Can You Use AI-Generated Images in Marketing? Ownership, Copyright, Disclosure, and a Policy You Can Paste

The short answer is yes, and the reason the question keeps getting asked is that “yes” comes with three different fine prints that people run together: whether you are allowed to use the image (a contract question), whether you own it in the copyright sense (a law question that got clearer in 2026), and whether you have to say it was generated (a regulation question that changed on August 2, 2026). Most of the confusion in the “can I use AI images” articles comes from answering one of the three and implying the other two.

This post separates them, gives you a per-platform table as of September 2026, names the two traps that actually get marketers in trouble, and ends with a five-line policy you can paste into your team’s wiki. It is not legal advice; it is the map you bring to whoever gives you legal advice, so the conversation takes ten minutes instead of a quarter.

Three questions that get confused

1. May I use it? This is a permission the platform grants you by contract, in its terms of service, and it operates independently of copyright. A platform can let you use commercially something nobody owns, and it can forbid commercial use of something you technically own. Read the terms for your plan.

2. Do I own it, in the copyright sense? In the US, the answer for a purely AI-generated image is no, and 2026 settled it. On March 2, the Supreme Court declined to review Thaler v. Perlmutter, leaving in place the rule that copyright requires human authorship. AI-assisted works can be protected to the extent a human exercised meaningful creative control over the expressive elements; the Copyright Office evaluates case by case, and writing a prompt and accepting the first result is unlikely to qualify. The practical consequence for a marketer: you may use the image, and you generally cannot stop anyone else from using it too. No exclusivity.

3. Must I disclose it? This is regulation and platform policy, and it is where 2026 moved fastest. Covered below.

Keep the three apart and most of the “can I” articles collapse into a table.

The per-platform table (as of September 2026)

Four things to take from the table. Ownership language is nearly universal and nearly meaningless, because ownership without copyright is a permission, not a property. Indemnification is rare and it is the thing a risk-averse legal team actually wants; Firefly is the mainstream option and its coverage has explicit exclusions worth reading. Default visibility matters more than people expect: a Midjourney Basic or Standard user’s prompts and drafts are public. And watermarks and content credentials are becoming the disclosure mechanism, which connects to the next section.

The two traps that actually cause trouble

Almost no marketer will be sued over copyright in a generated image. The exposure is in what goes into it.

Likeness. Generating a recognizable real person (a celebrity, a public figure, a real customer, a competitor’s founder) raises right-of-publicity claims that no platform’s terms shield you from, and platforms’ own policies prohibit it. The subtler version is the “customer testimonial” image of a person who does not exist presented as one who does, which is a deception problem before it is a legal one; the trust guide covers why it costs more than it saves.

Trademark and trade dress. Never generate a competitor’s logo, packaging, or distinctive look, never generate a recognizable brand character, and never use another brand’s image as a reference. The reference-images workflow is powerful precisely because the model reproduces what it sees, which is why the references must be yours.

Avoid those two inputs and most of the practical risk is gone. What remains is the training-data litigation against the model providers, which is real, ongoing, and mostly not your exposure as a user; if your legal team disagrees, Firefly’s indemnification is the answer they are asking for.

Disclosure: what changed on August 2, 2026

The EU AI Act’s Article 50 transparency obligations took effect on August 2, 2026. For a marketer, the relevant part: anyone who publishes AI-generated or AI-manipulated images that realistically depict people, objects, places, or events in a way that could falsely appear authentic must disclose it in a clear and distinguishable manner, at the latest at first exposure. Providers must embed a machine-readable marker (SynthID and C2PA both qualify); deployers who publish must additionally make the disclosure perceptible to a human. Systems deployed before August 2 have until December 2, 2026 for the machine-marking part. Guidance from advertising lawyers reads the rule as expanding disclosure duties for advertisers and PR teams well beyond deepfakes in the tabloid sense.

Platforms moved earlier. Meta requires disclosure for political, electoral, and social-issue ads that use realistic AI imagery, and Meta, TikTok, and YouTube auto-detect and label content carrying C2PA credentials, which many generators now embed by default. In the US there is no federal labeling law for commercial images as of this writing, several states have narrower rules, and the FTC’s deception authority applies to any image that misleads about a material fact.

The practical reading: if the image realistically depicts something a viewer could believe is real, assume disclosure is required somewhere you publish, and label it. If the image is plainly illustrated concept art, it generally falls outside the deepfake framing, but the platforms may label it anyway via credentials, and that is fine. The rule that keeps you out of trouble is the same one that keeps you out of the trust penalty: never present a generated image as a real thing.

The five-line policy

Paste this into the team wiki, adjust the plan names, and have whoever advises you on legal matters spend ten minutes on it.

AI IMAGE POLICY (v1, September 2026)

1. We generate images only on platforms whose paid terms grant commercial use
   ([list: e.g. OpenAI, Adobe Firefly]). No Free-tier generation for published work.
2. We never generate real people, competitors' marks or trade dress, or any
   image presented as a photograph of our team, premises, customers, or events.
3. Our own approved images are the only reference images we use.
4. Realistic generated images are labeled wherever we publish them; illustrated
   concept images are labeled where a platform or market requires it.
5. Every published generated image is recorded with its model, prompt, references,
   and approver, so we can answer "where did this come from" a year later.

Line 5 is the one teams skip and the one that saves you. The day a question arrives about an image, the answer is either a record or a shrug.

Where provenance comes from

That fifth line is easy to write and hard to keep by hand: a folder of PNGs with random filenames records nothing. It is one of the reasons Marqeable generates images inside the piece they belong to and records, on each image, the model, the prompt, the brand-profile version, and the reference images that produced it, alongside who approved the piece. Generation runs on OpenAI’s GPT Image 2.5 models under OpenAI’s terms; your real photos are matched from the library before anything is generated, so the “team” and “truck” images are always the real ones; and the logo is composited after generation rather than drawn by the model. The policy above is the one we designed for. We are in private beta with a small early cohort: get early access if you would rather have line 5 kept for you.

Frequently asked questions

Can I trademark a logo made with AI?

Trademark protects a mark as a source identifier and does not require the copyright-style authorship test, so a generated logo can often be registered as a trademark if it is distinctive and in use. You still cannot stop someone from copying the artwork under copyright if it was purely AI-generated, and generated logos can resemble existing marks; have it searched.

Do I need Adobe Firefly for indemnification?

Only if your legal or procurement team requires a contractual indemnity. Most small teams accept the platform terms and manage risk by controlling inputs (the two traps above). Enterprises and agencies delivering to enterprises often do need it, and Firefly’s enterprise entitlement is the mainstream option.

Are ChatGPT images copyright-free?

They are yours to use under OpenAI’s terms, and they are not copyrightable in the US if purely generated, which together means “free to use, not exclusive.” Substantial human editing of the output can create protectable elements.

What about images I edit heavily after generation?

The more human creative control in the final work, the more of it may be protectable, evaluated case by case. Keep the layered file and a note of what you changed; that is the evidence of authorship if it ever matters.

The bottom line

You can use AI-generated images in marketing. You will not own them the way you own a photograph, you must read the terms of the plan you are on, and since August 2026 you must label the ones that could pass for real in the EU and on several platforms. None of that is the risk that matters. The risk is generating a real person or someone else’s mark, and presenting a made image as a real thing. Avoid those, keep a record of every image you publish, and the legal question becomes a ten-minute conversation instead of a reason not to start.

Marqeable runs your campaigns, answers every visitor, text, and email in seconds, and turns them into booked jobs and meetings - even at 9pm on a Saturday. We’re in private beta with a small early cohort. Get early access