The short answer
Rolling out AI to your team works best in six steps: survey current use, choose one business plan that fits your software, configure it, pilot it with five to ten people for four weeks, train everyone by role, then review quarterly. In the EU, Article 4 of the AI Act has required AI literacy measures for staff since 2 February 2025, with national enforcement from 2 August 2026. No certificate is needed, but keep a record of training.
What you need before you start
- One owner. A named person who decides which tools are approved, answers questions and keeps the rules current. In a small company that can be the founder or operations lead.
- A budget for business seats. Personal plans are the wrong foundation, because their training defaults and lack of admin controls are the main risks.
- Two or three real tasks per team that take time every week, such as first drafts, meeting follow-ups, report summaries or support replies.
- A short AI policy. You can write it during the pilot. An outline is below, and the full version is in how to write an AI policy for your company.
The legal part: the EU AI Act's AI literacy duty
If your company uses AI systems in the EU, or its AI use affects people in the EU, Article 4 of the EU AI Act applies to you. According to the European Commission, the AI literacy obligation has applied since 2 February 2025. As amended by the AI Omnibus, it requires providers and deployers of AI systems to take measures to support the development of AI literacy of their staff and other people who use AI systems on their behalf, such as contractors, taking account of their knowledge, experience, training and the context in which the AI is used. The Commission stresses that it does not require you to guarantee any specific level of AI literacy for any individual.
What the Commission's AI literacy Q&A says this means in practice:
- A company whose employees use ChatGPT to write advertising text or translate is in scope, and should inform staff about the specific risks of the tool, such as hallucination.
- No certificate is required and no specific level is mandated. The Commission says organisations can keep an internal record of their training and other guidance instead.
- No AI officer or governance board is required.
- National market surveillance authorities supervise and enforce the rules from 2 August 2026, and the Commission says a sanction is more likely if an incident can be traced to inadequate training or guidance.
- The AI Omnibus, which entered into force on 27 July 2026, amended Article 4 but kept the obligation on providers and deployers, with the Commission and Member States supporting them with practical examples.
The Commission suggests a programme cover a general understanding of AI, your organisation's role, the risks of the specific tools you use and how to reduce them, tailored to each group's knowledge and work. That maps neatly onto the rollout below.
How to roll out AI to your team in 6 steps
- Find out what people already use. Send a two-question survey: which AI tools do you use for work, and on which account? Expect personal accounts. The goal is to move that use somewhere safe, not to punish it.
- Choose one main platform with a business plan. Match it to the software your company already runs on (see the table below). One tool everyone knows beats four nobody masters.
- Set it up before inviting anyone. Turn on single sign-on where your plan supports it, decide which connectors are allowed, and confirm the training and retention settings in the admin console.
- Run a four-week pilot with five to ten people. Pick volunteers from different teams, give each one a real weekly task, and meet for 30 minutes every week to share what worked. Collect the good prompts.
- Train everyone, briefly and by role. A 60-minute session covers what the tool is, what data goes where, how to check outputs and three prompts for that team's work. Record who attended: that is your Article 4 record.
- Roll out, then review every quarter. Tools change monthly. Two examples from this autumn: OpenAI retires custom GPTs on 11 December 2026, and Gems on personal Google accounts become read-only on 17 November 2026. Someone has to notice changes like these.
Which business plan fits your company
| If your company runs on | Start with | What the plan gives you |
|---|---|---|
| Microsoft 365, Outlook and Teams | Microsoft 365 Copilot or Copilot Chat with work accounts | Enterprise data protection; prompts not used to train foundation models |
| Google Workspace, Gmail and Docs | Gemini with Workspace, or Gemini Enterprise (its Business edition is aimed at small businesses and teams) | No human review and no training outside your domain without permission |
| A mix, or nothing in particular | ChatGPT Business (2 to 200 people) | Standard and Premium seats, SAML SSO, admin, no training on business data by default |
| Long documents, writing and analysis | Claude Team (2 to 150 users) | Standard and Premium seats, SSO and admin controls, no training by default |
Larger organisations should look at ChatGPT Enterprise or Claude Enterprise, which adds SCIM, audit logs and custom data retention. See ChatGPT vs Copilot, ChatGPT vs Claude and AI tool stacks.
A policy outline you can adapt
- Purpose and scope: who the policy covers, including contractors.
- Approved tools and accounts: the platform you chose, work accounts only, and how to request a new tool.
- Data rules: public, internal, confidential, personal; what may go into which tool.
- Allowed and not allowed uses: for example drafting and summarising yes; decisions about people, legal advice to customers and final financial figures without review no.
- Human review and accountability: the person who sends it owns it.
- Transparency: when to tell customers and colleagues that AI was used.
- AI literacy and training: who is trained, how often, and where the record is kept.
- Incidents: what to do if data is shared by mistake or an AI error reaches a customer.
- Owner and review date.
Worked example: a 25-person agency
A marketing agency on Google Workspace finds, from its survey, that 18 people already use personal ChatGPT or Gemini accounts, several with client briefs. The operations lead approves Gemini with the company's Workspace accounts, so client work stays under Workspace data terms, and allows ChatGPT Business for two strategists who rely on it. Six volunteers pilot it for four weeks on proposals, social copy and meeting notes. The best 12 prompts go into a shared doc. A one-hour training per team follows, with attendance logged in a sheet, and the two-page policy goes live the same week. The quarterly review is in the calendar.
Common mistakes and how to fix them
- Banning AI outright. People keep using it on personal accounts, where you have no control. Fix: approve one tool and make it easy.
- Buying seats and stopping there. Usage fades without training and examples. Fix: give each team three prompts for its own work, such as the ones in Yuzoki kits.
- Training once. Fix: a short refresher each quarter covering what changed in your tools.
- No record of training. Fix: a simple sheet with name, date and topic, the kind of internal record the Commission suggests keeping.
- Measuring hours saved by guesswork. Fix: time two real tasks before and after the pilot, and report those.
For role-by-role ideas, see the best AI for small business and the best AI for project management.
Prompts to copy
Find the tasks worth handing over
I lead a team of [number] [role, e.g. account managers]. Here is a typical week: [list recurring tasks with rough hours]. Rank the five tasks where an AI assistant could save the most time with the least risk, say what the human must still check for each, and suggest a first prompt for the top one.
You get: A pilot shortlist with the checks built in
Build a 60-minute training session
Create a 60-minute AI training session for [team] who use [tool] on a business plan. Cover: what the tool is and is not, our data rules [paste], how to check outputs, and three hands-on exercises from their real work: [tasks]. Give timings, and end with a five-question check I can record.
You get: A ready-to-run session plan and a record you can keep
Draft the policy from the outline
Write a two-page AI policy for a [size] [industry] company using this outline: [paste outline]. Our approved tool is [tool] on [plan]. Plain English, short sentences, one example per rule, and a one-paragraph summary for staff at the top.
You get: A first policy draft to review with your legal or HR adviser
Write the survey
Write a short, friendly, anonymous survey for staff about how they use AI at work: which tools, on which accounts, for which tasks, and what worries them. Six questions maximum, with answer options, and an intro that makes clear nobody will be punished.
You get: A survey that surfaces real usage honestly