If your Swiss or Italian SME uses AI in any part of its marketing — and most of you do, even if you don’t think of it that way — the EU AI Act has a deadline you need to know about: 2 August 2026.
That’s when Article 50 of Regulation (EU) 2024/1689 takes effect. It introduces transparency obligations for providers and deployers of certain AI systems, including ones used in marketing and content production. The good news: most marketing AI use cases fall under “limited risk” — not the heavy compliance regime that applies to high-risk systems. The bad news: “limited risk” still comes with real obligations, and ignoring them can trigger fines of up to €15 million or 3% of annual worldwide turnover.
This guide is the practical version. No legal jargon. No Annex-by-Annex walkthrough. Just what a Swiss SME marketing team actually needs to do between now and August 2026.
Why Swiss SMEs should care about an EU regulation
Switzerland is not an EU member state. But the AI Act applies extraterritorially — it covers any AI system placed on the EU market or whose outputs are used in the EU. If your marketing reaches EU customers, or if you use an AI tool whose provider is based in the EU, you’re likely in scope.
In practice, most Swiss SMEs that market across borders are in scope. The Swiss Federal Council is also preparing its own AI regulation, expected to be closely aligned with the EU Act. Treating the EU AI Act as the compliance baseline now will save you a second migration later.
How the AI Act classifies AI systems
The Act organises AI systems into four risk tiers:
- Unacceptable risk — banned entirely (e.g., social scoring, manipulative subliminal techniques). Marketing teams won’t encounter these.
- High risk — permitted but heavily regulated (e.g., AI used in hiring, credit scoring, critical infrastructure). Most marketing AI does not fall here.
- Limited risk — permitted with transparency obligations (e.g., chatbots, AI-generated content, emotion recognition). This is where most marketing AI lives.
- Minimal risk — permitted with no obligations (e.g., spam filters, automated translation). Most of the rest of marketing AI lives here.
The critical question for a marketing team is: which of our AI uses fall into “limited risk”? Those are the ones that trigger Article 50.
What Article 50 actually requires
Article 50 sets out four transparency obligations. Two apply to providers (the companies building AI systems — OpenAI, Anthropic, Google, and so on), and two apply to deployers (the companies using AI systems — that’s you). Here’s what matters for marketing teams.
1. Chatbots and interactive AI: disclose you’re talking to a machine
If you deploy an AI system that interacts directly with users — a website chatbot, a WhatsApp assistant, a voice agent — you must inform users they’re interacting with an AI, unless it’s already obvious from the context.
Marketing impact: Any conversational AI on your site or in your DMs needs clear disclosure. “Chat with our AI assistant” in the UI is usually enough. Hiding it is not.
2. AI-generated content: make it detectable
If you use AI to generate synthetic audio, image, video, or text content, the provider of that system must ensure outputs are machine-readable and detectable as AI-generated. As a deployer, you don’t build the detection layer — but you should choose tools whose providers comply.
Marketing impact: When you generate blog posts, social images, or video with AI, prefer tools that embed provenance metadata (C2PA, watermarking). If your tool doesn’t, document it internally and label externally where it matters.
3. Deepfakes and manipulated media: label clearly
If you publish AI-generated or AI-manipulated image, audio, or video that resembles real people, places, or events — a “deepfake” — you must clearly disclose that the content has been artificially generated or manipulated.
Marketing impact: If you use an AI-generated image of a fake person in an ad, label it. If you produce a video with an AI-generated voiceover of a real person, label it. The EU has published voluntary transparency icons for this purpose. Use them.
4. AI-generated text on matters of public interest: disclose
If you publish AI-generated or AI-manipulated text intended to inform the public on matters of public interest, you must clearly disclose that the content has been artificially generated or manipulated.
Marketing impact: This applies narrowly. It does not apply to standard marketing copy, product descriptions, or promotional content. It does apply to corporate reports, investor communications, and editorial content that claims to inform the public. If you’re publishing a white paper or research report with AI assistance, disclose it.
Important exemption: Content that has been subject to human review or editorial control is exempt from the disclosure obligation. This is why a human-reviewed AI workflow — where a person reads, edits, and approves AI output before publishing — is both a quality practice and a compliance shortcut. At Ai-Fi, every deliverable is human-reviewed. That’s not just a brand promise; it’s a compliance posture.
A practical compliance checklist for marketing teams
Here’s what to do between now and August 2026.
Step 1: Inventory your AI use in marketing
List every tool, platform, and workflow where AI is involved in producing or distributing marketing content. Include:
- Content generation tools (blog posts, social, ad copy, email)
- Image and video generation tools
- Chatbots and conversational agents
- Personalisation and recommendation engines
- Analytics and predictive tools
For each, note: what it does, what data it processes, who the provider is, and whether the output is published externally.
Step 2: Classify each use case
For each item on your inventory, sort it into one of three buckets:
- Minimal risk — no Article 50 obligation (e.g., translation, spam filtering, internal analytics).
- Limited risk — Article 50 applies (e.g., chatbots, AI-generated published content, deepfakes).
- High risk — unlikely in marketing, but flag anything that touches hiring, credit, or personal data at scale.
Step 3: Implement disclosure for limited-risk uses
For each limited-risk use case, implement the disclosure that Article 50 requires:
- Chatbots: Add clear AI disclosure to the interface.
- AI-generated content: Prefer tools with provenance metadata. Where that’s not possible, add visible labels on the published content.
- Deepfakes: Add disclosure to any AI-generated media depicting real people or places. Use the EU’s voluntary transparency icons.
- Public-interest text: Add disclosure to AI-assisted reports or editorial content. Note that human-reviewed content is exempt — document your review process.
Step 4: Document your review workflow
For AI-generated marketing content, maintain a record of human review. Who reviewed it? When? What changes did they make? This isn’t strictly required for marketing copy (which falls outside the public-interest text rule), but it is the cleanest way to demonstrate compliance if challenged — and it produces better content anyway.
Step 5: Brief your team
Article 50 obligations apply from 2 August 2026. Make sure everyone who touches marketing — internal team, external agencies, freelancers, and yes, managed AI marketing providers — knows the rules and follows them.
What you can ignore
There’s a lot of noise around the AI Act, and most of it doesn’t apply to marketing teams at an SME. You can safely set aside:
- High-risk system requirements (conformity assessments, CE marking, registration in the EU database) — unless you’re building AI for hiring, credit, or critical infrastructure.
- GPAI model obligations (Article 53–55) — these apply to the companies training frontier models, not to companies using them.
- The proposed implementation delays — the Article 50 transparency obligations for deployers are not affected by the EU Digital Omnibus delay proposals. The only piece that may shift is the provider-side content-labelling requirement for systems placed on the market before August 2026, which may move to 2 December 2026.
The bottom line
For most Swiss SMEs, EU AI Act compliance for marketing comes down to three things:
- Disclose chatbots. If a user is talking to an AI, make sure they know.
- Label AI-generated media that looks real. Deepfakes, AI portraits, synthetic voiceovers — label them.
- Human-review what you publish. Not only does it produce better marketing, it exempts you from the public-interest text disclosure obligation.
You don’t need a six-month compliance project. You need an inventory, a classification, a set of disclosures, and a review process. That’s a week of work, not a quarter.
How Ai-Fi handles this for its clients
Every piece of content Ai-Fi produces is human-reviewed before publication. Our MarKomX team operates a documented review workflow that means our clients are already on the right side of Article 50 — for the content we produce. For your other AI use cases, we can help you run the inventory and set up the disclosures. Book a diagnostic call and we’ll walk through it with you.
Compliance isn’t a moat. But doing it well — simply, early, and without over-engineering — is a signal that you run your business the way your customers would want you to. That’s worth more than the fine you’ll never pay.