gap: Your AI-Made Marketing May Not Belong to You

A competitor lifts your best-performing ebook. They don't paraphrase it. They change the logo, swap the color palette, and publish the same 4,000 words and the same diagrams under their own brand. Your team drafts the takedown letter, and then legal asks a question nobody in marketing can answer: how much of that ebook did a person actually write?

If the honest answer is "the outline, some edits, and the prompts," you may have very little to enforce.

This isn't a hypothetical edge case anymore. According to the Content Marketing Institute's B2B Content and Marketing Trends report for 2026, 95% of B2B marketers say their organizations use AI-powered applications, and almost nine in ten of them use AI to produce written content. Half use it to create or edit images, video, and other visual assets. The World Federation of Advertisers' State of AI in Marketing 2026 study, published on September 30, found the same thing among the world's largest advertisers: 96% now use generative or agentic AI, up from 45% in 2023, and 80% use it for content creation.

Production moved fast. The legal footing under that production did not move with it.

For CMOs, Content Leaders, Brand and Creative Teams, and the Legal Partners Who Sign Off on Their Work: this piece covers what U.S. copyright law now says about AI-generated marketing assets, where the real exposure sits (it's on both the output side and the input side), and how to decide which assets need a human author on the record and which ones don't.

A quick note before going further. I'm a marketer, not a lawyer, and nothing here is legal advice. The point is to give you enough of the picture to ask your counsel the right questions before a dispute forces them.

The rule got settled in 2026

For a couple of years the question of whether AI output could be copyrighted felt open. It's much less open now.

In January 2025, the U.S. Copyright Office published Part 2 of its report on copyright and artificial intelligence, which dealt specifically with copyrightability. Its central finding is blunt: prompts alone don't make you the author of what the model produces, no matter how detailed they are. The Office's reasoning is that the user doesn't control how the system turns instructions into fixed expression. The model decides the expressive elements. So purely AI-generated material, without enough human control over those elements, isn't protected.

Then the courts closed the door from the other side. The D.C. Circuit ruled in Thaler v. Perlmutter in March 2025 that the Copyright Act requires a human author. Stephen Thaler asked the Supreme Court to take the case, and on March 2, 2026, the Court declined. That leaves the human authorship requirement in place across every U.S. jurisdiction.

Thaler was the easy case, though, since he argued his AI system was the sole author. The harder question for marketing teams is the one in Allen v. Perlmutter, still pending in federal court in Colorado. Jason Allen created his prize-winning image "Théâtre D'opéra Spatial" by entering prompts into Midjourney at least 624 times, then cleaning it up in Photoshop and upscaling it. The Copyright Office refused registration anyway, arguing that time, effort, and iteration are not substitutes for human creation. Summary judgment briefing finished earlier this year, and as of August there was no ruling.

Think about what 624 prompts means for your team. That is far more iteration than any content marketer puts into a blog header image or a webinar promo graphic. If that level of effort may not be enough, a single prompt and a light edit almost certainly isn't.

What you can still own

The picture isn't all bad, and the Copyright Office report is fairly clear about where protection survives.

Human authors keep copyright in their own expression when it's perceptible in an AI-assisted work. They can also own the creative selection, coordination, or arrangement of material, and creative modifications they make to AI output. The Office says it has registered hundreds of works that incorporate AI-generated material since 2023, with the registration covering the human contribution.

The best-known example is "A Single Piece of American Cheese," an image made with Invoke AI using a technique called inpainting, where the creator regenerated small regions of the image repeatedly and chose what to keep. The Office first rejected it in September 2024, then registered it on January 30, 2025, because the creator showed enough human authorship in how he selected, coordinated, and arranged the AI-generated fragments.

There's a catch for anyone hoping to register an asset. Under the Office's registration guidance, if a work contains more than a de minimis amount of AI-generated material, you have to disclose that and explicitly exclude the AI-generated portions from your claim. You get protection for what the human did. You don't get it for the rest.

In practice, your copyright in an AI-assisted asset covers your edits, your structure, your original writing, and your arrangement. It does not cover the paragraphs or images the model produced on its own. A competitor who copies only the AI-generated parts may be copying material nobody owns.

Why B2B teams should care more than they do

My honest first reaction to all of this was a shrug. Most B2B content is a blog post that has a shelf life of about six months. Nobody is going to steal your "7 signs your CRM needs a cleanup" listicle, and if they did, the traffic would go to whoever ranks better, not whoever owns the copyright.

That's a fair point for a lot of content. It falls apart for a smaller set of assets that carry real commercial weight:

The proprietary frameworks your sales team uses in every deck. The maturity model that anchors your category story. The original research report that took a quarter to produce. The product illustrations and diagram system that make your brand recognizable. The certification curriculum in your customer academy. The messaging house your agency built for the rebrand.

These are the assets competitors copy, partners repackage, and former employees take with them. They're also the assets most likely to have been accelerated with AI in the past 18 months, because they're big and expensive to produce by hand.

There's also the quality question sitting underneath the legal one. In the CMI research, 12% of marketers said the quality of their content went down after they started using AI, and a fifth said AI wasn't improving creativity or quality at all. The assets that are cheapest to produce with AI tend to be the ones that are both hardest to protect and least differentiated. That's a bad combination for anything you want to build a brand on.

The input side is where the money moved

Ownership of outputs is half the problem. The other half is whether the tools themselves were built in ways that create liability for the people using them.

The scale of that question became concrete this summer. On July 20, 2026, a federal judge in the Northern District of California gave final approval to the $1.5 billion settlement in Bartz v. Anthropic, covering roughly 482,000 books that the company had downloaded from pirate sites. It's the largest copyright class settlement on record. The settlement dealt with how training data was acquired, not with what end users did, but it put a price tag on input-side risk that any legal team reviewing AI tools will have noticed.

Courts haven't moved in one direction. In the U.K., the High Court ruled against Getty Images on its secondary copyright claim against Stability AI in November 2025, after Getty dropped its primary copyright claims during trial. Getty won only a narrow trademark point. In the U.S., Disney and Universal sued Midjourney in June 2025, alleging the model reproduces their characters on demand, and Warner Bros. Discovery joined that September. Summary judgment motions in that case are due November 23, 2026.

For a B2B marketer, the Midjourney case is the one to watch. It's about outputs that resemble protected work closely enough to infringe. If your designer asks an image model for "a friendly robot mascot in the style of" a well-known animation studio, owning the result is the smaller worry. Someone else may already own what it looks like.

Read the indemnity before you rely on it

The major AI vendors saw this coming and offered indemnities. Microsoft has its Customer Copyright Commitment for Copilot. Google offers generative AI indemnification for its paid services. Adobe sells contractual IP indemnification for select Firefly outputs through certain enterprise plans. OpenAI announced Copyright Shield for business customers.

These are real protections, and they're conditional in ways most marketing teams never check. Microsoft's commitment depends on customers not trying to get around the guardrails and content filters it builds in. Google's coverage applies to generally available paid services, excludes previews and free versions, and doesn't apply if you disabled guardrails. Adobe's protection covers certain outputs on certain plans, not everything anyone on your team ever generated.

The shadow tool problem makes this worse. If your content team uses a free image generator a designer found on their own, or a consumer account instead of the enterprise one, none of those vendor commitments apply. The indemnity you heard about in the procurement meeting covers the tool procurement bought, used the way the contract describes.

Before you treat vendor indemnity as your safety net, ask three things. Which tools is the team actually using, including the ones nobody approved? Are we on the plan tier that carries the indemnity? And are we using the tool in a way that keeps us inside its conditions?

The agency question nobody asked

A lot of B2B content isn't made in-house. It comes from agencies, freelancers, and production shops. That's where the ownership gap tends to be widest, because the old contract language assumed a human made the work.

A standard work-for-hire or assignment clause transfers whatever copyright exists. If the agency's deliverable was mostly AI-generated, there may be little copyright to transfer, and the assignment clause hands you an empty box.

Large advertisers have started to fix this, slowly. In 2024 the WFA found that only 36% of major brands had introduced terms prescribing how their agency partners could use generative AI, and just 29% had reviewed their creative and media contracts to add AI clauses. In the 2026 study, 59% have introduced AI-related contract clauses covering issues such as IP, confidentiality, and training data, and 69% require agencies to disclose when AI is used. But only 13% say they have a great deal of visibility into how agencies use AI.

In June 2025, the Association of National Advertisers and Venable LLP released an AI rider template for advertiser and agency contracts. One provision gives the client copyright in AI-generated work made on its behalf, with the obvious exception for work that can't be protected in the first place. That exception is the whole story. A contract can allocate ownership. It cannot create copyright that the law says doesn't exist.

What a contract can do is require disclosure of which tools were used and where, require that hero assets carry meaningful human authorship, shift indemnity for infringing outputs to the agency, and forbid feeding your confidential material into tools that train on customer data. Those are the levers worth pulling. If your agency agreements are more than two years old, they probably don't address any of them.

A tiering model for AI-assisted assets

You don't need to stop using AI. The 96% figure tells you that ship has sailed, and it should have. What you need is a way to decide where human authorship matters enough to require it and document it.

I'd sort marketing assets into three tiers.

Tier one is disposable content. Social posts, routine blog articles, ad variants, email subject line tests, event recap posts. The value is in timeliness and distribution, and it decays fast. If someone copies it, you lose very little. Use AI freely here, keep a human editor for accuracy and brand voice, and don't spend legal time on ownership.

Tier two is durable content. Pillar pages, ebooks, webinar decks, case studies, comparison guides. These earn traffic and pipeline for a year or more, and a copied version could pull real search traffic or confuse buyers. Use AI for research, outlines, and first drafts, but make sure a person writes or substantially rewrites the final text and builds the structure. Keep the drafts.

Tier three is crown-jewel content. Proprietary frameworks, original research reports, brand illustration systems, certification curricula, category narratives, anything that might end up in a dispute or a licensing deal. These need a named human author, documented creative decisions, and a version history that shows what the person did. AI can help with research and analysis, but the expression that matters should come from someone on your payroll or under a contract that covers it.

The documentation part is what most teams skip. If you ever need to register or enforce a tier-three asset, you'll have to show the human contribution. Version history in your docs tool, design files with layers, and a short note on who made which creative calls cost almost nothing to keep and are very hard to recreate later.

Use the right kind of protection for brand assets

Copyright isn't the only tool, and for some marketing assets it isn't the best one.

Your brand name, logo, tagline, and product names function as source identifiers, and trademark law protects them based on use in commerce rather than on who or what drew them. That makes trademark a more reliable path for brand marks than copyright when AI was involved in the design. It isn't automatic, since a mark still needs to be distinctive and actually used, so talk to counsel before your next rebrand rather than after.

For research and data assets, the protection was always thin. Facts aren't copyrightable no matter who finds them. Original research earned its moat by being first, getting cited, and tying a methodology to your brand, and the copyright on the numbers had little to do with it. That kind of advantage survives AI just fine, and it's arguably stronger now, because so much competing content is generic. The protectable part is your written analysis, which is one more reason to have a person write it.

The governance most teams are missing

The WFA's 2026 study shows big brands building practical controls: 75% require human oversight of AI-generated outputs, 70% maintain approved or prohibited tool lists, and 57% require review and approval of AI outputs. Governance and legal requirements were named as a top barrier to scaling AI by 43% of brands.

In the mid-market B2B companies I talk to, almost none of this is written down. Usually there's an AI acceptable use policy from IT that covers data security and says nothing about who owns the output.

A workable version fits on one page. List the approved tools and the plan tier for each, and note which ones carry indemnity. Define your three content tiers with examples from your own library. State the human authorship and documentation expectations for tiers two and three. Ban prompts that reference named artists, studios, or competitor brands. Require agencies and freelancers to disclose AI use per deliverable. Name one owner in marketing ops who keeps the list current and who legal can call.

Then audit backward. Pull your top 20 assets by pipeline influence or brand importance, and for each one find out how it was made. You'll probably discover a few where nobody can say. Those are the ones to rework or document now, while the people who made them are still around.

What this means for the next 12 months

Several things are still unresolved. The Allen decision could clarify how much iteration counts as authorship, though it may not help much given how far apart the two sides are. The Midjourney case could set the first major U.S. marker on output infringement. Courts outside the U.S. are working through their own answers, and they don't all agree.

None of that uncertainty changes the core of the problem. The legal rule for outputs is now settled at the level that matters: no human author, no copyright. Most B2B marketing teams have spent two years scaling content production on tools that, used casually, produce assets they may not own and can't enforce.

For most of your content, that's fine. Speed was the point, and nobody's coming for your Tuesday blog post. For the small set of assets that carry your brand, your category story, and your competitive position, it isn't fine. You don't need to slow down everywhere. You need to know which assets are which, put a human author on the ones that count, keep proof that you did, and make sure your contracts and vendor terms match how your team actually works.

The ebook a competitor copies next year was probably drafted this quarter. Decide now how much of it you want to be able to call yours.