AI in marketing: what the law says

Important note first

In marketing, law often feels like a brake. With AI it’s more of a railing: most rules ask for honesty towards customers, care with data and respect for other people’s work. If you want that anyway, there’s little to fear.

EU AI Act: deadlines and duties

As of Nov 14, 2026, per the timeline overview and article texts.

What Art. 50 means for marketing

• Deepfakes (para. 4): as a deployer, if you generate or manipulate image, audio or video content that appears authentic, you must disclose that it’s artificial. For evidently artistic, satirical or fictional works, a notice that doesn’t spoil the work is enough.
• Texts on matters of public interest (para. 4): AI texts meant to inform the public must be disclosed, unless a person reviewed or edited them and a person or organisation holds editorial responsibility.
• Chatbots (para. 1): systems that talk directly to people must be designed so users know they’re talking to AI, unless it’s obvious. This duty is the provider’s; check that your chatbot vendor implements it.
• Machine-readable marking (para. 2): a duty for providers of image, audio, video and text generators, not for you as a user.

For typical marketing content: an AI video with realistic-looking people gets a label. A blog post you reviewed, with clear responsibility, doesn’t need an AI label under Art. 50. Platforms like Meta add their own rules; more in AI in advertising and AI for social media.

AI literacy under Art. 4

Per the European Commission, no specific certificate is needed; internal records of training are enough. The duty also covers service providers using AI on your behalf. How to put it into practice is in AI marketing strategy.

GDPR: personal data in AI tools

• Data minimisation: put only the data a task needs into an AI tool. An anonymised extract is often enough.
• Check settings: whether inputs are used for training is in the vendor’s settings and terms.
• Marketing profiles: lead scoring and personalisation are usually not Art. 22 decisions as long as a person decides in the end and nobody is excluded from an offer by the score alone.

More on processor agreements, US vendors and self-hosting in marketing automation with AI, on chatbots in AI chatbot for your website.

Competition law: misleading claims, labelling, marketing email

• Misleading claims (§ 5 UWG): untrue or deceptive statements about a product, price or features are unlawful, also if an AI phrased them. Made-up testimonials, test results or top ratings are off-limits.
• Commercial purpose (§ 5a(4) UWG): it’s unfair not to disclose the commercial purpose of a business act. If someone acts for another company, consideration is presumed until the opposite is credibly shown.
• Marketing email (§ 7 UWG): email advertising without prior express consent is an unacceptable nuisance, with a narrow exception for existing customers. That includes AI-personalised cold emails, also to businesses.

Add § 6(1) no. 1 of the German Digital Services Act (DDG): commercial communication in digital services must be clearly recognisable as such. For affiliate links that means “Ad” or “Advertising” next to the link.

More: AI cold outreach on § 7, AI copywriting on misleading claims and copyright.

Copyright: AI output and others’ works

Do you own what AI creates?

Under § 2(2) of the German Copyright Act, works are only “personal intellectual creations”. What an AI produces without a defining human contribution is therefore generally not protected by copyright. For you: others can often reuse a purely AI-generated image or text, and you can hardly stop them. The more of your own design you add (selection, editing, combination), the more likely protection arises for your share.

May you use what AI outputs?

If an AI reproduces protected works in whole or in part, such as lyrics, logos or well-known characters, using them can infringe others’ rights, regardless of who wrote the prompt. So check outputs that resemble known works, and don’t use prompts that deliberately recreate others’ works.

The key rulings

As of Nov 14, 2026. Neither case is concluded; the legal position may change.

For marketing teams the practical consequence is clear: the risk lies less in training, which vendors are responsible for, and more in using outputs that reproduce protected works. More in AI images for marketing.

Personality rights: faces and voices

• Images (§ 22 KUG): likenesses may only be distributed with the depicted person’s consent, with statutory exceptions. That includes AI images in which a real person is recognisable.
• Voices: on Aug 20, 2025 (2 O 202/24), the Berlin Regional Court II ruled that an AI-generated voice closely resembling a well-known voice actor’s infringes his general personality right without consent. Decisive was that the audience could assume he had spoken the text. A notional licence fee was awarded.
• AI faces: don’t have AI create people who resemble real, well-known individuals. Generic, recognisably invented people are the safer pattern.

For AI avatars and UGC ads, also: an AI actor must not be presented as a real customer. More in AI avatars and UGC ads and AI video.

Labelling ads and influencers

On Sep 9, 2021, Germany’s Federal Court of Justice set the line for influencer posts in three cases (I ZR 90/20, I ZR 125/20, I ZR 126/20):

Since then, § 5a(4) UWG presumes consideration when someone acts for another company. For AI marketing the principle is unchanged: advertising is labelled as advertising, whether a person or an AI wrote it.

• Affiliate links: “Ad” or “Advertising” right at the link.
• Creator partnerships: set out labelling in the contract.
• AI influencers: label ads as for people; disclose realistic AI people additionally under Art. 50.

Platform rules come on top

Besides the law, platform terms apply: Meta, Google, TikTok and LinkedIn have their own rules for AI content, political ads and labelling. They’re often stricter than the law and change more often.

Details on ad platforms in AI in advertising, on agents in AI agents for marketing and marketing to AI agents.

How to check a campaign in 15 minutes

Before any campaign with AI content, a short, fixed pass is worth it. It doesn’t replace a legal review, but it catches the most common mistakes.

1. Claims: every number, comparison and superlative has a source you opened today.
2. People: real people only with consent; AI people resemble nobody well known.
3. Voices: no voice modelled on a real person without their consent.
4. Works: no logos, characters, lyrics or visual styles recognisably taken from others.
5. Labelling: advertising recognisable, affiliate links marked, realistic AI content disclosed.
6. Data: audiences and lists come from consent or lawful sources; tools have processor agreements.
7. Platform: the platform’s ad policies for AI content have been read.
8. Sign-off: a person saw everything before it went live, and that’s recorded.

If a point stays unclear, that’s exactly the question to ask a professional. With this list, the conversation is short.

Checklist: legally clean with AI

• Team trained in AI literacy, attendance documented
• List of approved tools with processor agreements
• No personal data in tools without an agreement
• Privacy policy names the AI services used and where data goes
• Realistic AI images, videos and voices labelled
• No AI people resembling real individuals; consent for real people
• All product claims checked, no made-up reviews
• Ads and affiliate links labelled
• Marketing email only with consent (double opt-in)
• AI outputs resembling known works not used
• Platform terms read before switching on AI automation

12 prompts for a legal pre-check

These prompts help with preparation and self-checks. They don’t replace legal advice, and language models regularly get legal questions wrong.

Check this ad copy for claims I must be able to prove (numbers, superlatives, test results, comparisons). List them: [text].

Could this text mislead customers about the product’s features, price or origin? Name critical passages and more neutral wording: [text].

This post contains [affiliate link/partnership/own product]. How do I label it clearly and briefly for [platform]?

We’re planning this AI video: [description]. Could viewers take it for real? What notice would be appropriate?

We use these AI tools with this data: [list]. Which personal data flows where, and where do I probably need a processor agreement?

Draft a privacy policy section on [AI service, purpose, data], for review by a professional.

Draft a clear consent text for a newsletter with product recommendations and open tracking, for review by a professional.

Describe whether this AI image resembles known brands, logos, characters or people, and what I should change: [description or image].

Draft a one-page internal policy for AI in marketing: allowed tools, data classes, sign-off, labelling, handling mistakes.

Create a short template to document AI literacy training: date, attendees, content, tools.

List the points a contract with a creator should cover on AI use, labelling, usage rights and voice/likeness, as a basis for a conversation with a lawyer.

I’m planning [campaign]. Write five precise questions for a lawyer about the legal risks, so the conversation is short and focused.

The most common mistakes

1. “The AI wrote it” as an excuse. Whoever publishes is responsible.
2. Made-up testimonials. Misleading and unlawful, also with an AI avatar.
3. Customer data in every tool. Without a processor agreement, a GDPR problem.
4. Real people’s voices and faces. Without consent, a personality-rights violation.
5. Unlabelled ads. Applies to AI influencers as to people.
6. Asking only AI about the law. Fine for a pre-check, not for decisions.