Verified as of August 2026: EU and UK rules are actively changing, and some widely-shared claims are out of date. This is operational orientation, not legal advice; confirm with official sources or your data protection adviser.
From 2 August 2026, the EU AI Act's Article 50 transparency obligations apply. In marketing terms this means two concrete things: an AI system that interacts with people must make clear they are dealing with AI unless that is obvious from context, and synthetic image, audio, video and certain text output must be marked in a machine-readable way as artificially generated or manipulated. Deepfake content and AI-generated text published to inform the public on matters of public interest carry additional disclosure duties. What Article 50 does not do is require you to slap an "AI-assisted" label on every blog post you drafted with a model's help. That is the most repeated over-claim in the marketing press, and it is wrong.
Key Takeaways
- The AI Act has been in force since August 2024. Prohibitions and AI-literacy duties applied February 2025; general-purpose AI model obligations applied August 2025.
- Article 50 transparency applies from 2 August 2026 and was NOT postponed. The "AI Omnibus" amendment delayed high-risk deadlines only.
- Standalone Annex III high-risk obligations moved to December 2027; embedded Annex I high-risk to August 2028. Saying "the AI Act is delayed" without that qualifier is inaccurate.
- A reported four-month grace period applies to watermarking for systems already placed on the market before 2 August 2026, confirm scope before relying on it.
- Chatbots must disclose they are AI. Synthetic media must be machine-readably marked. Deepfakes need a visible disclosure.
- Ordinary AI-assisted copywriting, where a human edits and takes editorial responsibility, is not what the text-labelling duty targets.
The Timeline, Stated Accurately
The AI Act entered into force on 1 August 2024 with staggered application dates.
- February 2025, prohibited practices (Chapter II) and AI literacy obligations (Article 4) applied.
- August 2025: obligations for providers of general-purpose AI models applied, along with governance provisions.
- 2 August 2026: Article 50 transparency obligations apply, alongside the general application of most remaining provisions.
- December 2027: revised deadline for standalone Annex III high-risk systems, following the AI Omnibus amendment.
- August 2028, revised deadline for Annex I high-risk systems embedded in regulated products.
This is the part the trade press keeps flattening. There is no single "AI Act deadline." There are several, and the one that changed is not the one that affects marketers most.
Correcting Error One: "The AI Act Was Delayed"
An amending package commonly referred to as the AI Omnibus adjusted the Act's timetable. What it postponed was the application of high-risk system obligations. Article 50's transparency duties were not part of that postponement and take effect on 2 August 2026 as originally scheduled.
Why the error spread: headlines reporting "EU delays AI Act" are accurate about high-risk timelines and useless without that qualifier. A marketing team reading those headlines concludes it has until 2027 to think about chatbot disclosure. It does not.
Note also that the AI Omnibus is a different instrument from the Digital Omnibus data package covering GDPR and ePrivacy: which remains under negotiation and is not law. Conflating them produces confident, wrong statements in both directions.
Correcting Error Two: "Every AI-Assisted Blog Post Needs a Label"
Article 50(4) addresses AI-generated or manipulated text published with the purpose of informing the public on matters of public interest. It contains an explicit carve-out where the content has undergone human review or editorial control and a natural or legal person holds editorial responsibility for publication.
Two things follow.
First, the provision is aimed at synthetic content in a public-information context, the disinformation problem, not at commercial copywriting generally.
Second, even within that scope, human-reviewed content under editorial responsibility is carved out. A product page, a case study, or a how-to article that a marketer drafted with model assistance and then edited and signed off is not the target of this rule.
What remains true regardless: platform policies, advertising standards bodies, and consumer protection law all continue to apply. If AI-generated content misleads consumers about a product, the problem is unfair commercial practices law, not Article 50.
The Obligations Table
| Obligation | Who it applies to | Applies from | What a marketing team actually does |
|---|---|---|---|
| Disclose that a person is interacting with an AI system | Providers of AI systems intended to interact directly with natural persons | 2 Aug 2026 | Label your site chatbot and AI voice agent clearly at the start of the interaction; no disclosure needed where it is obvious to a reasonably informed person |
| Mark synthetic audio, image, video or text in machine-readable format as artificially generated or manipulated | Providers of AI systems generating synthetic content | 2 Aug 2026 | Ensure your generation tooling embeds provenance metadata; ask vendors in writing |
| Disclose deepfake content (image, audio, video resembling real persons, objects, places or events) | Deployers | 2 Aug 2026 | Visible disclosure on any synthetic likeness in a campaign |
| Disclose AI-generated/manipulated text published to inform the public on matters of public interest | Deployers | 2 Aug 2026 | Applies to public-interest publishing; carve-out where human review and editorial responsibility exist |
| Inform people subject to emotion recognition or biometric categorisation | Deployers | 2 Aug 2026 | Rare in marketing; if you use audience emotion analytics, this is your provision |
| AI literacy, ensure staff have a sufficient level of AI literacy | Providers and deployers | Already applied (Feb 2025) | Documented internal training; this one is already live |
| Prohibited practices (manipulative techniques causing significant harm, exploiting vulnerabilities, certain biometric categorisation) | All | Already applied (Feb 2025) | Audit any persuasion or targeting technique against Chapter II |
| Grace period on watermarking for systems on market before 2 Aug 2026 | Reported at approximately four months | From 2 Aug 2026 | Do not treat as a planning assumption; verify current text |
| High-risk obligations, standalone Annex III | Providers/deployers of Annex III systems | December 2027 (postponed) | Marketing rarely in scope; HR/recruitment tooling can be |
| High-risk obligations, Annex I embedded | Providers of regulated products | August 2028 (postponed) | Not a marketing concern |
Where Marketing Teams Are Actually In Scope
Conversational interfaces
If you run a chatbot, an AI SDR, a voice agent, or an AI-driven support widget that talks to prospects, you are in the clearest area of scope. The disclosure must come at or before the point of interaction. "Hi, I'm an AI assistant" at the top of the chat window satisfies the spirit. A disclaimer in the privacy policy does not.
The obvious-from-context exemption is narrower than teams hope. A widget branded with a human first name and a photograph is not obvious.
Synthetic creative assets
AI-generated imagery in ads, AI voiceover in video, AI-generated video, and synthetic presenters all fall under the marking obligation. The machine-readable marking duty falls primarily on the provider of the generating system, which means your practical action is procurement: ask your tooling vendors, in writing, how they implement machine-readable marking and whether their output carries provenance metadata.
Synthetic likeness
If a campaign uses an AI-generated likeness of a real person, a founder, a celebrity, a customer, you are in deepfake territory and a visible disclosure is required, on top of every publicity-rights and consent question that already applied.
Recruitment marketing
Employer branding teams that use AI to screen or rank applicants are looking at Annex III high-risk classification, not Article 50. Those deadlines moved to December 2027, which is breathing room, not exemption.
What To Do Before and After 2 August 2026
Inventory
List every AI system touching a customer-facing surface: chat widgets, personalisation engines, generative creative tools, voice agents, review-response tools. For each, record whether you are the provider or the deployer, the obligations differ.
Fix interface disclosure
This is the cheapest and highest-value item. One line of copy at the start of every conversational interface.
Send vendor letters
Ask each generative tool vendor: do you embed machine-readable provenance markings in output, in what standard, and does it survive common processing? Keep the replies. If they cannot answer, that is a procurement signal.
Write a labelling policy that is accurate rather than fearful
The over-cautious response, labelling every page "AI-assisted", has real costs. It invites reader distrust, gives no legal benefit where the obligation does not apply, and sets a standard you will fail to apply consistently. Write down what you actually label and why.
Handle AI literacy
Article 4 already applies. A short documented internal session on capabilities, limits and permitted use is a proportionate answer for a marketing team.
Interaction With Everything Else
Article 50 sits alongside, not instead of, other regimes. GDPR still governs personal data used in AI-driven personalisation. National consumer protection law still governs misleading claims. Platform policies, Meta, Google, LinkedIn, TikTok, already require synthetic-media disclosure on their surfaces and enforce faster than any regulator. The DSA imposes duties on the platforms, which is why the platform rules keep tightening even though your own website has no DSA obligation.
Frequently Asked Questions
Was the EU AI Act delayed?
Partly, and not the part that matters most to marketers. The AI Omnibus postponed high-risk obligations, standalone Annex III systems to December 2027 and Annex I embedded systems to August 2028. Article 50 transparency obligations were not delayed and apply from 2 August 2026.
Do I have to label blog posts written with AI assistance?
Generally no. Article 50(4) concerns AI-generated text published to inform the public on matters of public interest, and carves out content subject to human review where a person holds editorial responsibility. Ordinary edited commercial content is not the target.
Does my website chatbot need to say it's AI?
Yes, unless it would be obvious to a reasonably informed person that they are interacting with an AI system. Disclose at the start of the interaction, in the interface, not in a policy document.
What counts as a deepfake under the Act?
AI-generated or manipulated image, audio or video content that appreciably resembles existing persons, objects, places, entities or events and would falsely appear authentic. Using a synthetic likeness of a real person in a campaign puts you here.
Who is responsible for watermarking, me or my AI tool vendor?
The machine-readable marking obligation falls primarily on providers of the generating system. As a deployer you have your own disclosure duties for deepfakes and certain public-interest text. Practically, verify your vendors' implementation in writing.
Is there a grace period on watermarking?
A four-month grace period has been reported for systems already placed on the market before 2 August 2026. Verify the current text and its scope before treating it as a planning assumption.
Does the AI Act apply to a company outside the EU?
It can. The Act applies to providers placing systems on the EU market and to deployers established in the EU, and extends to situations where output is used in the EU. Non-EU agencies serving EU clients should assume relevance.
What are the penalties?
Tiered, with the highest bands reserved for prohibited practices and reaching into the tens of millions of euros or a percentage of global turnover. Transparency breaches sit in a lower band but are not trivial.
Is the AI Omnibus the same as the Digital Omnibus?
No. The AI Omnibus amended AI Act timelines. The Digital Omnibus is a separate data-protection package covering GDPR and ePrivacy that remains under negotiation and is not in force. They are routinely confused.
Does AI-generated content hurt my SEO?
That is a search-quality question, not a regulatory one. Search engines reward content that demonstrates genuine experience and usefulness regardless of how it was drafted. Thin AI output performs badly because it is thin, not because it is AI.
Further Reading
- European Commission, AI Act policy hub
- European Commission, AI Act implementation timeline
- Search Engine Land for ongoing coverage of AI content and search policy
If you want AI used in your marketing in a way that survives both regulatory scrutiny and reader scrutiny, that's a conversation worth having. More on how I work at younusfardeen.com.