mail@crispycontent.de
Last updated on August 17, 2026 at 06:53 AM.
AI labelling refers to a visible and machine-readable notice indicating that content was generated wholly or partly by an AI system. From 2 August 2026, Art. 50 of the EU AI Act obliges every deployer to disclose AI involvement – violations carry fines of up to €15 million or 3 % of global annual turnover. This article provides channel-specific copy templates for websites, LinkedIn, Instagram, newsletters, video and AI avatars that can be adapted directly. Each template specifies when the wording applies and when an exception exists.
What does the AI labelling obligation under the EU AI Act mean?
Art. 50 of the AI Act defines transparency obligations for two addressees: providers (those who develop and place an AI system on the market) and deployers (any natural or legal person that uses an AI system under its own authority in a professional capacity). The obligation applies EU-wide from 2 August 2026 as a regulation – no national transposition is required. Anyone using an AI tool for text, images or chatbots today will be a deployer within the meaning of the law from that date.
Shadow AI is rarely announced. It arrives quietly, one prompt at a time, until sensitive data sits in tools no one has vetted. The question is not whether teams use AI, but whether anyone can vouch for how they use it. Crispy Content puts AI usage on GDPR-compliant ground through audits, policies, and secure integrations, which is less about restricting people than about making a promise you can actually keep to your customers and regulators.
The four key terms at a glance: A deepfake is AI-generated or AI-manipulated image, audio or video that convincingly resembles real persons, objects or events. Editorial control means substantive review and assumption of responsibility by a human – it can eliminate the labelling obligation for text. Machine-readable metadata are technical markers (IPTC, C2PA, EXIF) that make AI provenance detectable by systems without requiring human visibility. And the number that makes this urgent: 67 % of German employees already use AI tools regularly – the labelling question affects virtually every marketing team.
| Content type | Labelling required | Exception |
|---|---|---|
| AI-generated image (photorealistic) | Yes – deepfake under Art. 50(4) | Obviously unrealistic (comic, collage) |
| AI-generated text (public-interest relevance) | Yes – Art. 50(4) subpara. 2 | Human editorial final review |
| Chatbot / AI avatar | Yes – Art. 50(1) | Obviously recognisable as a bot |
Which wording meets the legal requirements?
The AI Act requires "clear and distinguishable" disclosure to the persons concerned (Art. 50(5)) but deliberately leaves room for choice of wording, placement and format. The decisive factor is whether the label is comprehensible to the specific target audience – a hashtag may suffice for social-media-savvy professionals, while a consumer portal needs a fully formulated sentence. The Wettbewerbszentrale recommends German-language labels for the German market, drawing on Higher Regional Court case law on influencer advertising: what applies to "Anzeige" (advertisement) can be transferred to "KI-generiert" (AI-generated). Art. 50(5) sentence 2 additionally requires that the label be accessible without barriers.
Core principles for AI disclosure wording
- Place before the content: The notice appears above or at the beginning of the AI-generated element – not in the footer, not in the legal notice, not after scrolling.
- Self-explanatory: No technical English for B2C audiences. "AI-generated content" is not "clear and distinguishable" within the meaning of the regulation for a German-speaking consumer audience.
- Supplement with machine-readable data: Alongside the visible notice, technical metadata (IPTC, EXIF, C2PA manifest or XML markup) belong in the file or source code – they enable automated detection by platforms and supervisory authorities.
Good to know: A hashtag such as #AIgenerated may suffice for social-media-savvy audiences. For less tech-savvy recipients, the Wettbewerbszentrale recommends more prominent, standardised notices – for example a visually distinct label placed directly above the content.
AI labelling for websites – copy templates for corporate pages
On websites, companies face the task of correctly labelling AI-generated images, text and chatbots simultaneously. Placement must occur before the content and be visually perceptible – a greyed-out notice in 8-point type inside an alt attribute does not meet the requirement. Those using a CMS can set up the label as a template component and manage it editorially.
Template for AI-generated images on the website
The standard-case template: "This image was created with the assistance of an AI system." Where the editorial team has subsequently edited the image, the variant works well: "This image was generated with AI support and reviewed by our editorial team." Both formulations meet the comprehensibility criterion because they name the AI contribution without presupposing technical knowledge. In addition, a C2PA manifest or an IPTC field "digitSourceType: trainedAlgorithmicMedia" should be embedded in the image file.
Template for AI-generated text on the website
For partially AI-assisted text: "This text was edited with the help of AI." For fully AI-generated content without editorial final review: "AI-generated text" – short, unambiguous, placed as a label above the paragraph. Once a human reviews the text substantively, edits it and assumes responsibility, the obligation may no longer apply. Voluntary labelling remains worthwhile nonetheless because it builds trust.
Chatbot labelling
The template for the start of interaction: "You are chatting with an AI assistant. For personal advice, we are happy to connect you with our team." The notice "Supported by AI" is insufficient – Art. 50(1) requires that users learn they are interacting with an AI system, not merely that AI is involved somewhere. The label must appear before the chatbot's first response.
| Measure | One-off effort (hrs) | Ongoing effort/month |
|---|---|---|
| Audit of existing AI content | 8–16 | – |
| Template integration (CMS) | 4–8 | 1–2 |
| Set up metadata workflow | 4–6 | 0.5–1 |
AI disclosure for LinkedIn and Instagram – social media templates
Social media platforms offer their own AI labels – LinkedIn since 2024, Meta with the tag "AI generated". These labels do not replace the deployer's obligation under Art. 50 of the AI Act. They are a technical tool provided by the platform, not a legally compliant disclosure by the deployer. Companies remain responsible for clear, audience-appropriate disclosure within their own posts.
LinkedIn copy templates
For the post opening: "[AI-assisted] This post was written with the support of generative AI and reviewed by our editorial team." The square brackets create visual prominence in the feed. For image captions: "Visual: AI-generated (Midjourney) | Editorial responsibility: [Company]" – naming the tool is voluntary but increases credibility. Additionally, activate the platform's own label so that algorithmic detection also takes effect.
Instagram copy templates
For the caption: "📌 AI-generated image. Concept & editorial review: our team." For story overlays: "Created with AI – reviewed by [name/department]" – the personalisation signals editorial responsibility. Instagram offers a platform-native "AI generated" label that can be activated via advanced settings. Both measures – your own copy template and the platform label – belong together.
| Platform | Native AI label available? | Does it replace the Art. 50 obligation? | Recommendation |
|---|---|---|---|
| Yes (since 2024) | No – use as supplement | Own copy template + platform label | |
| Instagram/Meta | Yes ("AI generated") | No – use as supplement | Caption notice + activate Meta label |
| YouTube | Yes (disclosure checkbox) | No – use as supplement | On-screen notice in video + checkbox |
AI disclosure for newsletters – transparency in email marketing
Newsletters fall under the labelling obligation when they contain AI-generated text or images that carry public-interest relevance – i.e. are capable of influencing opinion formation. Where editorial final review has been applied, the obligation may not apply. A voluntary transparency notice is still worthwhile: the Edelman Trust Barometer 2026 reveals a growing trust crisis in which proactive disclosure becomes a differentiator. Those who label voluntarily invest in brand trust – measurable through open rates and unsubscribe rates over time.
Newsletter copy templates
- Header notice: "Parts of this newsletter were created with AI assistance and reviewed by our editorial team."
- Footer disclaimer: "Transparency notice: Generative AI was used to produce individual sections. Editorial responsibility lies with [Company]."
- Fully AI-generated: "This newsletter article was produced by an AI system."
The header notice is the stronger option because it appears before the content is consumed. The footer disclaimer is suitable for newsletters in which only individual elements are AI-assisted and overall editorial responsibility rests with the sender.
AI disclosure for video and AI avatars – labelling in moving image
Videos containing AI-generated elements – synthetic voices, AI avatars, manipulated scenes – are subject to the labelling obligation under Art. 50(4) of the AI Act. AI avatars that interact in a human-like manner must additionally be disclosed as an AI system under Art. 50(1), specifically at the very start of the interaction. This dual obligation (deepfake labelling plus AI system disclosure) makes avatars the most complex use case.
Video copy templates
- On-screen notice (intro): "This video contains AI-generated elements."
- Lower third: "Voice: AI-generated | Presenter avatar: synthetic"
- Video description: "Notice: Parts of this video were created with generative AI (image generation: [tool], voice: [tool])."
The intro notice must be visible for at least 3 seconds and must not be obscured by other elements. The lower third is suitable for longer videos in which AI elements appear only during the course of playback.
AI avatar labelling
The template for the start of interaction: "You are speaking with an AI avatar. It simulates a human interaction but is entirely controlled by an AI system." The exception: obviously non-human avatars – such as a robot design or an abstract figure – do not require labelling because there is no risk of confusion with a real person.
Templates are not a universal remedy – the use case decides
The copy templates in this article are starting points, not copy-paste solutions. The Wettbewerbszentrale emphasises: whether content requires labelling depends on the individual case – in particular on the target audience, the potential for deception and the context of publication. A photorealistic AI image on a news site carries a different deception potential than the same image in an internal training document.
The factors for selecting the right wording:
- Tech-savviness of the audience: A B2B specialist audience understands "generative AI" without explanation; B2C consumers need "created with artificial intelligence".
- Content lifespan: A social media post disappears in the feed; a landing page remains live for years. The latter requires more robust labelling.
- Platform conventions: What works as a label on LinkedIn may look out of place on a corporate website.
- Industry standards: Pharma, financial services and media are subject to additional transparency requirements beyond the AI Act.
A worked example: A B2B company with 4 social media channels, a weekly newsletter and a website chatbot needs at least 6 different labelling templates – each adapted to format, audience and platform.
A documented AI disclosure strategy eliminates approval loops and protects against fines. Those who prefer not to build the template library in-house can develop it with a specialist content marketing agency such as Crispy Content®.
Trends – where is AI labelling heading?
The European Commission published a draft guideline on Art. 50 on 8 May 2026; a binding code of practice is expected by the end of 2026. Platforms are simultaneously implementing their own technical standards – C2PA and Content Credentials are establishing themselves as infrastructure. Regulation is not yet complete, but the direction is clear: more automation, more standardisation, more enforcement.
- Automated labelling by AI tools: Watermarks and metadata are embedded directly at the point of generation. Deployers must verify whether automatic labelling meets legal requirements – it does not replace visible disclosure.
- Proactive transparency as a trust signal: Consumers expect disclosure even below the mandatory threshold. Those who label voluntarily position themselves as trustworthy – the Edelman Trust Barometer 2026 confirms the link between transparency and brand loyalty.
- National enforcement from 2027: In Germany, the Bundesnetzagentur (BNetzA) is expected to serve as the market surveillance authority and will specify enforcement practice. First fine proceedings are anticipated for 2027.
| Date | Milestone |
|---|---|
| 01.08.2024 | EU AI Act enters into force |
| 02.02.2025 | AI literacy obligations apply |
| 08.05.2026 | Draft guideline on Art. 50 published |
| 02.08.2026 | Labelling obligation under Art. 50 becomes enforceable |
| End of 2026 | Binding code of practice expected |
AI transparency as brand equity
Companies that treat AI labelling not as a compliance chore but as a trust investment position themselves favourably with audiences and regulators alike. The copy templates in this article form the foundation of a labelling template library differentiated by channel, content type and audience. Methods deliver guarantees – here too. Those who systematise labelling before the deadline arrives save not only fines but gain something that cannot be bought after the fact: credibility.
| Criterion | Mandatory (Art. 50 AI Act) | Voluntary (best practice) |
|---|---|---|
| Applies to | Deepfakes, public-interest AI text, chatbots | All AI-assisted content |
| Penalty for non-compliance | Up to €15 million / 3 % of turnover | None – but loss of trust |
| Recommendation | Mandatory from 02.08.2026 | Implement immediately |
| Metric | Value | Source |
|---|---|---|
| Employees in Germany who regularly use AI | 67 % | BCG AI at Work 2025 |
| Employees worldwide | 72 % | BCG AI at Work 2025 |
| Max. fine for Art. 50 violation | €15 million or 3 % of turnover | EU AI Act Art. 99 |
Sources
Wettbewerbszentrale e.V. (2026): Kennzeichnung KI-generierter Inhalte – Leitfaden. URL: https://www.wettbewerbszentrale.de/wp-content/uploads/2026/02/2026_2_Leitfaden_KI_generierte_inhalte_1-1.pdf (accessed 20.07.2026).
Ecovis (2025): EU AI Act: Kennzeichnungspflicht für KI-generierte Inhalte ab August 2026. URL: https://de.ecovis.com/unternehmensberatung/eu-ai-act-kennzeichnungspflicht-unternehmen/ (accessed 20.07.2026).
European Commission (2026): Code of Practice on Transparency of AI-Generated Content. URL: https://digital-strategy.ec.europa.eu/de/policies/code-practice-ai-generated-content (accessed 20.07.2026).
Boston Consulting Group (2025): AI at Work 2025: Momentum Builds, but Gaps Remain. URL: https://www.bcg.com/publications/2025/ai-at-work-momentum-builds-but-gaps-remain (accessed 20.07.2026).
Future of Life Institute (2026): The EU AI Act's Transparency Rules (Article 50). URL: https://artificialintelligenceact.eu/de/transparency-rules-article-50/ (accessed 20.07.2026).
Edelman (2026): 2026 Edelman Trust Barometer. URL: https://www.edelman.com/trust/2026/trust-barometer (accessed 20.07.2026).
European Parliament / Council of the EU (2024): Regulation (EU) 2024/1689 (AI Act). URL: https://eur-lex.europa.eu/legal-content/DE/TXT/?uri=CELEX:32024R1689 (accessed 20.07.2026).
Gerrit Grunert
Gerrit Grunert is the founder and CEO of Crispy Content®. In 2019, he published his book "Methodical Content Marketing" published by Springer Gabler, as well as the series of online courses "Making Content." In his free time, Gerrit is a passionate guitar collector, likes reading books by Stefan Zweig, and listening to music from the day before yesterday.
.png)
.jpg)
-1.jpg)
-1.jpg)
.jpg)
.jpg)
.jpg)
