AI Literacy Fundamentals for Staff: What Every Employee Should Know

AI literacy is the floor — the minimum every staff member needs before they touch a generative AI tool at work. Done well, it gives people the judgement to use AI competently, the vocabulary to ask good questions, and the instincts to flag when something is off. Done badly, it is a 12-slide deck that nobody remembers. This is what a defensible staff AI literacy curriculum actually looks like.

What literacy is and is not

Literacy is not training someone to be a prompt engineer. It is not a deep dive on transformer architectures. It is also not the acceptable-use policy read out loud.

Literacy is a shared baseline that lets the rest of your AI program function: it means a sales rep, a payroll officer, and a project manager all understand the same five or six things about the tools they are now allowed to use. Everything role-specific — briefing, drafting, analysis, support workflows — assumes literacy is already in place. If you skip it, role-specific training has to keep restarting from first principles, which is expensive and inconsistent.

For the wider program context, see the cluster pillar on AI education for organisations.

The five things every employee needs to understand

A staff literacy module that earns its 90 minutes covers five threads, in this order.

1. What these tools actually are

Plain-English explanation of large language models and the broader generative AI category. Two ideas matter: they are pattern predictors, not knowledge bases; and they are confidently wrong in ways that look right. People do not need the maths. They do need a mental model that explains why the tool sometimes makes up a case citation or a customer name.

2. Where they help and where they hurt

A short list of high-confidence use cases (drafting, summarising, reformatting, brainstorming, explaining) and a short list of low-confidence ones (precise calculations, recent facts without retrieval, anything safety-critical without review). Use examples from your own organisation. Generic examples wash off; "remember that incident with the contract numbers" sticks.

3. How to verify

This is the most important section and usually the shortest in bad training. Cover:

• The "would I sign my name to this?" test.
• Cross-checking facts, figures, and citations against source documents.
• The two-pass pattern: AI drafts, human reviews; or human drafts, AI critiques.
• What to do when verification is impossible (escalate, do not ship).

4. What data is allowed in

Map every approved tool to a data classification. People want to know, in concrete terms: can I paste a client email into this? An employee performance review? A board paper? Give specific yes/no answers, not abstract principles. Tie it to your Privacy Act obligations and any sector rules (APRA, health records, education).

5. Where to ask, where to flag

A named channel, a named owner, and a no-blame reporting path for "I think the AI got this wrong" or "I am not sure if I can use this here". Without this, shadow use grows and you lose the signal.

A 90-minute agenda you can adapt

A workable structure for a facilitated session of 10–20 people:

• 0–10 min — why we are doing this, what changes after today.
• 10–25 min — the mental model: what the tools are, in plain language, with two short demos.
• 25–45 min — capability tour: live walkthrough of approved tools on a non-sensitive task from the participants' work.
• 45–65 min — the verification drill: participants are given AI outputs containing planted errors and asked to spot them.
• 65–80 min — data rules and approved-tool map, with quick scenarios ("can I paste this?").
• 80–90 min — where to ask, what is coming next, and how this connects to role-specific training.

The verification drill is the part most programs skip. It is the highest-retention section we run. People remember being wrong about an AI output far more clearly than being told AI can be wrong.

Delivery format

For Australian mid-market organisations, the most cost-effective shape is:

• A short asynchronous primer (20–25 min video plus a quick check) covering threads 1, 2, and 4.
• A live facilitated session (60–75 min) covering threads 3 and 5, with the verification drill.
• A 30-day follow-up artefact: a one-pager pinned in the team space, plus a clinic for questions.

Pure e-learning under-delivers on the verification drill, which is the behavioural change you actually want. Pure live training is expensive and inconsistent across cohorts. The hybrid is the right default.

Common failure modes

The four ways we see staff literacy programs fail in the wild:

• Treating it as a comms exercise. Sending a video and a policy link does not constitute literacy.
• Skipping the data rules. Without explicit allowed-data examples, people guess, and they guess optimistically.
• No follow-through. A single session with no clinic, no community, and no refresh decays in weeks.
• Wrong facilitator. If the person running the session has not used the tools in real work, the room knows within five minutes.

How this connects to the rest of the program

Literacy is necessary but not sufficient. It unlocks role-specific work — see generative AI for marketing teams or the customer support track — and it must be paired with the responsibility layer covered in AI safety and responsibility training. It sits underneath the executive layer too: leaders need a more strategic curriculum, not the staff module with a fancier slide.

What to do next

If you are starting from zero, build the literacy module first, ship it to a single team as a pilot, gather what broke, then roll it organisation-wide. Do not try to perfect the curriculum before contact with reality.