Where AI Actually Helps in L&D (And Where It Doesn't)
Recently someone commented on one of my videos on YouTube how L&D is a profession that can and will be replaced by AI. This isn’t the first time someone has brough up the question. So I thought I’d spend some time explaining why this is both true and false.
AI can make parts of our work quicker. That’s true. But deciding what people need to learn, whether the material is useful, and how they'll apply it still requires a human touch. Let's look at both sides of this coin.
AI is a tool with a job to do
Let’s get one thing straight – AI is a tool not the job itself. Here’s what I mean – generative AI is software that produces content, such as text or images, in response to instructions or prompts. Think of it as a very fast drafting assistant: useful, enthusiastic, but occasionally also far too confident about things it hasn't checked.
Speed is valuable when you already know what you're trying to achieve. And this last part is key! If your goal is vague, AI can help you produce the wrong thing at an impressive speed.
Before opening the tool, finish this sentence: “I need help with ___ so that ___.” For example: “I need three practice scenarios so new supervisors can rehearse asking clarifying questions during shift handovers.” If you can’t, go back to the L&D basics before being able to give this tool a task and make yourself a standard for judging the result.
Five places to put AI to work
First, prepare for a learning needs analysis conversation. Learning needs analysis means finding out what people need to learn to do their work well. And the single most important thing you need are the right questions. Imagine you're meeting the warehouse manager about finding errors. Type into your Gen-AI tool of choice “Draft six open questions about picking errors within this context [describe the business context here]. Ask about expected performance, current errors, tools, and working conditions. Do not assume training is the solution.” Now you have some questions to review before the meeting, and you are more prepared to go with the flow and be present during the actual conversation.
Second, organise feedback. If you have comments you're authorized to use, AI can suggest themes. You can start with: “Group these comments by recurring issue. Include a supporting comment for each theme and keep conflicting views visible.” Check the original comments yourself. A tidy summary can quietly hide the one person who spotted the real problem. There, you’ve saved a considerable amount of time and you have more user0friendly data to work with.
Third, brainstorm activities. You can’t be an expert in every single field of the business. This is where AI can come in handy. For example, for supervisors practicing delegation, ask: “Suggest three short exercises where a supervisor must explain a task, agree a deadline, and check understanding. Use everyday retail situations.” AI can spit up several ideas for you to choose from that would fit the audience and available time, then write the details. But again, I want to make this as clear as possible – unless you know what you’re doing, even the best list of activities won’t produce the right results.
Fourth, create a first draft. Let’s say you need a job aid (a short resource people consult while doing a task). Try: “Draft a one-page checklist for handling a returned item using only the approved process below. Flag missing information instead of inventing steps.” The process owner must check it before anyone uses it. Simple, easy and, mostly importantly, quick so you can get something out to your stakeholders for review as fast as possible without waiting for the perfect version, only to be disassembled at first glance.
Fifth, adapt existing material. For example, you can ask: “Rewrite this approved explanation for employees who are new to stock control. Define technical terms, keep every procedural requirement, and list changes separately.” This will quickly allow you to compare both versions and choose the one that would fit best. Simpler language is helpful; accidentally simplifying away an essential condition is less helpful. AI is superb at this types of tasks.
Where things get slippery
AI doesn't see the workplace unless you provide information about it. It may suggest communication training when the actual problem is that two shifts never overlap. It may recommend a role-play without knowing that participants need language support or that the example doesn't fit their work.
It can also produce plausible inaccuracies. NIST's generative AI risk guidance discusses this problem, sometimes called confabulation (aka “hallucinations” or “fabrications”). Check factual claims against approved policies, reliable sources, or the relevant expert. Asking the same tool “Are you sure?” isn't an independent check.
Treat employee information carefully. Use your organization's approved tools and rules. Removing names doesn't necessarily make comments anonymous: a job title, location, and unusual incident may identify someone perfectly well.
You also need to understand the people you're designing for: how adults learn, what motivates them, and how they're likely to respond to different activities. A role-play might look brilliant on paper, but will this group feel comfortable giving it a go? Is there space to move around, or will everyone be wedged between tables like passengers on a budget airline? How does the session connect to what they've already learned and what they'll do next? AI can help you explore options, but you're the one who has to connect those details and judge what will work with these people, in this room, on this particular day.
A review process you can actually use
Follow this sequence:
Give context
Generate a draft
Review it
Improve it
Validate it
Suppose AI suggests “Discuss good feedback in groups.” You review it against the goal and realize nobody actually practices giving feedback. Replace it with a fictional handover problem. Ask participants to explain the missed information, describe its effect, and agree a next step. Give observers a short checklist. Then ask an experienced supervisor to check whether the situation sounds realistic and pilot it with two people.
Our L&D Fundamentals course helps you build the learning design knowledge needed to make those decisions. The downloadable resources in The Ultimate L&D Toolkit can also give your planning a useful starting structure.
Try three small experiments this week
Prepare questions for one meeting. Improve one draft activity. Simplify one approved paragraph. Use fictional content where appropriate, and keep your original version for comparison.
For each experiment, record the drafting time, the checking time, and what needed correcting. Ask whether the final result is clearer and more useful. If checking takes longer than writing it yourself, you've learned something worth knowing.
If you're unsure where AI fits into your own L&D work, bring one workflow to a 1:1 L&D consultation. Start with a real task and a clear purpose. The shiny presentation can wait.
