Can Brands License Influencer Content for AI Training?

Two separate rights that get bundled together

When a brand licenses content from an influencer, it typically negotiates usage rights for something specific: can the video run in paid ads, on the website, for how long. That is an agreement about where and for how long the content can appear. Using the same footage to train an AI model is a third, separate right — and it is not automatically included just because the brand can show the content in its own ads.

The distinction matters in practice: a licence to display an image is not the same as a licence to let a system learn from that image's pixels, composition and style to generate new content. The two questions are governed by different rules, so they have to be agreed separately.

The EU default: AI training is lawful text-and-data mining unless the creator opts out

Legally, AI training falls under what EU law calls text-and-data mining (TDM). Article 4 of the DSM Directive (2019/790) permits TDM on copyright-protected material for any purpose — including commercial AI training — unless the rightsholder has expressly reserved that right in a machine-readable way (Article 4(3)). That makes it an "opt-out" system: content can be used for AI training by default until someone actively says no.

Denmark implemented the rule into its copyright act (ophavsretsloven) with effect from 1 June 2023: anyone with lawful access to a work may carry out text-and-data mining unless the rightsholder has reserved the right "in an appropriate manner" — and a reservation is generally considered appropriate if it is machine-readable, e.g. in metadata or in a website's or service's terms of use.

For an influencer, that means: unless the person (or whoever they have assigned rights to) has actively flagged their content as reserved from AI training, it is lawful by default for others to mine it — whether that's a third-party AI provider scraping publicly available content, or a brand that already has access to the material.

What counts as a "machine-readable" reservation in practice?

The most widely used tool is still robots.txt, where a website blocks named AI crawlers. A more precise, purpose-built alternative is TDMRep (the Text and Data Mining Reservation Protocol) — a W3C community specification that lets a website or platform declare, in machine-readable form, whether content may be used for data mining, and that is explicitly referenced as a valid reservation mechanism alongside the DSM Directive and the AI Act.

Whether a reservation written in plain natural language — e.g. inside a set of terms and conditions — also counts as "machine-readable" has been contested. Germany's Hanseatic Higher Regional Court ruled in Kneschke v. LAION (10 December 2025, case 5 U 104/24) — the first appellate-level EU ruling on the question — that a natural-language reservation did not meet the standard for the specific 2021 use in that case, while suggesting a more generous reading could apply going forward. The law here is still settling, and there is no unified EU-wide opt-out registry yet: the European Commission published only an initial feasibility study into whether one should even exist on 13 July 2026.

Denmark's own Ministry of Culture-convened expert group on copyright and AI has concluded that rightsholders struggle in practice to enforce reservations, and has recommended that Denmark push to change Article 4 to an "opt-in" model — requiring explicit consent for AI training instead of a reservation — if the practical problems are not resolved as part of the EU's planned review of the DSM Directive in 2026.

The AI Act adds transparency, not permission

The EU AI Act does not change copyright law itself; it layers a transparency requirement on top of it. Providers of general-purpose AI (GPAI) models have had to publish a public summary of their training data since 2 August 2025, using a template the European Commission released on 24 July 2025 — including what data sources were used and how the provider respected reservations made under the DSM Directive's text-and-data-mining exception. Enforcement of these duties took full effect on 2 August 2026, when the EU AI Office gained the power to check compliance.

For a brand weighing whether to let influencer content flow into AI training: if the material ends up inside a third-party AI model, that use will — in theory — show up in that provider's public summary. It does not change the underlying question of whether the use was lawful in the first place; that is still decided by copyright law and, where one exists, the contract.

Does a standard usage-rights clause cover AI training?

In short: no, not unless it says so explicitly. A usage-rights clause written for paid ads, the website and social channels typically does not address AI training at all — and a right that isn't mentioned isn't granted.

If the brand wants to train its own AI model on the content

Keep two tracks separate here. The Article 4 exception governs when someone without a direct agreement with the rightsholder may mine publicly available content. If a brand already has a direct contract with the influencer for the content, it's the contract's wording — not the law's default rule — that decides what the brand may do. A contract can grant fewer rights than the law would otherwise allow (by explicitly excluding AI use) or more (by explicitly permitting it), but it doesn't remove the need to say so directly.

If a brand wants to use influencer content to train an internal AI model — for example, a tool that generates product imagery in a particular influencer's style, or a model that mimics a specific creator's tone — that needs its own explicit clause. That applies whether the model stays internal to the brand or the brand wants to resell or sub-license access to an external AI vendor.

What the clause should actually say

• Whether AI training is permitted at all — yes or no, stated, not implied.
• Which kind of AI use — internal use, a public-facing AI model, or both.
• Whether sub-licensing to a third party is allowed — can the brand give an external AI provider access to the material?
• Whether AI-generated output based on the influencer can be used commercially — e.g. synthetic images of the creator's face or voice, which raise the same name-and-likeness questions as traditional marketing.
• Duration and revocation — can the permission be withdrawn, and what happens to a model already trained on the material?
• Compensation — AI-training rights are their own negotiable right and should be priced separately, not folded for free into a standard content deal.

See the full contract checklist in what to put in an influencer contract.

If a creator doesn't want their content used for AI training

An influencer who doesn't want their publicly available content mined for AI training has to take an active reservation — it doesn't happen automatically. In practice that means either a robots.txt block against named AI crawlers, a TDMRep declaration on their own site, or, where relevant, an explicit term in the platform's conditions where the content is hosted (e.g. a portfolio page). For content that only exists with a specific brand after a paid collaboration, it's simpler: the contract controls, and a creator can require that AI training is explicitly excluded.

Decision framework

FAQ

Is it illegal to train an AI model on influencer content without permission?

Not automatically. Under the EU's current opt-out model (DSM Directive Article 4), it's lawful by default unless the rightsholder has made a machine-readable reservation. That could change if the EU moves to an opt-in model, as Denmark's own expert group has proposed, during the DSM Directive's planned 2026 review.

Does GDPR also apply if the influencer's face is part of the training data?

Yes — if the material includes a person's face or voice, that's personal data, and GDPR's lawful-basis rules apply alongside copyright law. See influencer marketing and GDPR for the full picture.

Can an influencer charge for their content being used to train an AI model?

Yes, if it's agreed in the contract. AI-training rights aren't a statutory payment obligation, but a negotiable right like any other usage right — the price is set by the parties, not by the law.

Is this the same as the AI and virtual influencers article?

No. AI and virtual influencers vs human creators covers fully synthetic personas with no real person behind them. This article covers using a real influencer's existing content as training data for an AI model — a different, copyright-driven question.

What about cloning a creator's voice to dub content into another language?

That's a related but separate question — see our article on AI voice cloning and dubbing of creator content, which covers the EU AI Act's separate disclosure duty on top of the rights question this article covers.

Make Influence's perspective

Make Influence does not currently offer AI-training licensing as part of its product, and we generally advise brands to keep AI-training rights as a separate, explicit line item in any influencer contract rather than assume it's covered by the standard usage-rights clause. This area is moving fast — both in the EU's planned DSM Directive review and in case law like Kneschke v. LAION — so a clause that's precise today should be revisited as the rules shift. This article is general guidance, not legal advice; talk to an IP-specialised lawyer for a specific contract.