FAQ

For Chinese companies entering Europe, what do GDPR and the EU AI Act require of websites and AI applications?

Two separate tracks. For the website, it is mostly GDPR: collecting emails, form and behavioural data needs a lawful basis, the privacy policy must be real (not a copied template), consent must be evidenced, and data-subject rights (access, erasure) need a working process. For AI applications, the EU AI Act grades by risk: most enterprise applications are limited-risk, where the core duties are transparency (users must know they are dealing with AI) and record-keeping; high-risk uses such as recruitment, credit or critical infrastructure carry far heavier duties and must be classified at project start. The Data Act and NIS2 also affect connected products and digital supply chains. This page is a practical framework, not legal advice — formal conclusions need European counsel.

Figure 01 · GDPR and AI Act: key dates

  • 2018-05-25

    GDPR applies (Art. 99(2))

  • 2024-08-01

    AI Act enters into force

  • 2025-02-02

    Prohibitions and AI literacy obligations apply

  • 2025-08-02

    GPAI model obligations apply

  • 2026-08-02

    General application; transparency duties (Art. 50) apply

  • 2027-12-02

    Annex III high-risk duties apply (deferred by the Digital Omnibus)

  • 2028-08-02

    Annex I product-embedded high-risk duties apply

High-risk dates as amended by Regulation (EU) 2026/1744 (EUR-Lex original, in force 27 July 2026); the original 2 Aug 2026 / 2 Aug 2027 dates no longer apply.

The website track: four practical GDPR items

GDPR (Regulation (EU) 2016/679, applicable since 25 May 2018) sets four practical requirements:

  • Lawful basis and minimisation (Art. 6, Art. 5(1)(c)). Every processing operation needs one of the six lawful bases, and data collected must be limited to what is necessary. Marketing and consultation consent are separate, never bundled (Art. 7(4)).
  • A privacy policy that matches reality (Art. 12–14). The policy must match what the site actually does — a copied template claiming unused practices is itself a risk. Art. 13 requires purposes, retention, recipients and rights to be stated at collection.
  • Consent evidence (Art. 7). Who consented, when, and to what must be retrievable — Art. 7(1) requires the controller to demonstrate consent. This site's lead system is built to that standard (timestamp plus source path). Withdrawal must be as easy as giving consent (Art. 7(3)).
  • A rights process (Art. 15–17, Art. 20, Art. 12(3)). Access, rectification, erasure and portability requests need a defined entry point and timeline. Art. 12(3) requires action within one month, extendable by two months for complex cases, reasons given within the first month.

The AI track: classify first, then build

The EU AI Act (Regulation (EU) 2024/1689, in force since 1 August 2024) grades applications by risk: unacceptable (prohibited), high, transparency, and minimal or no risk — the Commission's four levels.

Conversational support, content generation and internal assistants are usually not high-risk, but they carry transparency duties: a chatbot user must be told they are interacting with an AI system (Art. 50), and AI-generated content must be identifiable, with deep fakes clearly labelled. The industry calls this tier "limited risk"; the regulation uses no such name.

High-risk has two routes (Art. 6): AI as a safety component of an Annex I product needing third-party assessment (Art. 6(1)); and the Annex III use cases (Art. 6(2)): recruitment screening (4(a)), creditworthiness assessment (5(b), fraud detection excluded), and others. High-risk systems require risk management, data governance, documentation, logging, deployer transparency, human oversight and post-market monitoring (Art. 9–14, 72); deployers must assign oversight to competent, trained, authorised natural persons (Art. 26). Design this in at project start — retrofitting costs far more.

Two details matter. First, Art. 6(3) excepts Annex III systems posing no significant risk and doing only one of four narrow functions — but profiling is always high-risk. Second, the timeline: generally applicable since 2 August 2026, but the Digital Omnibus (Regulation (EU) 2026/1744, in force 27 July 2026) pushed Annex III duties to 2 December 2027, Annex I to 2 August 2028.

Four points most often missed on the website side

  • Analytics must sit behind consent. Statistics and marketing scripts loading before consent are the most common European website violation. The consent gate is ePrivacy Directive Art. 5(3) (as amended in 2009): accessing information on a user's device requires prior consent. Consent management must block unconsented loading, not just show a banner.
  • Contact forms must say what happens. Purpose, retention period, who can access the data, and how users exercise deletion and export rights belong next to the form and in the privacy page (Art. 13).
  • AI chat and support widgets need a data-flow check. Where visitor input goes, who processes it, where it is stored, and whether it trains models — covered by a data processing agreement (Art. 28).
  • Cross-border transfers need an assessment on file. When headquarters sit outside Europe and data flows back to China, Chapter V (Art. 44–49) requires a valid transfer mechanism (adequacy decision, safeguards such as SCCs, or a derogation); a transfer impact assessment is usually filed as well.

Where this meets GEO

Solid compliance is a visibility asset: consistent, factual privacy and company pages help engines judge trustworthiness. Technical foundations matter: in Negency's flagship measurement of 30 Chinese exporter companies' global homepages (a single-day sample on 30 October 2026, measuring technical website visibility only, not brand mention rates in AI answers), machine readability averaged 69.2 (median 68.5), 17 of 30 sites had no JSON-LD at all. Source: Flagship Report 2026, Issue 1. The Negency team does this work in European automotive and manufacturing.

Penalties: know the scale first

GDPR fines for core breaches run up to €20 million or 4% of worldwide turnover, whichever is higher (Art. 83(5)); under the AI Act, prohibited practices draw up to €35 million or 7%, other infringements up to €15 million or 3% (Art. 99). Those are ceilings, not typical fines.

This is a summary of common requirements, not legal advice; implementation needs counsel familiar with EU data law. Preparing the engineering facts first — a data-flow map and a retention table — speeds that up.

Bring the agent solution into your workflows

Read the answer — the next step is judging it against your scenario. Discuss an agent solution: tell us the business problem and your systems for a preliminary suggestion.

Related questions

How is GEO different from SEO?

SEO optimises a page's rank in a list of links, aiming for clicks. GEO (Generative Engine Optimization) optimises how discoverable, understandable, trustworthy and citable a company is inside AI-generated answers — the goal is to be in the answer itself. They share technical foundations (crawlability, structured data, content quality), but GEO additionally requires governed entity facts, a machine-readable surface (llms.txt, Markdown mirrors) and a third-party mention ecosystem. A site that ranks well is not automatically visible to AI: engines cite the companies whose facts they can fetch, parse and verify. The concrete differences sit in four dimensions — goal, object, content structure and evaluation — compared item by item in the body below.

How can Chinese companies going global build AI search visibility?

For Chinese companies going global, the visibility gap is rarely content volume — it is the machine-readable surface. Our audit of 30 Chinese outbound companies found machine-readability the weakest category (69.2/100 average): 17 had no JSON-LD on the homepage, 15 had a missing or thin sitemap, 13 had a missing or duplicated H1. The path has four steps: audit bot reachability and machine-readability first; govern entity facts so Chinese and English statements agree; publish question-shaped content built on real overseas buyer questions; then earn third-party mentions (media, communities, partners) and verify with periodic resampling against a fixed question set.

What should a corporate website change for the AI search era?

Five changes. (1) Bot reachability: robots.txt explicitly allows AI crawlers and live fetches are not stopped by firewalls or challenge pages. (2) A machine-readable surface: an llms.txt fact file (optional aid), a complete sitemap, Markdown versions of core pages (optional). (3) Entity structure: JSON-LD naming entity types such as Organization, clean title/H1 discipline, hreflang pairing on bilingual sites. (4) Question-shaped content: pages organised around buyer questions, short answers first, FAQ schema attached. (5) Measurability: analytics that separate human and bot traffic and attribute AI-referred inquiries. Negency's own site was rebuilt to exactly this standard, and the public audit tool scores against it.

What is llms.txt, and should every company website have one?

llms.txt is a plain-text brief at the site root written for AI systems: who the company is, what it does, where the core pages live, and the canonical facts and definitions. It does not by itself raise citation rates — in Ahrefs' May 2026 study of 137,210 domains, 97% of published llms.txt files received zero requests that month, so its direct effect is near zero — but it is the cheapest fact-governance move available: one file anchors a single version of the truth that crawlers can rely on. Verdict: worth adding, useless alone; it works as part of entity-fact governance, structured data and question-shaped content. Negency's llms.txt is a full fact bible: company definition, offering boundaries, sourced cases and citation rules.

What should an enterprise prepare before deploying AI agents?

Four preparations, and we advise against starting without any of them. (1) Scenario selection: pilot on a workflow that is frequent, comparatively rule-clear and verifiable — not the most glamorous one. (2) Knowledge and data foundations: govern the knowledge base first (deduplication, versioning, permissions); a dirty knowledge base only scales errors. (3) Governance design: permission boundaries, human approval points and audit logs decided before launch, not after. (4) A measurement baseline: record the manual process's time and error rates before the pilot, or value can never be proven afterwards. Negency's method is a small controlled pilot, verified before scaling, with governance as a design input rather than a patch.