What does the EU AI Act mean for AI search and marketers in 2026?

Why it matters
- The EU’s GPAI provider obligations began applying on 2 August 2025, and Commission enforcement powers became applicable on 2 August 2026.[2]
- Article 50 transparency obligations generally apply from 2 August 2026, but they do not require a universal label on every AI-assisted marketing sentence.[1]
- The GPAI Code and the transparency Code are separate voluntary compliance instruments addressing different duties.[19]
- The DMA covers competition issues such as search-data access, while the Commission’s AI Overviews and AI Mode publisher-content investigation proceeds under Article 102 TFEU.[9]
- Google says its Search generative AI control became available worldwide on 31 August 2026, but publisher-content remedies, remuneration and a confirmed global click-data date remain unresolved in the reviewed record.[26]
In this brief10 sections
- 01What happened
- 02Why it matters
- 03What applies in 2026?
- 04What is the DMA angle on AI search?
- 05What do Google’s AI-search controls mean for publishers?
- 06What we know and what is not confirmed
- 07What should brands and publishers do now?
- 08Timeline: the dates that determine what applies
- 09Frequently asked questions
- 10Sources
The short answer is that, as of 17 November 2026, the EU AI Act governs several different parts of AI search rather than creating one universal “AI search law”: general-purpose AI model obligations are live, Article 50 transparency obligations generally apply, and competition authorities are separately examining Google’s AI search practices under the DMA and Article 102 TFEU.[1] For brands and publishers, the central question is not whether AI was used, but which organization is providing or deploying the system, what content it generates, and whether a specific transparency duty or competition proceeding applies.[1]
What happened
The EU’s general-purpose AI model obligations began applying on 2 August 2025 to models placed on the market from that date, while the European Commission’s enforcement powers for those obligations began on 2 August 2026.[2] These are primarily provider obligations covering matters such as technical documentation, downstream information, copyright policies and public summaries of training content.[3][4]
The separate transparency regime in Article 50 generally became applicable on 2 August 2026.[5] It covers notices when people interact with certain AI systems, machine-readable marking of synthetic outputs, and disclosures for specified AI-generated or manipulated content.[1]
Those dates do not mean that every AI-assisted article, product description or marketing sentence must carry a visible “AI-generated” label. Article 50 contains role-specific duties and exceptions, including provisions concerning human review, editorial control and editorial responsibility.[1][6]
AI search is also being addressed through competition law. On 9 December 2025, the Commission opened an Article 102 TFEU investigation into Google’s use of publisher and YouTube content for AI, expressly including AI Overviews and AI Mode.[7][8] That investigation is separate from DMA proceedings involving search-data access, interoperability and the treatment of publishers.[9][10]
Why it matters
AI search combines several legal relationships that are easy to conflate:
- A model provider may have obligations under the GPAI provisions.[11]
- A provider or deployer of a qualifying interactive AI system may have transparency duties under Article 50.[1][6]
- A publisher may face a disclosure question when it publishes AI-generated or manipulated public-interest text.[1][6]
- A search platform may face competition scrutiny over how it uses publisher content or supplies data to rivals.[12][13]
- A website owner must distinguish AI-search inclusion from model training, ordinary indexing and snippet controls.[14]
The practical consequence is that “EU AI Act compliance” is not a single checklist for a marketing department. A company buying access to ChatGPT, Gemini, Google AI Mode or another AI-search service does not automatically inherit every obligation imposed on the model provider.[1]
The distinction matters for search visibility as well. Google says its separate Search generative AI control became available worldwide on 31 August 2026 and can exclude a site’s links and content from AI Overviews, AI Mode and related generative Search features.[15][16] Google also says the control does not affect inclusion or ranking in other parts of Search.[15]
Exclusion, however, removes a site’s links and traffic from the affected AI features.[17] That makes the choice commercial as well as technical, particularly while publishers continue to seek more information about AI-search referrals and remuneration.[12][7]
What applies in 2026?
General-purpose AI model obligations
The GPAI rules apply primarily to providers of general-purpose AI models, not automatically to every business using an AI product.[1] Article 53 requires qualifying providers to maintain technical documentation, provide information to downstream providers, adopt a copyright-compliance policy and publish a sufficiently detailed summary of training content.[3]
Providers of models presenting systemic risk face additional duties relating to evaluations, risk mitigation, incident reporting and cybersecurity.[4]
The Commission’s framework therefore distinguishes the model supplier from downstream providers and deployers. Documentation is intended to flow through the supply chain, rather than transferring every provider responsibility wholesale to a brand or publisher using an AI tool.[1]
A limited transition remains relevant. GPAI models placed on the market before 2 August 2025 have until 2 August 2027 to comply with the applicable obligations.[18][2] This is a model-level transition and should not be confused with the Article 50 deadline for certain pre-existing AI systems.[18][1]
Article 50 transparency duties
Article 50 generally applies from 2 August 2026.[1] Its obligations vary by system and role:
| Rule | Applies from | Who | What |
|---|---|---|---|
| AI interaction notice | 2 August 2026 | Providers or deployers of qualifying interactive AI systems | Inform people that they are interacting with AI, unless this is obvious; the information must be clear and accessible by the first interaction or exposure.[1][6] |
| Machine-readable output marking | 2 August 2026, subject to a legacy deadline | Providers of systems generating synthetic text, images, audio or video | Ensure outputs are marked in a machine-readable format and detectable as artificially generated or manipulated.[1][5] |
| Public-interest text disclosure | 2 August 2026 | Deployers publishing AI-generated or manipulated text | Disclose the artificial generation or manipulation where the text informs the public on matters of public interest, subject to the relevant human-review and editorial-responsibility exception.[1] |
| Deepfake disclosure | 2 August 2026 | Deployers of AI-generated or manipulated image, audio or video | Disclose the artificial generation or manipulation, with tailored treatment for artistic and similar works.[1] |
| Legacy Article 50(2) systems | 2 December 2026 | Providers of relevant systems placed on the market before 2 August 2026 | Meet the applicable marking obligation by the transition deadline; this is not a general postponement of Article 50.[5][1] |
The machine-readable marking duty and a publisher’s visible disclosure duty are separate. A provider’s technical marking does not automatically satisfy a publisher’s publication disclosure, and a publisher’s label does not replace the provider’s obligations.[1]
The public-interest text rule is narrower than a blanket requirement to label all AI-assisted marketing copy. Article 50 addresses material published to inform the public on matters of public interest and recognizes human review or editorial control together with a person holding editorial responsibility.[6][1]
The Commission’s guidance also indicates that superficial spelling or grammar checking is not enough to constitute meaningful human review.[1] Whether a particular workflow qualifies depends on the system, the content and the actual editorial process.
The GPAI Code and transparency Code
The voluntary GPAI Code of Practice was published on 10 July 2025.[4][19] It contains chapters on transparency, copyright, and safety and security for GPAI providers.[19]
The separate Code of Practice on Transparency of AI-Generated Content was published on 10 June 2026.[5] It supports compliance with Article 50 marking and labelling requirements.[5]
Neither code should be described as replacing the binding legal obligations. The GPAI Code supports model-provider compliance, while the transparency code concerns marking and labelling duties under Article 50.[5]
What is the DMA angle on AI search?
The Digital Markets Act is relevant to AI search, but it is not the same legal track as the AI Act’s transparency rules or the Commission’s Article 102 investigation into publisher content.[9][7]
The Commission opened proceedings on 27 January 2026 concerning Android interoperability and Google Search data-sharing obligations under DMA Articles 6(7) and 6(11).[10] On 16 July 2026, it adopted a decision specifying Google’s search-data-sharing obligations; qualifying AI chatbots that provide online-search-engine functionality were included among potential beneficiaries.[11]
The search-data provisions concern access to anonymized ranking, query, click and view data on fair, reasonable and non-discriminatory terms.[11] They are aimed at competition and access to search infrastructure, not at creating a publisher-content licensing scheme.
The Commission also opened a separate DMA investigation on 13 November 2025 into Google’s demotion of publishers’ content under its site-reputation-abuse policy.[9] That matter should not be merged with the December 2025 investigation into Google’s use of publisher and YouTube content for AI Overviews and AI Mode.[20]
On 23 July 2026, the Commission announced combined DMA fines of €890 million against Google: €460 million concerning Search self-preferencing and €430 million concerning Google Play anti-steering.[21] Those fines were not an AI Overviews content-use fine.[21][22]
The distinction is therefore:
- AI Act: model governance and transparency duties.[18][6]
- DMA: obligations concerning designated gatekeepers, competition, interoperability, ranking conduct and search-data access.[10]
- Article 102 TFEU: the Commission’s investigation into whether Google’s use of publisher and YouTube content for AI may constitute anticompetitive conduct.[23]
What do Google’s AI-search controls mean for publishers?
Google’s Search generative AI control lets website owners include or exclude their sites’ links and content from AI Overviews, AI Mode and generative AI features in Discover.[16] Google says the control became available worldwide on 31 August 2026.[24][15]
Google also says the control does not affect ordinary Search inclusion or ranking signals.[15] That distinguishes it from broader controls:
| Control | Governs | Limitation |
|---|---|---|
| Search generative AI control | Links and content in specified generative Search features | Exclusion also removes the site’s links and traffic from those features.[25] |
| Google-Extended | Training of models used for generative Search responses | It is separate from the generative Search inclusion switch.[15][17] |
nosnippet, data-nosnippet, max-snippet |
Search-result snippets | These do not perform the same function as the generative Search control.[6] |
noindex |
Google Search inclusion generally | It is broader than opting out of generative Search features.[17] |
The control changes the practical decision for publishers, but it does not resolve the broader competition dispute. The European Publishers Council’s February 2026 complaint sought meaningful control, transparency about use and impact, and a fair licensing or remuneration framework.[12][9]
A technical opt-out and commercial remuneration are therefore different questions. Google’s documentation describes what the control does, while publisher complaints ask whether those controls provide an adequate answer to market power, content use, traffic displacement and compensation concerns.[7][13]
Google also introduced generative AI visibility insights covering impressions, pages and countries.[16] However, reporting dated 9 November 2026 said publishers were still awaiting click data, and a confirmed worldwide delivery date for that data was not published in the reviewed report.[26]
What we know and what is not confirmed
Confirmed
- GPAI provider obligations began applying on 2 August 2025 for models placed on the market from that date.[2][1]
- Commission enforcement powers for those GPAI obligations became applicable on 2 August 2026.[2]
- Article 50 transparency obligations generally apply from 2 August 2026.[1]
- The transparency Code of Practice was published on 10 June 2026.[5]
- The relevant legacy Article 50(2) deadline is 2 December 2026.[5][1]
- Google says its Search generative AI control became available worldwide on 31 August 2026.[24][15]
- The Commission’s Article 102 investigation expressly covers AI Overviews and AI Mode.[9]
- DMA proceedings separately address search-data access and related competition questions.[9][19]
Not confirmed in the reviewed record
- A final Commission decision or binding remedy in the Article 102 investigation concerning Google’s use of publisher content for AI Overviews and AI Mode.[27]
- A Commission-awarded publisher compensation or licensing framework for that investigation.[7]
- A confirmed worldwide date for Google click-data delivery to publishers.[26]
- That Article 50 requires a visible label on every AI-assisted marketing sentence.[1]
- That the July 2026 DMA search-data decision created a publisher-content licensing right.[11][28]
What should brands and publishers do now?
- Map the role in the AI supply chain. Record whether the organization is providing a model, providing an AI system, deploying an AI system or publishing AI-generated material.[1]
- Separate model documentation from content disclosure. Obtain relevant provider documentation, but maintain a distinct review process for Article 50 publication and interaction duties.[1][6]
- Classify content before applying a label. Identify interactive AI notices, synthetic media, deepfakes and public-interest text instead of treating every AI-assisted sentence identically.[1]
- Document editorial control. Where a publisher relies on a human-review exception, retain evidence of substantive review and editorial responsibility.[1]
- Separate search inclusion from training. Evaluate the Search generative AI control, Google-Extended, snippet controls and
noindexas different mechanisms.[12][9] - Measure more than AI visibility. Track citations, AI-feature impressions, organic referrals, conversions and revenue separately because visibility does not equal a click or a commercial outcome.
- Keep legal tracks separate in reporting. Identify whether a development concerns the AI Act, DMA, Article 102 TFEU or a private or industry complaint.[9]
Timeline: the dates that determine what applies
| Date | Development | Relevance |
|---|---|---|
| 10 July 2025 | GPAI Code of Practice published | Voluntary provider-compliance instrument.[5] |
| 2 August 2025 | GPAI provider obligations begin for relevant models | Model-level duties become applicable.[18][5] |
| 10 June 2026 | Transparency Code of Practice published | Supports Article 50 marking and labelling compliance.[5] |
| 2 August 2026 | Article 50 generally applies; Commission GPAI enforcement powers begin | Transparency duties and enforcement are live.[1][2] |
| 31 August 2026 | Google says its Search generative AI control became globally available | Publishers can separately manage inclusion in specified AI-search features.[15] |
| 2 December 2026 | Legacy Article 50(2) deadline | Limited transition for relevant pre-existing systems.[18] |
| 2 August 2027 | Legacy GPAI model deadline | Separate transition for models placed on the market before 2 August 2025.[18] |
| 2 December 2027 | Relevant amended high-risk obligations | Not a GPAI or general Article 50 deadline.[29] |
| 2 August 2028 | Relevant Annex I product-related obligations | Separate from the 2026 AI-search transparency timetable.[27] |
Background: why the dates are easy to confuse
The EU AI Act contains different obligations for different actors and systems. GPAI rules concern model providers, while Article 50 concerns specified transparency duties for providers and deployers of certain AI systems.[1]
The dates also include separate transition periods. The 2 December 2026 date concerns relevant pre-existing systems subject to Article 50(2), whereas 2 August 2027 concerns GPAI models placed on the market before 2 August 2025.[1][29]
Competition proceedings follow another timetable and legal basis. The DMA addresses gatekeeper conduct and access obligations, while the Article 102 investigation concerns possible anticompetitive conduct relating to Google’s use of content for AI Overviews and AI Mode.[23]
Frequently asked questions
Does the EU AI Act require every AI-written marketing sentence to be labelled?
No. Article 50 is not framed as a universal label requirement for every AI-assisted marketing sentence.[1] The applicable duty depends on the system, the role of the organization, the type of output and whether an exception for human review or editorial control applies.[1]
When did the EU AI Act’s GPAI obligations apply?
GPAI provider obligations began applying on 2 August 2025 to models placed on the market from that date.[2][1] The Commission’s enforcement powers for those obligations became applicable on 2 August 2026, while certain older models have until 2 August 2027 to comply.[18][2]
What AI Act transparency rules apply to publishers in 2026?
From 2 August 2026, Article 50 generally applies to specified AI-generated or manipulated content, including certain public-interest text and deepfakes.[1] For public-interest text, the provision recognizes an exception linked to human review or editorial control and a person holding editorial responsibility.[1]
Does the DMA regulate Google AI Overviews?
The DMA is relevant to AI-search competition, including search-data access and certain conduct by designated gatekeepers.[11][21] The Commission’s specific investigation into Google’s use of publisher and YouTube content for AI Overviews and AI Mode was opened under Article 102 TFEU, not as a DMA or AI Act proceeding.[9][7]
Can publishers opt out of Google AI Overviews and AI Mode?
Google says its Search generative AI control can exclude a site’s links and content from AI Overviews, AI Mode and related generative Search features, and became available worldwide on 31 August 2026.[15][16] Google also says the control does not affect ordinary Search ranking or inclusion, but exclusion removes the site’s links and traffic from those AI features.[15]
Sources
- 1Transparency obligations under Article 50 of the AI Act | Shaping Europe’s digital future — digital-strategy.ec.europa.eu
- 2Guidelines for providers of general-purpose AI models | Shaping Europe’s digital future — digital-strategy.ec.europa.eu
- 3What legal obligations do providers of general-purpose AI models face under the AI Act as of 2 August 2025? | AI Act Service Desk — ai-act-service-desk.ec.europa.eu
- 4General-Purpose AI Models in the AI Act – Questions & Answers | Shaping Europe’s digital future — digital-strategy.ec.europa.eu
- 5Code of Practice on Transparency of AI-generated Content | Shaping Europe’s digital future — digital-strategy.ec.europa.eu
- 6Article 50: Transparency obligations for providers and deployers of certain AI systems | AI Act Service Desk — ai-act-service-desk.ec.europa.eu
- 7EPC Press Release on Google AI Complaint — cdn.prod.website-files.com
- 8European Publishers Council files formal antitrust complaint against Google over AI Overviews and AI Mode — epceurope.eu
- 9IUST | Policy Department for Justice, Civil Liberties and Institutional Affairs — europarl.europa.eu
- 10Commission opens proceedings to assist Google in complying with interoperability and online search data sharing obligations under the Digital Markets Act - Digital Markets Act (DMA) — digital-markets-act.ec.europa.eu
- 11EUR-Lex - 52026DMA100209 - EN - EUR-Lex — eur-lex.europa.eu
- 12Executive Summary on EPC's Google AI Complaint — medianama.com
- 13A new case against Google in EU and UK hopes to give publishers an opt-out of crawling - WAN-IFRA — wan-ifra.org
- 14AI Features and Your Website | Google Search Central | Documentation | Google for Developers — developers.google.com
- 15Search generative AI control - Search Console Help — support.google.com
- 16New opportunities, control and insights for website owners — blog.google
- 17AI in Search Reduces Publisher Referrals Without Improving User Experience: Experimental Evidence — arxiv.org
- 18AI Act | Shaping Europe’s digital future — digital-strategy.ec.europa.eu
- 19The General-Purpose AI Code of Practice | Shaping Europe’s digital future — digital-strategy.ec.europa.eu
- 20French press body challenges Google over AI summaries — aibc.world
- 21Commission fines Google €890 million for breaches of the Digital Markets Act — ec.europa.eu
- 22EU antitrust regulators quiz publishers on Google's AI search opt-out | Reuters — reuters.com
- 23Commission opens investigation into possible anticompetitive conduct by Google — ec.europa.eu
- 24Five French publisher groups take Google's AI Overviews to Brussels - MACCUS — maccus.fr
- 25EU Complaint Filed Over Google AI Overviews — palai.media
- 26As publishers await click data, Google’s AI visibility push complicates the opt-out decision - Digiday — digiday.com
- 27Google faces EU complaint from five French publisher groups over AI Overviews — ppc.land
- 28Commission fines Google €890 million for breaches of the Digital Markets Act — ec.europa.eu
- 29Regulation - EU - 2026/1744 - EN - EUR-Lex — eur-lex.europa.eu