Teach Before You Trust: The Case for AI Literacy Ahead of Unsupervised Use

By John

A lesson from thirty years of training rooms, applied to the newest tool on the desk

I have spent most of my career designing training for people who are about to be trusted with something consequential: a mortgage recommendation, a compliance judgement, a difficult customer conversation.

In every one of those cases, the sequence was the same. Understanding came first. Independent practice came second. Nobody was handed the keys and told to work it out as they went along.

Generative AI has broken that sequence almost everywhere I look.

Business owners have given entire teams access to tools that can draft a contract clause, summarise a policy, or answer a customer query in seconds, and in most organisations, that access was switched on before anyone was taught how the tool behaves.

As an educator, that sequencing problem is the thing that concerns me most. Not the technology itself, but the gap we have left around it.

Why this is a Training Problem, not a Technology Problem

It is tempting to treat AI adoption as an IT rollout: provision the licence, send a how-to guide, move on.

A Large language Model (LLM) is highly complex & limitless structures and can be wrong in ways that are fluent, well-punctuated, and delivered with total confidence.

That is a comprehension problem before it is a technical one, and comprehension problems are what good and engaging training exists to solve.

Put simply: we would never let a new financial adviser give financial guidance to a customer without first teaching them what good advice looks like and where the boundaries sit.

Therefore, irrespective of industry, we should not be letting staff take AI-generated output at face value without the equivalent grounding.

Three Things Every Professional Should Understand Before Using AI Unsupervised

Whether you run a five-person business or lead learning and development for a large organisation, there are a small number of concepts that I believe should be non-negotiable before anyone is left to use AI tools independently.

1. How the tool produces its answers

Staff do not need to understand the mathematics behind a language model, but they do need to understand its basic character: it predicts plausible text, it does not verify facts, and it can be entirely wrong while sounding entirely sure. Once someone genuinely understands this, their instinct shifts from ‘the AI said so’ to ‘let me check that’, which is exactly the instinct we are trying to build.

2. Where the line sits on data and confidentiality

Every professional handling client, patient, or commercially sensitive information needs a clear, practical answer to one question: what am I allowed to type into this tool? Left untrained, people default to convenience, and convenience is precisely where data protection incidents begin. This is not an abstract compliance point; it is a day-to-day judgement call that untrained staff are currently making alone, several times a day, on your behalf.

3. When a human needs to check the work

Not every AI output carries the same risk. A first draft of internal meeting notes is not a client-facing regulatory communication. Good AI literacy training gives people a working sense of proportionality: what can reasonably go out with a light review, and what demands the same scrutiny you would apply to any other high-stakes piece of work.

What This Looks Like in Practice

Treat AI literacy as mandatory and something that’s completed before independent use begins, not offered afterwards as a courtesy.

Use real examples from your own sector. A generic AI awareness session teaches less than twenty minutes spent reviewing an AI-drafted email against your own standards.

Make the data and confidentiality rules concrete and specific, not a general reminder to ‘use good judgement’.

Revisit the training periodically. The tools change quickly, and one session delivered eighteen months ago will not reflect what staff are using today.

A Closing Thought

None of this is an argument against using AI at work. Used well, these tools are a genuine gain in capability. My argument, as someone who has spent a career designing the training that sits ahead of consequential decisions, is simply this: capability without understanding is not a shortcut, it is a liability wearing the costume of progress.

Teach first. Trust follows.