Teaching a Team to Use AI Without a Training Budget

The gap is judgement, not technique

The common failure is not that people write poor prompts. It is that they do not know when to trust the output, and either over-trust it — shipping invented facts — or under-use it for tasks it would genuinely help with.

Training should therefore be mostly about calibration: what these tools are reliable at, and what they are confidently wrong about.

Four things everyone should understand

1. It predicts plausible text. Fluency is not accuracy, and it will invent details with complete confidence.
2. It is strong at transformation, weak at facts. Give it the material; do not ask it to supply the material.
3. It does not know your business unless you tell it, and it does not remember between conversations unless configured to.
4. Everything you paste goes somewhere. Which is why the approved tool and the data rules matter.

Teach with their own work

Generic training produces generic use. Thirty minutes spent taking three real tasks from someone's actual week and doing them together is worth more than a full-day course on prompting.

For each role, find the three tasks that are mechanical transformations — summarise this, draft that from these notes, turn this into that format — and demonstrate those.

Show the failures deliberately

Ask it for a statistic and watch it invent a plausible source. Ask about your own company and watch it get details wrong. Show a confidently incorrect answer in their own domain.

People calibrate far faster from seeing a confident error than from being told errors happen. This is the single most valuable fifteen minutes of any AI training session.

Create a shared library of what works

When someone finds a prompt or a workflow that works well for a recurring task, put it somewhere everyone can use it. This spreads competence far faster than training does.

Keep it practical: the task, the prompt, and a note about what to check in the output.

Set the expectation about review

The rule that should survive every training session: you are accountable for anything you send, whoever or whatever drafted it. That single sentence does more for output quality than any technique.

Frequently asked questions