AI Is Making Bad Training Faster

AI is making bad training faster when teams use it to speed up production without improving the learning strategy. 

That sentence sounds blunt, but I think it is the most useful way to talk about the risk. The problem is not that AI can help create training. The problem is that AI can help create training before anyone has done the harder work of deciding what the training is meant to change. 

I have seen this pattern long before AI. A business problem appears. A training request follows. The solution becomes a course because a course is the thing the organization knows how to ask for. AI now makes that response faster, smoother, and more polished. It does not automatically make it right.

Polished output can hide weak thinking 

One reason AI-generated training feels risky is that it can sound competent even when the underlying design is thin. It can write learning objectives, create scenarios, draft quiz questions, and summarize key points in a professional tone. On the surface, the output looks usable. 

That is exactly why L&D teams need to be careful. The issue is not whether the writing is clean. The issue is whether the training is anchored to the right performance need. 

A polished course about objection handling does not mean salespeople will handle objections better. A tidy module on coaching does not mean managers will have stronger coaching conversations. A clean checklist for customer service does not mean people can use judgment when a customer is frustrated and the answer is not obvious. 

The design has to reach beyond explanation. It has to help people practise the behaviour the organization needs.

Speed is useful only after the problem is clear

AI can save time once the problem is clear. It can help a team build faster after the team has defined the workplace moment, the learner group, the target behaviour, the constraints, and the evidence of success. 

Without that clarity, speed becomes a liability. It allows teams to create more material around a poorly framed problem. 

For example, “improve sales confidence” is not a clear enough design target. Confidence in what situation? With which buyer? At what point in the conversation? What does confident performance look like? What should the learner say or do differently? What feedback would help them improve?

Those questions matter because they turn a broad training topic into a usable design brief. Once the brief is strong, AI can help. Before that, it may only accelerate ambiguity. 

The training that looks efficient but changes little

There are a few common ways AI can make weak training easier to produce:

• It can turn every business request into a course instead of questioning whether training is the right solution.
• It can generate generic examples that sound realistic but do not reflect the actual sales process, customer expectations, or manager behaviours.
• It can create role plays that feel conversational but do not elicit the specific competency the organization wants to build.
• It can produce quizzes that check recall while the real problem is judgment or application.
• It can make old content look new without improving transfer to the job. 

None of these problems are caused by AI alone. They are existing L&D problems with a faster production engine behind them.

The role play example shows the difference

Role play is a useful example because the gap between generic and well-designed practice is easy to see. 

A weak AI-generated role play might ask a learner to “handle an upset customer” or “sell a product to a hesitant buyer.” That may sound fine, but it is too broad. It does not tell the designer or learner which skill is being developed. 

A better role play is built around a precise behaviour. The scenario might be designed to test whether a salesperson can ask a follow-up question after a vague objection, link the response to the buyer’s business need, and avoid discounting too early. The customer persona, prompts, scoring, and feedback are all built to elicit that exact behaviour. 

To the learner, the practice feels natural. From a design standpoint, it is intentional. 

That is the difference between using AI to create activity and using AI to support skill growth.

What I would put in place before scaling AI training

Before scaling AI-generated or AI-assisted training, I would put a simple quality filter in place. It does not need to be complicated. It does need to be non-negotiable.

 Every program should be able to answer these questions: 

1. What business problem is this meant to support? 

2. What workplace moment are we designing for? 

3. What observable behaviour should improve? 

4. Where will learners practise that behaviour? 

5. What feedback will they receive? 

6. How will we know whether the skill improved? 

7. What human review is required before launch? 

If those answers are missing, AI should not be used to move faster. The team should slow down long enough to define the work.

A better way to use AI in L&D

The better use of AI is not to produce more training. It is to make good training easier to design, personalize, and scale. 

That means AI can help draft scenario variations, create practice prompts, summarize feedback, identify patterns in learner performance, and support faster iteration. It can help L&D teams move from one-size-fits-all content toward more targeted practice. 

But the human role remains central. Someone has to decide what matters. Someone has to challenge whether the output is accurate and relevant. Someone has to protect the learning experience from becoming a pile of attractive but low-impact content. 

AI can make L&D faster. The question is whether it will also make it better. That depends less on the tool and more on the discipline of the team using it. 

Soft CTA: Beyond Role Plays uses AI where it helps and learning expertise where it matters. We create focused role plays, scoring, feedback, and rollout support so practice is tied to the exact behaviours an organization wants to build.