Why Would Anyone Pay for AI-Generated Content? 348 Survey Responses Have the Answer

A while back, I had dinner with a friend who works in consumer goods.

He was bursting to show off: their product-seeding posts (short recommendation posts designed to spark the urge to buy) now come out dozens a day. Images drawn by AI, copy written by AI, even the layout done by AI. What used to take a content team a whole month, he now knocks out in a week with two AI tools.

I said, congrats—you beat the cost problem.

He waved it off: his boss was more anxious than ever. Output was up, sure, but nobody could promise that all this AI-generated stuff was something users genuinely liked—rather than something they'd see through at a glance, swipe away, and block.

My goodness, this problem is as classic as it gets.

Since ChatGPT appeared at the end of 2022, AI-generated content—AIGC, as the industry calls it—has practically rebuilt the entire content production pipeline. Writing copy, drawing pictures, cutting videos: it handles everything, fast and cheap. Even a giant like McDonald's has already folded AI tools into its social media strategy. And Ogilvy has said that many functions of marketing can now be handed over to AI.

The stronger the tools get, the more fundamental the question:

Can AI-generated content actually influence consumer decisions? And what is it that makes the difference?

People have been arguing about this for centuries... okay, for a few years now, and everyone just talks past everyone else.

The good news is that in 2025, a research team took the whole question apart, properly. They ran surveys, crunched the data, cross-validated with a neural network, and came back with a remarkably clear answer.

I've been through this study several times, and it gets more interesting with every pass. Today, I'll walk you through it in plain language.

First, an Old Model from 50 Years Ago

Where to begin? Let's start in 1974.

That year, two psychologists proposed a model called SOR: Stimulus, Organism, Response.

What's a stimulus? Something that happens out there in the world. Say you're scrolling your phone and a post crosses your feed.

What's the organism? What happens inside you. A little voice mutters: ooh, this is kind of interesting. Or: here comes another scam.

What's the response? The action you take. Place an order, or swipe away.

It's that simple. Run marketing through this model and it turns transparent: everything a brand does is the "stimulus"; the scale weighing things in the user's mind is the "organism"; whether they buy is the "response."

So here comes the question: what kind of stimulus does a piece of AI-generated content need to tip that inner scale toward "buy"?

That is exactly what this study set out to answer.

How Did They Study It?

First, the method—and it's solid.

From February to March 2025, the team distributed 400 questionnaires, then carefully cleaned the returns, cutting 52 careless ones, leaving 348 valid responses.

Each respondent first looked at two types of content—human-written and AI-generated—worked out which was which, and scored them.

Then they broke the content down into 6 features and tested them one by one: do these features actually make users feel something is "worth it," feel they can "trust" it, and finally want to "buy"?

Which 6? Jot them down—everything that follows revolves around them:

Fun to consume, easy to talk back to, on-trend, gets you, vouched for by others, and good-looking.

In proper research terms: entertainment value, interactivity, trendiness, customization, electronic word of mouth (eWOM), and visual aesthetics.

By the way, this framework wasn't pulled out of thin air. Marketing scholars had already distilled the first five dimensions back in 2012. This study added the sixth: good looks. Why add it? Because generating beautiful visual content is exactly what AI is best at—and it's precisely what past research most overlooked.

To make sure the findings weren't a fluke, they used two completely different methods: one called structural equation modeling (SEM), the traditional statistics route; the other a neural network, the machine-learning route. The results computed by the two methods were nearly identical.

This is no longer "I reckon." This is the data talking.

The Results: Five Pass, One Upset

14 hypotheses, 13 held up.

Let's start with the ones that held.

Fun is the number-one driver of "worth it." Among all the factors influencing perceived value, entertainment value ranks first in importance, at roughly 86%. The content has to be fun before users decide it's interesting—and that the brand means it.

Talking back works too. Content you can comment on, respond to, and take part in makes people feel it's worth it more easily than one-way broadcasting.

Good looks really do work. This newly added sixth dimension lifts both "perceived value" and "trust." The first instant a user scrolls past your content, the impression score has already been given. Looks are the content's front door.

Word of mouth is the number-one driver of "trust or not." This one is the most interesting. Among all the factors influencing trust, first place goes to electronic word of mouth—again at roughly 86% importance. However polished your content is, it can't beat one honest sentence from another user. Think about your own habits: heading to a restaurant you've never been to, what do you check first? The bad reviews. What other people say beats what the seller says—a hundred times over.

Finally, what pushes people to the checkout counter is "worth it." Of the two factors influencing purchase intention, perceived value has a predictive importance as high as 99%, leaving trust far behind. Trust is the foundation; whether it's worth it is the final kick that scores.

All of the above was expected.

The really interesting part is the one that didn't hold.

The Algorithm Gets You, but You Won't Trust It for That

The research hypothesis: AI-personalized content will increase users' trust.

The data says: it doesn't hold.

Personalization can lift perceived value. Users feel "this recommendation really gets me," feel it's worth it. But it can't buy trust. Not a shred of it.

Why?

The study's explanation: AI's personalization is "technical precision." What it lacks is "emotional care."

What's technical precision? The algorithm has watched your 100 clicks and pushes you the 101st thing you might like. Fast and accurate.

What's emotional care? Your mom remembers you don't eat cilantro. Your old friend knows what you've been wrestling with lately.

The gap between the two is the entire human touch.

The algorithm can say "I get you." But you know perfectly well that what it understands is your data, not you as a person. Being read correctly by data and being cared about by a person are two completely different feelings. The former makes it pleasant to use; the latter is what makes you willing to actually trust.

Put bluntly: precision can buy efficiency, but it can't buy hearts.

So, How Should You Use It?

The study offers several practical recommendations, and I think every one of them is worth stealing.

1. Set the ratio: AI 7, human 3. For everyday content, let AI run 70% of the volume to keep efficiency, and leave 30% to humans to keep the warmth. Note that this is for everyday content. In high-stakes industries like healthcare and finance, the human touch needs to be dialed up even more—don't let users feel that what's sitting across from them is a robot.

2. Label it, don't pretend it's human. When you publish AI-generated content, honestly mark it "AI-generated." Sneaking around is the real trust-killer; transparency actually earns you points.

3. Serve different dishes to different platforms. On looks-driven platforms like Xiaohongshu and Douyin (China's TikTok), good-looking AI-generated content is right in its element; on places like Weibo and Zhihu that run on information density, add more human review. One semantic mistake, and the comment section will roast you alive.

To sum up this study in one sentence:

AI owns "worth it." Humans own "trust."

Finally, Back to My Friend

Later I met up with him for another chat and walked him through what those 348 survey responses revealed.

He thought for a moment and said: no wonder. We poured all our effort into "volume," believing that more content meant winning. But users only have two questions in their heads: what's in this for me? And why should I trust you?

The first question, AI can answer for you, fast and well.

The second question, AI cannot answer for you. That takes real word of mouth from real users, genuine caring, and sincerity accumulated bit by bit.

Oh, and one more thing—the research team was honest about its own limits. They said this survey was conducted in China, and the conclusions may not transfer directly to other markets.

But that direction? I believe in it.

May your content carry both AI's sharpness and a human's warmth.