Someone asked me recently whether you can teach discernment to people who use AI.
Not whether you can teach prompting. Not whether you can teach workflows or shortcuts or which model to use for which task. Specifically discernment. The ability to look at what the AI produced and know whether it's good.
I've been thinking about this for weeks. And I have a conclusion, but it's not the one most people want to hear.
The Problem Nobody's Talking About
There are a lot of people right now producing a lot of AI-assisted content. Some of it is good. A lot of it is what I'd call slop. Technically correct, structurally sound, and completely lifeless. The kind of thing that reads as AI the moment you hit the second paragraph.
Most of the people producing slop don't know they're producing slop.
That's the core problem. Not that people are using AI. Not that the tools aren't capable of producing good output. It's that a meaningful percentage of people using these tools don't have enough exposure to what good actually looks like to recognize when they're not producing it.
They accept the first draft. They publish it. They wonder why it doesn't land.
And the answer isn't that they need a better prompt. The answer is that they need to have read more things. Seen more examples of what good writing looks like. Developed a sensitivity to the difference between something that actually sounds like a human and something that sounds like a language model doing an impression of one.
That sensitivity is what I mean by discernment. And you cannot shortcut it.
Can You Teach It?
This is the question that's been sitting with me. And I think the honest answer is: not directly.
You can teach someone what to look for. The AI tells. The em dashes that appear too frequently. The adjective triads. The paragraph that starts with "It's worth noting that." The sentence that technically says something but doesn't actually mean anything. You can give someone a checklist and tell them to audit their output against it.
But that's not discernment. That's pattern matching against a list of known tells. It's useful, but it's a different skill.
Actual discernment is when you read something and the response is immediate and physical. Something's off. I can't tell you exactly which rule it's breaking, but I know this isn't right. That comes from exposure. From having read thousands of pieces of writing. From having written a lot yourself. From having had your own work critiqued enough times that the standards are now internal rather than external.
You can give someone a fishing rod. What you can't give them is ten years of fishing.
The 10,000 Reps Problem
There's a phrase I keep coming back to: taste is reps.
The people with the best taste in any domain, writers, designers, photographers, editors, have almost universally put in the volume. They've looked at a lot of bad work. They've produced a lot of bad work. They've seen what good looks like from close up, often by working alongside people who had better taste than them and being corrected until the correction became automatic.
You cannot download that. There's no prompt that installs it. There's no course that compresses it to an afternoon.
What you can do is accelerate the exposure. Read more. Read widely. Specifically read things that are widely regarded as excellent in the category you're working in. If you're producing marketing copy, read the best marketing copy that exists. If you're writing essays, read people who write essays well. If you're building educational content, study the teachers who make hard things feel simple.
The AI tools are useful precisely because they produce so much output. But the person sorting through that output, deciding what to keep and what to throw away, what to publish and what to rewrite, needs the taste to make those calls. And the taste comes from the reps.
What This Means for How You Use AI
The people who are getting the most out of these tools right now are not the ones who prompt best. They're the ones who can tell, quickly and reliably, when the output is good enough to use and when it needs work.
That's a skill that compounds over time. Every time you read something good, your bar moves. Every time you catch something off in AI output and fix it, you're training your own sensitivity. Every time you publish something and then look back at it six months later and cringe, you're raising your floor.
It's slow. It doesn't respond to urgency. You can't hustle your way into taste.
But here is the part worth sitting with: most people are not doing this work. Most people are accepting the first draft, or skipping the audit entirely, or not reading enough quality input to know what they're comparing against. That means the gap between people with real discernment and people without it is growing as the tools get better, not shrinking.
The output quality ceiling for someone without discernment is roughly the average quality of AI training data. The ceiling for someone with discernment is whatever they've trained their taste to recognize as excellent.
Those are very different ceilings.
The One Thing That Actually Helps
If I had to give you one concrete thing: go make a thousand pieces of content.
Not all of them good. Not all of them with AI. Just make them. Post them. Get feedback. Look at what lands and what doesn't. Develop the sensitivity that only comes from doing the thing repeatedly.
Call it taste. Call it discernment. Call it whatever you want.
But know that there's no shortcut to it, and the people who have it are going to be the ones who matter in five years when everyone has access to the same tools and the differentiator is purely the judgment of the person using them.