The average is now free: What students need to know about AI
AI can write your email. Build your presentation. Generate your ad campaign. Summarize your reading. And, increasingly, produce an endless supply of perfectly adequate content.
There’s even a term for the worst of it: “AI slop.”
Anuj Kapoor has been studying what happens when artificial intelligence meets human behavior. An assistant professor of marketing at the University of Missouri’s Robert J. Trulaske, Sr. College of Business, Kapoor’s research interests include digital video advertising, AI-human interaction and digital platforms.
In a recent field experiment involving more than 21,000 consumers, Kapoor and his fellow researchers found that personalized video advertisements created with generative AI generated greater engagement than more traditional digital advertising formats.
His expertise also comes at an important moment for Trulaske students. Transformative Technologies is one of the college’s three strategic pillars, reflecting a commitment to preparing students not only to use rapidly evolving tools such as AI, but also to understand how those technologies are reshaping business.
That work is playing out in Kapoor’s classrooms. This fall, he is teaching two sections of Social Media Marketing as well as Python for Marketing Analytics. In both, students are learning to use AI without surrendering the fundamentals that allow them to determine whether its output is actually good.
That raises a bigger question: If everyone has access to the same powerful tools, what makes you valuable?
We asked Kapoor about AI slop, good judgment and why struggling with a blank page might be more important than students realize.
As AI-generated content becomes easier and cheaper to produce, does distinctly human work become more valuable?
Yes, and I would put it more strongly.
Picture a machine that can draw a thousand clean, pretty pictures in a minute. Once that exists, an OK picture is worth almost nothing because everyone has the machine and all the pictures start to look the same. What people still pay for is the picture only you would have thought of.
The model gives you the average good answer, so the average is now free. Everything that stands out has to come from the part the model cannot reach, which is the part that is actually you: a real point of view or a detail about a specific customer that was never sitting in the training data.
In my world, a thousand brands can generate the same on-trend caption by lunchtime. The ones that are successful sound like a person who has actually met the customer and is willing to be a little sharp about it.
The machine can copy what already exists. It cannot be you.
If every graduating student can use ChatGPT, Claude and other AI tools, what will distinguish the students employers really want?
Imagine every student is handed the same calculator. Being able to press the buttons does not make anyone special because everyone can press the buttons.
What makes a student valuable is knowing when the answer on the screen is silly. If the calculator says one slice of pizza costs a million dollars, you want to be the person who says that is wrong.
AI is like that. It will give you a confident answer, and the rare skill is catching the moment it is confidently wrong. That takes real knowledge of the business, which a prompt cannot hand you.
The students who can do the underlying work without AI end up being the best at using it because they can see where it went off the rails. A student who never learned to write cannot tell good writing from AI slop, so the tool quietly makes the work worse while they think it is making it better.
How are you putting that philosophy into practice with Trulaske students?
This fall I’m teaching Social Media Marketing and Python for Marketing Analytics, and the two make a nice contrast because one is about words and ideas and the other is about code and data. AI is reshaping both.
In Social Media Marketing, I lean straight into the slop problem. Students use the tools to generate content, and then the real work begins, which is judging it. We pull apart why so much of what the model produces is fluent and forgettable, and what it takes to turn a generic post into something a real customer would actually stop for.
The point is not to ban the tool. It is to make them the editor rather than the typist, the person with the taste to know when the output is good and the nerve to throw it out when it is not.
In Python for Marketing Analytics, the tools will now write a lot of the code for you. So I make students earn the fundamentals first, before they lean on an assistant. Once they understand what the code is actually doing, AI makes them much faster. If they skip that step, they cannot tell when the analysis is quietly wrong, and in analytics a confident wrong answer is the dangerous kind.
Struggle first, then accelerate. That is the rule in both courses.
You’ve said, “The judgment is ours.” What does good judgment look like when working with AI?
The tricky thing about these systems is that they always sound sure, even when they are inventing. Good judgment is being the person who can tell a confident wrong answer from a genuinely good one.
Think of a tall tower of Lego. One brick near the bottom is holding the whole thing up, and if you pull it, the tower comes down. The model cannot tell you which brick that is. It hands you the whole tower and treats every piece as equal.
Knowing which one actually matters is the judgment, and the model does not have it.
So how do students develop that judgment?
You get that judgment the way you get anything real: by doing the thing yourself enough times that you can feel it.
It is like tasting cake. You can only tell whether a cake came out right if you have baked and tasted enough of them to carry a sense of “good” in your head. There is no shortcut.
I tell my students to treat the model as a sparring partner rather than a ghostwriter. Argue with it. Make it prove its answer and check what it says against something real.
The ones who do that get sharper. The ones who paste and submit slowly lose the ability to notice when they are being handed nonsense.
Do students risk losing important skills if they rely too heavily on AI?
Yes, and the worry is specific. The skills most at risk are the boring, hard ones because those are exactly the ones students are quickest to hand off.
The blank page. The problem you cannot crack for 20 minutes. That stuck feeling is not a sign something has gone wrong. It is the exercise. It is the mental version of lifting something heavy, and the strain is the whole point.
Writing is the clearest case because writing is how you find out what you actually think. Hand it off and you never form the thought. You just receive one.
So the order matters. Do the heavy lifting yourself first, then use the tool to go faster.
The problem is not that students use AI. It is that many reach for it before they have built the muscle to judge what it gives back, and by then the muscle they skipped never grew.
What do today’s students need to understand about AI that they may not fully appreciate yet?
The first thing is uncomfortable. The model is basically everyone’s average blended together, so if your work sounds like the model, you have quietly told the world you are easy to swap out for it.
Be the part that is only you.
The second is that responsibility does not move to the tool. If the machine does the work and the work is wrong, it is still your name on the page and you are the one who gets asked about it. “The AI said so” has never once sounded good in a room that mattered.
And people assume AI is a great equalizer that lifts everyone the same, but it gives the biggest advantage to the people who already know what they are doing.
Simply knowing the standard answer will be worth very little because the machine will know it, too. What holds its value is the nerve to bet against the standard answer and the discipline to actually pull the bet off.
Those are two things the machine cannot copy.
Co-authored by Madhav Kumar. “Frontiers: Generative AI and personalized video advertisements” appeared in Marketing Science (2025).