Article 4 does not ask for a certificate. It asks for a file you can describe.

Who this is for

This is for the Learning and Development manager who has been asked to evidence AI literacy and does not yet have a folder they can describe. Legal has already confirmed that the organisation deploys AI systems. Someone then forwarded Article 4 and asked what you can show. You are not being asked for a slide about culture. You are being asked for a file: who was trained, on what systems, to what standard, and whether that mapping matches the job.

If you are still deciding what the duty means, read L&D has to evidence staff AI literacy under Article 4 first. That page is the explainer. This page is the filing checklist.

Experrt trains those cohorts live and keeps the export with you. Read this before you buy a badge, including from us.

What changed

Article 4 of the EU AI Act requires providers and deployers to take measures supporting AI literacy among staff and others who use the systems on the organisation's behalf. National market surveillance authorities began supervising that duty on 2 August 2026. It is now September. A smile sheet from a Copilot rollout is a weak answer. "We bought licences" is not better.

The Commission AI literacy Q&A is explicit: you still have to take measures that support AI literacy, taking account of the people and the systems they use. You are not required to guarantee a score for each person. You are required to be able to describe what you did.

What they will actually ask is four things. Who was in scope. On what tools and what work. To what standard. And whether that standard matches the job. If you cannot open a dated pack that answers those four, you do not have a measure. You have a story.

What AI literacy actually means at work is the definition you should already have written down. This note is what to file next to it.

What belongs in the folder

The pack is not a marketing folder. It is the set of documents you can hand to counsel, an auditor, or a customer questionnaire without rewriting the story each time. Miss one of the fields below and the rest become decoration.

A role and risk map, not a headcount list. Write which roles already use AI on customer or regulated work, which tools they use, and what happens if an output goes out unchecked. A receptionist who summarises notes and a claims handler who uses a model in a file are not one audience. A list of names without that mapping is a roster, not a measure.

Named courses mapped to those roles. Staff who use the tools on their own work, line managers who approve AI-assisted work, and the sponsor who has to describe the measures later do not sit the same session. Map AI Foundations for Every Role and Responsible AI Use at Work to the people who already use the tools. Map AI Governance and Oversight for Managers to the people who sign the work out. Map Sponsoring an AI Literacy Programme to the person who owns the pack. How to commission workforce AI training and still have a record is the buying brief for that mapping.

Attendance, as a register. Who was invited, who attended, which session, which date. A certificate of attendance without a register is a badge. If someone sent a deputy, write that down. If a session was missed, write that down too.

Submitted and marked work, or an observed outcome. Literacy includes whether someone can handle a real file, not only whether they sat in the room. If a session was practice-only, write that down and say why. If work was graded, keep the grade. An observed outcome from a facilitator who watched the work is enough when a written submission would have been theatre.

Facilitator credentials. Name the person who ran the room, and what qualifies them to run it against your policy and your tools. An unnamed vendor webinar does not give you that field. If the facilitator has not seen your policy, the session was a product tour.

A dated export the organisation owns. Not a screenshot in a trainer's inbox. A file you can reopen next quarter: cohort, course, dates, people, outcomes. Store it where you already keep the AI policy. Name the owner. A pack that only the vendor can regenerate is not yours.

A written limit on what the programme claims. You can say you took measures supporting literacy for the roles you trained. You cannot say the organisation is "Article 4 certified," because no such certificate exists. Write the limit in the pack so a later slide does not invent it.

What does not belong

A vendor badge on its own proves that someone paid for a logo. It does not prove who was trained, on what work, or to what standard. An LMS completion percentage with no role map and no marked work is the same problem in a dashboard. A generic e-learn sent to every role treats literacy as one product. It is not.

Do not file assessment results as if they were the duty. The AI readiness assessment is a diagnostic. Completing it can tell you which roles to train first. Filing the scores as Article 4 measures is a category error. They are not a record of literacy training.

Do not file Calendly screenshots, or any other proof that someone was invited rather than trained. An invitation is not attendance. A hold in a diary is not a register.

In-person AI training versus an LMS module is the longer argument for why a completion tick is a weak measure. Managers, not champions, after you buy the tools is why a champions-only cohort does not cover the people who already use the tools on live work.

How this differs from the Article 4 briefing

The Article 4 briefing for L&D explains the duty: what changed, why a certificate is the wrong product, and who should sit which live cohort. This page assumes that argument is settled. It is the filing checklist: the folder you should be able to open when someone asks what you did.

Keep both. The explainer is what you send when someone still thinks a webinar discharges the duty. The pack is what you open when they have moved on to evidence.

How to build the pack this quarter

Start with the roles that already use AI on customer or regulated work. Do not start with a company-wide e-learn. Write the role and risk map first, even if it is one page. If you cannot name the tools those roles already use, you are not ready to buy training.

If you need to aim the first cohort, run the AI readiness assessment. Treat it as a diagnostic, not the measure. Then commission live training from AI literacy training: a practitioner cohort for the people already on the tools, a manager session for the people who approve the work, and a sponsor session for the person who will have to describe the pack later.

File the dated export next to the AI policy the week the cohort ends, not "when we have time." Name the owner. How to measure whether AI training actually stuck is useful after the first cohort. It is not a substitute for the register.

If you want the programme scoped against the roles you already have, start on the AI literacy training page. Bring the list of roles and the tools they already use. You can also write to ag@experrt.com.

Cost honesty

For the roles that already use AI on customer or regulated work, a live programme of this kind usually sits in the £20,000 to £30,000 a year range. It is not a six-figure monthly transformation project, and it should not be sold as one.

That figure is for a scoped literacy programme you can file, not for an academy rebuild. If a vendor quotes a transformation day rate before they will name the courses, send it back.

Experrt will not tell you that a course makes you compliant. We will run the cohorts, take the register, grade the work, and give you the pack. The brief is on the AI literacy training page.