How to choose an AI tool, without chasing a moving target
A teacher asked me this during Q&A after this morning's talk: "Are there resources for knowing which AI tool to use for which purpose?" Here's my honest answer, and why I'm not giving you a list of app names.
The tempting answer is a chart: use this tool for essays, that one for lesson plans, this other one for data analysis. I'm not going to give you that chart, and here's why: whatever list I write today will be wrong in six months. Tools merge, rebrand, get better or worse overnight. A list is obsolete before most people even finish reading it.
So instead of a list, here's a framework, one you can use on any tool that exists today, or any tool that gets released next year. For each use case, ask the same handful of questions before you trust the output.
The framework: ask, verify, decide
- Can I see multiple distinct options, not just one polished answer?
- Does it help me think of angles I hadn't considered, or just restate the obvious?
- Am I using it to generate a draft I'll heavily rework, or to skip the thinking entirely?
- Does it show its sources, and can I actually click through and verify them?
- Does it distinguish between what's well-established and what's contested or uncertain?
- What happens when I ask it a question it doesn't actually know the answer to, does it say so, or does it guess confidently?
- Can it show its work, the actual calculation or method, not just a final number?
- Would a student who used this be able to explain how the answer was reached?
- Does it flag when the data is incomplete or the sample is too small to trust?
- Does it explain what the code does, or just hand over a block to paste and run?
- Can the student trace every line back to something they understand?
- Does it acknowledge when it's uncertain about a library, version, or edge case?
- Is the feedback specific to this student's actual work, or generic advice that could apply to anyone?
- Does it explain why something is a strength or weakness, not just label it?
- Who sees the student data this tool collects, and how long is it kept?
The one red flag that matters most
Across every category, there's a single tell that outranks all the others: does the tool ever say "I don't know" or "I'm not sure"? A tool that always sounds confident, regardless of the question, is optimizing for how it sounds, not for being right. That's the tool to be most careful with, no matter what category it's in.
That's the whole framework. Not a list to memorize, a set of questions to ask every time, about any tool, forever.