If you're doing a pre-2-December compliance sweep, here's the item that predates everything on your list: since 2 February 2025, Article 4 of the EU AI Act has required companies that use AI to ensure their staff have a sufficient level of AI literacy. Not the AI team. Not legal. Your staff — which in 2026 means your marketing team, the heaviest AI users in most companies.
The 2 December 2026 deadline getting all the attention is the machine-readable marking one — the grace period ends then for generative systems already on the market before August 2026. We cover that separately. This page is the duty that has been live for almost two years, that most teams still haven't discharged, and that no deadline delay will rescue: the Digital Omnibus (Regulation (EU) 2026/1744) pushed high-risk obligations to 2027/28 and left Article 4 exactly where it was.
Here's the minimum: what the duty requires, who needs training, and how to evidence it.
What Article 4 Actually Requires (Plain English)
Three points, no legal varnish:
- Who it binds: providers and deployers of AI systems. If your team uses AI — chatbots, Copilot or Workspace AI, AI features inside your CRM, CMS or ATS — you're a deployer. No size threshold. No SME exemption. UK and Irish firms are in scope wherever their AI-affected output reaches the EU market.
- Who must be literate: everyone who operates or uses AI systems on your behalf, or whose work is affected by one. In a marketing team that's copywriters, designers, performance marketers, analysts, ops — not one nominated "AI person".
- What "sufficient" means: proportionate to context and role. Technical depth for the people configuring systems; judgement-level awareness for everyone editing AI-assisted output. Regulators expect evidence you took action — not perfection.
Why Marketing Teams Are the Test Case
Marketing is where AI usage went mainstream first: generated copy, generated images, synthetic video "fixes", chatbots on pricing pages, scoring and segmentation inside the automation platform. That means marketing carries a triple exposure:
- Article 4 literacy — live since February 2025 (this page).
- Article 50 transparency — fully applicable since 2 August 2026: chatbot disclosure, deepfake labelling, synthetic-content marking. Our AI Act Article 50 checklist for marketing covers the disclosure duties.
- The 2 December 2026 marking deadline — the grace period for machine-readable marking on generative systems placed on the market before August 2026 ends then. If your team is auditing its generative stack this month, fold literacy into the same sweep — the 2 December deadline guide has the calendar.
And before anyone points at the delay headlines: the Digital Omnibus postponed high-risk obligations to 2027 and 2028 — it did not touch literacy, transparency, marking, or penalties. Marketing's duties are all in the "unchanged" column.
Who Needs Training (and How Deep)
Role-tiered, not one-size:
| Tier | Who | Depth | Format |
|---|---|---|---|
| Awareness | Everyone whose work touches AI output | What AI is, what it's bad at, disclosure duties, what not to paste into a prompt | 60–90 min, all-hands |
| Applied | Daily AI users (copy, design, performance, ops) | Tool-specific capability and limits, hallucination checks, data-handling rules, marking/labelling workflow | Half-day, hands-on |
| Owner | System owners and approvers | Vendor assessment, incident response, evidence keeping, policy maintenance | Half-day + quarterly refresh |
The tier structure matters more than the hours. A single generic e-learning module satisfies no one — not the regulator, not the auditor, not the tender panel.
The Minimum Viable Programme
If you do nothing else, do these four:
- Baseline: inventory who uses which AI tools at what depth. If you don't have that list, start with a shadow AI audit — the ungoverned tools are the ones training is for.
- Deliver the three tiers above. Weeks, not quarters.
- Write it down. Records are the difference between "we did something" and evidence.
- Tie it to policy. Literacy training that references no AI usage policy is a course, not a control.
How to Evidence It (What Auditors and Tender Panels Ask For)
Four artefacts, in order of how often they're requested:
- Attendance record — who was trained, when, on what.
- Training outline matched to roles — showing the awareness/applied/owner split, not a single deck.
- Policy and inventory links — the training points to a named AI usage policy and a tool inventory.
- Refresh cadence — a dated schedule, because literacy decays as tools change.
This evidence pack is also what client procurement questionnaires are starting to ask for explicitly. Blank answers lose deals now, before any regulator calls.
The Governance Wrinkle Most Teams Miss
Someone senior must own this. "The team" is not an owner, and AI literacy that lives in nobody's job description quietly expires. If that sentence just described a gap, copy the pattern from our governance model: who owns what in your marketing stack — named owners per domain, RACI-style, survives team changes. Then walk the training records into your existing security and compliance checklist for marketing automation; literacy slots into the same file as access reviews and vendor attestations, and the delta is smaller than you fear.
Get Your Team Covered
Article 4 is the cheapest AI Act obligation to fix and the most embarrassing to be caught without — a live duty since February 2025, fixable in weeks.
Our AI Literacy Training is a half-day session built around the tools your team actually uses, ending with an attendance record and training outline for your compliance file — the evidence pack above, produced for you. From £1,500 (online, up to 20 people).
Not ready to talk? Take the 2-minute AI Act quiz and see where your team stands. (Practical guidance throughout — not legal advice.)