I was reading a thread on the slack channel of the Learning Network (join – it’s a great community) about which tools people use and how people find them.

My two penn’orth that I added:

This is a really useful thread.

I’d add three things that often get missed when people say “we tried Claude / Copilot / ChatGPT and weren’t impressed”.

First, the environment matters. Free or basic access is usually not the same as working in a properly configured paid environment, especially if you need file handling, projects, memory/custom instructions, stronger models, privacy controls, or repeatable workflows.

Second, the tool matters. Claude, ChatGPT, Copilot, Gemini and Perplexity are not interchangeable. I use them for different jobs. Perplexity is useful for research and source discovery. Claude and ChatGPT are useful for drafting, analysis, structured thinking and working with larger bodies of material. Copilot can be useful inside the Microsoft environment, but it depends heavily on the organisational setup and permissions.

Third, the set-up matters as much as, if not more than, the prompt. The best results usually come when the tool has clear instructions about who you are, what kind of work you do, what good output looks like, what it should avoid, and what standards it should apply. That might be through custom instructions, projects, uploaded reference materials, examples of previous work, rubrics, assessment principles, or quality criteria.

So yes, giving an AI a storyboard and asking it to write course questions is unlikely to produce much of value. But that may be less a failure of the tool and more a weak use case set-up.

For assessment questions, I’d want the tool to understand the performance outcomes, assessment purpose, level, audience, context, common misconceptions, question-writing rules, answer feedback requirements, and any platform constraints. Even then, I’d treat the output as a rough first pass, not as finished assessment design.

AI is not a vending machine for learning content. It is closer to a capable but overconfident intern. It needs context, constraints, standards and review from someone who already knows what good looks like.

What are your thoughts? Am I on point, or spouting rubbish?