What Is It About AI-Generated Content That Actually Moves Consumers?

The other day I was scrolling through Xiaohongshu（RED, China's Instagram-like lifestyle platform）and a seeding post（product-recommendation post）caught my eye. The visuals were gorgeous, the copy was sharp — it was about a portable coffee machine. I almost placed an order.

It wasn't until I opened the comments that I realized the whole thing was AI-generated.

I froze for a second. Not out of anger at being tricked, but because a question popped into my head: why was I about to pay?

Was it the pretty pictures? The copy that hit a nerve? Or the chorus of praise in the comments? Which element actually pushed me from "seeing" to "wanting to buy"?

I'm not the only one curious about this. A group of researchers took it seriously — they used 348 survey responses to take this question apart with real precision.

What Do We Even Mean by "Moves Consumers"?

Let's get one thing straight first.

Between a consumer seeing a piece of marketing content and finally clicking "buy," there is no single straight line. There's a classic framework called SOR — Stimulus, Organism, Response.

In plain English: something external stimulates you (the content), something happens inside you (it feels worth it, it feels trustworthy), and finally you act (you place the order).

For that middle step — what happens inside — the researchers locked onto two key variables:

Perceived value — do you think this thing is worth it?

Trust — do you think this brand / content is credible?

These two are the bridge between "looking" and "buying."

So here's the question: across which dimensions can AI-generated content actually build those two bridges?

Six Keys

Earlier research had already found that social-media marketing content works (or doesn't) along five main dimensions: entertainment, interactivity, trendiness, customization, and electronic word-of-mouth.

But this study added a sixth: aesthetics.

Why add it? The logic is simple. What is AI best at? Making images. Generating things that look beautiful. The old framework was built before the AI-image explosion — without this sixth key, the puzzle is incomplete.

Let me break all six down for you.

Entertainment — is the content fun? Can it make you smile?

Interactivity — can you comment, share, talk back to the brand? It's not one-way broadcasting.

Trendiness — does the content ride the current moment? Does it feel "in step"?

Customization — does AI use your preferences to serve up "exclusive" content?

Electronic word-of-mouth — what are others saying? Are the comments good? Is anyone out there recommending it?

Aesthetics — is it nice to look at? Is the visual design refined? Does the imagery have texture?

All six keys are on the table. But they don't pull with equal force. Some open the lock faster than others. Which one is the master key?

The Answer from 348 People

The researchers used two methods to crunch the numbers. The first was partial least squares structural equation modeling (PLS-SEM), which excels at linear relationships. The second was an artificial neural network (ANN), which is good at capturing non-linear relationships. Two rulers cross-measuring, validating each other.

From February to March 2025, they distributed 400 surveys online via Wenjuanxing（a Chinese online survey platform）, and after screening ended up with 348 valid responses — an 87% effective response rate.

What did those 348 people tell us?

First, Perceived Value: Who's Pulling?

Entertainment pulls the hardest.

Path coefficient β = 0.176, ANN importance 86.28%. Both methods rank it first.

Think about it — when you're scrolling on your phone, what makes you stop? A dry spec sheet, or a clip that makes you laugh out loud? Definitely the latter. Entertainment triggers an emotional response, that emotional response warms you to the brand, and that warmth converts into "this thing is worth it."

Interactivity ranks second. β = 0.169. Interaction isn't just likes and comments — it turns you from a bystander into a participant. Once participation kicks in, you naturally feel this thing relates to you, that it's worth it.

Aesthetics ranks third. β = 0.140. What AI does best is generate refined visuals. A beautiful product photo — two glances and you already think "premium." That premium feel directly lifts perceived value.

Interesting, isn't it. In the world of AI-generated content, looking good is more persuasive than being useful when it comes to getting people to open their wallets.

Now Trust: Who's Building It?

The trust ranking is a completely different lineup.

Electronic word-of-mouth (EWOM) pulls the hardest. β = 0.192, ANN importance 85.94%.

In plain terms: if others say it's good, I believe it.

Think about it yourself. You see a polished AI-generated seeding post on Xiaohongshu — do you believe it right away? Probably not. You scroll down to the comments. If eight out of ten comments say "it really works," your defenses start to crack.

That's social proof. Word-of-mouth is the entry ticket to trust.

Trendiness ranks second. β = 0.164. When a piece of content lands on the current moment, you feel the brand is "alive in the present," not yesterday's news. That sense of being current is itself a signal: they're taking this seriously.

Aesthetics ranks third. β = 0.113. Visual texture signals professionalism. A brand whose visuals look rough — it's hard to believe it cares much about product quality. Flip it around: refined visuals themselves are saying "we mean business."

The One Surprise: Customization Doesn't Build Trust

Out of 14 hypotheses, 13 passed validation. Only 1 was rejected.

Customization's effect on trust is not significant. β = 0.072, p = 0.212.

This is the most interesting finding in the entire study.

Why? The researchers puzzled over it themselves. AI's customization power is enormous — algorithms can use your browsing history and purchase preferences to generate content "for you." But when consumers receive that content, it doesn't earn the brand any extra trust.

Why do you think that is?

Because "technical customization" and "emotional resonance" are two different things.

AI knows you like looking at coffee, so it serves you coffee. But it doesn't understand why you like coffee. Is it the morning caffeine fix? Or the memory of that one cup you had in Paris? The algorithm's precision is data precision, not emotional precision. And consumers can tell.

So customization can make you feel "this content is useful to me" (perceived value holds up), but it can't make you feel "this brand gets me" (trust doesn't get built).

This finding hits a blind spot a lot of people have. Marketers often assume: the more precise the AI customization, the better. Precision is genuinely useful — but don't count on it to build trust for you.

The Final Leg: Who Pushes You to Checkout?

Perceived value and trust have built the bridge — what gives you the final shove across?

Perceived value.

β = 0.344. The highest of all path coefficients. ANN importance 99.19% — almost a perfect score.

Consumers ultimately pay for "is it worth it," not for "do I trust them."

Trust matters too (β = 0.237), but its force is only about 70% of perceived value's.

What does that mean? It means even if a consumer is half-doubtful about the brand, as long as the content makes them feel "this thing is genuinely worth it," they'll still place the order. Conversely, trust alone without value transmission won't lift conversion rates.

What Does This Mean for Marketers?

The answers from 348 people, cross-validated by two algorithms, point to several very concrete actions.

First, the most important one: treat entertainment as your number-one productive force. Believe it or not, the data says entertainment's pull on perceived value is the strongest of all six dimensions. If your AI-generated content is merely "correct" but not "fun," you're leaving more than half its power on the table. Creative that makes people laugh, that stirs emotion — that's the secret weapon of AI marketing content.

Word-of-mouth? Word-of-mouth is the entry ticket to trust — don't treat it as an operations afterthought. EWOM's effect on trust far outstrips every other dimension. In an era when AI content is flying everywhere, the comments section matters more than the body of the post. One genuine user review is worth ten AI-generated copies. Rather than spending money on ever-more-polished AI generation, spend it on guiding real user feedback.

Don't skimp on visual precision either. Aesthetics — the sixth dimension this study added — is significantly and positively correlated with both perceived value and trust. AI's greatest strength is image-making; failing to leverage it is pure waste. On visually driven platforms like Xiaohongshu and Douyin（China's TikTok）, one polished image beats a thousand words of copy.

But don't count on customization to build trust. Precise AI recommendation is a good thing — it lifts perceived value. If you assume precise recommendation will make users trust the brand, the data says you'll be disappointed. Trust still has to come from word-of-mouth, from interaction, from genuine emotional connection.

The researchers also offered a practical ratio for everyday campaigns: use AI to generate 70% of content, with 30% created by humans — balancing efficiency and quality. But in sensitive sectors like healthcare and finance, the human share should go up. Brand warmth and credibility matter far more than efficiency in those settings.

Platform differences matter too. Xiaohongshu and Douyin are visually driven — AI-generated polished images and videos thrive there. Weibo（China's Twitter-like platform）and Zhihu（a Quora-like Q&A community）reward interaction and information density; AI-generated long-form needs human semantic review to reduce the risk of backfiring.

One last thing: when you should label, label. Proactively marking "AI-generated content" sounds like it would reduce appeal, but it's actually building trust. Transparent source-labeling is a signal of brand honesty in the AI era. Getting caught hiding it — the cost of a trust collapse is far greater than the cost of coming clean.

Back to That Post That Almost Made Me Buy

On my way home from work that day, I opened Xiaohongshu again and glanced at that AI coffee-machine seeding post.

The visuals really were pretty. The copy really was fun. The comments really did have people praising it.

It had assembled the three strongest keys. No wonder I almost bought.

But in the end I didn't — because I paged down to the fifth screen of comments and saw someone write, "the real thing looks nothing like the photos."

Word-of-mouth can build trust, and word-of-mouth can also destroy it.

That's the real predicament of AI marketing content: AI can help you get the content to 90 points, but whether the consumer ultimately trusts it and buys comes down to the last 10. And those last 10 are usually not in AI's hands.

What 348 people told us through their surveys boils down to one line:

AI's job is to make you stop and look. But what makes you open your wallet is always those three little words: is it worth it.