What should AI training cover, in short?
Cover five things, in this order of importance for most small teams: the rules (approved tools and information boundaries), checking output, giving the tool good context, the basics of how these tools work and fail, and practice on real tasks from the person's role. Practice is where the other four become habits.
Avoid starting with long histories of AI or lists of every tool on the market. Those are interesting but rarely change what someone does on Tuesday afternoon with a client email.
Rules
Which tools and plans are approved, and what information must stay out.
Checking
Verifying facts, sources and figures; separating confidence from evidence.
Context
Writing a brief with purpose, audience, source material and constraints.
Basics
What a language model does, why it can state false things confidently, and when not to use it.
Practice
Hands-on tasks from the learner's own role, reviewed against a stated standard.
Why should rules come before prompting?
People who are unsure what they may paste in either hold back or take risks, and neither helps. A short, clear rule on approved tools and never-enter information removes that uncertainty before anyone practises. It also gives the trainer something concrete to test, such as asking a learner to spot the confidential detail in a sample brief.
Why does checking output deserve so much time?
Because the main failure of these tools is plausible error. NIST's Generative AI Profile defines confabulation as the production of confidently stated but erroneous or false content, colloquially known as hallucinations. A fluent paragraph with one invented fact is more dangerous than an obviously poor one.
Good training makes checking concrete: open the source and find the supporting passage, recompute important figures, and keep a short record of what was checked. It also covers when a qualified human must decide, such as legal, financial or safety matters.
How should the content change by role?
The core skills are shared, but the tasks and risks differ by role, so the practice should too. A marketer practises briefs and brand voice; a recruiter practises job adverts and must learn where candidate data cannot go; a finance analyst practises variance commentary and must check every figure.
| Role | Example practice task | Main risk to train |
|---|---|---|
| Marketing | Draft a campaign brief from supplied facts | Invented claims about the product |
| Sales | Research summary before a call | Outdated or unsourced company facts |
| HR and recruiting | Rewrite a job advert for clarity | Personal data and biased wording |
| Finance | Draft commentary on a supplied table | Wrong figures or unsupported causes |
| Support | Draft a reply from the help centre text | Invented policy details |
Want to see how this connects to a role?
Explore the planned team trial ↗What does one good lesson look like?
At Kestrel Recruitment, a fictional 25-person agency, a lesson on job adverts starts with a two-minute explanation of why specific requirements beat generic ones. A worked example then shows a vague prompt, "write a job ad for an accountant", producing a generic advert with an invented salary.
The improved brief supplies the real requirements and says "use only these details; do not add a salary". The learner then writes their own brief for an open role, checks the output for invented details and biased wording, and submits it with a note on what they changed. The reviewer checks against a three-point standard.
Does the law say what AI training must cover?
In the EU, the European Commission's AI literacy questions and answers, dated 27 July 2026, says Article 4 of the AI Act asks organisations to take measures to support AI literacy, without a fixed format or certificate. It suggests considering the organisation's role, the risks of the systems used, and staff knowledge and context, and says legal and ethical aspects are encouraged. NIST's AI RMF Playbook similarly calls for personnel to receive AI risk management training to perform their duties. This is general information, not legal advice.
Where does Tosheo.ai fit?
Tosheo.ai is being built around a skill graph per role covering prompting, drafting, research, data summarising and checking AI output, with lessons that end in a hands-on task in the team's own tools. Growth plans are planned to use uploaded company documents so examples match your context. The content is planned scope, not a finished curriculum.
When is Tosheo.ai not the answer?
If your team needs deep technical training, such as building models or coding with AI, choose a specialist course. If you need to author your own course content, custom course authoring is not included. If you need lessons in languages other than English, those are not available at launch.