Scout7 logo

Scout7

Back

research_paper

Better, And Just Like Everyone Else's: What AI Writing Costs Founder-Led Marketing

September 25, 2026 · 9 min read · Scout7

300 writers got story ideas from AI. Their stories got better and more alike. Here is where the sameness comes from, and the part of a post to write yourself.

Better, And Just Like Everyone Else's: What AI Writing Costs Founder-Led Marketing

Three hundred people wrote short stories for a study. Some of them got story ideas from an AI. Those people wrote better stories. Their stories also came out more alike.

Both things happened at the same time. Better, and more like everyone else's.

Maybe you do founder-led marketing. That means you, the founder, do the marketing yourself. If you let an AI agent write your posts, this is about you.

What you will find out below:

  • who gains the most when an agent writes, and who gains nothing
  • why switching to another AI model does not fix it
  • the one part of a post you should never hand over, and why

The AI made the stories better. It also made them more alike

The study was done by Anil Doshi and Oliver Hauser. It was published in the journal Science Advances in July 2024.

Here is how it worked. Three hundred people each wrote a very short story. Some wrote alone. Some could ask ChatGPT for one starting idea. Some could ask for up to five ideas. Then six hundred other people read the stories and scored them.

The stories written with AI ideas scored higher. Readers found them better written and more fun to read.

Then the researchers compared the stories with each other. Stories by people who used one AI idea were 10.7 percent more alike than stories written alone. Each story got better. Together, they got more similar.

Now think about your own week. You build your product with Claude Code or Cursor. The same agent now writes your launch post and your blog. It saves you an hour, and the post reads well.

But many other builders ask the same kind of model for the same kind of post. A reader may see a post very like yours three times in one week. When yours arrives, it can feel like one they already read.

So what is going on here? And what can you do about it?

Where the sameness comes from, and what you will know by the end

Most advice on this says "add your own voice". That is not wrong. But it does not tell you what to do on a Tuesday morning with your agent open.

One study found where the sameness actually comes from. Once you know that, the fix takes about five minutes per post. We will get to it.

First, look again at the people in that story study. They did not all gain the same amount.

Who gains most when the agent writes, and who gains nothing?

Go back to the three hundred writers. Before they started, each one took a short test of how creative they were. So the researchers could see who the AI helped.

The writers who scored lowest on that test gained the most. With five AI ideas, their stories were judged up to 26.6 percent better written. They were up to 22.6 percent more enjoyable to read.

The writers who scored highest gained nothing the judges could see. Their stories did not get better with AI. But they still moved closer to everyone else.

This is the fair side of the story. If you find writing hard, the agent really helps. It gets you past the empty page. Your post comes out clearer than it would have.

If you already write well, the deal is worse. You give up some of what makes you different. You get nothing back.

Maybe the answer is just a better model. That is what most people try next.

Why switching to another model does not fix it

Many builders think the fix is to switch. Try a different model and you will sound different.

In March 2026, Emily Wenger and Yoed Kenett tested this. Their study is in the journal PNAS Nexus. They gave the same creativity tests to 102 people and to 22 different AI models.

One test was simple. Think of new uses for an everyday object, like a brick. Then the researchers measured how different the answers were from each other.

People's answers were spread far apart. The models' answers bunched close together. It was the same on all three tests, and across all 22 models.

A newer study, from August 2026, points to why. Alexandrine Fortier, Hazel Chen and Peter West found that the sameness shows up early in how these models are trained. It is not a setting you can switch off later.

So a different model may change a few words. It does not change the pull toward the middle.

And that pull has a direction. One study shows where it goes.

Why an Indian writer started to sound American

Dhruv Agarwal, Mor Naaman and Aditya Vashistha ran a test with 118 people. Some lived in India. Some lived in the United States. The study was shown at CHI 2025, a big meeting of computer researchers.

Everyone wrote about things from their own life and culture. Some wrote with AI suggestions popping up as they typed. Some wrote without.

With the suggestions, the Indian writers started to write more like the Americans. Not just the words. The way they told things changed too.

The Americans also got a bigger speed boost from the AI than the Indians did.

Think of a road with one wide lane in the middle. The AI keeps nudging you into that lane. That lane is the most common way of saying things.

If your customers chose you because you are different, that nudge costs you something.

This does not mean AI is no use for writing. There is one place where it clearly helps.

What does help: asking for ideas the right way

Lennart Meincke, Ethan Mollick and Christian Terwiesch work at Wharton, a business school in the United States. In 2024 they asked GPT-4 for product ideas. The ideas were for college students, and had to cost under 50 dollars.

First they asked in the normal way. The ideas were more like each other than ideas from a group of people.

Then they asked the model to think step by step before answering. The ideas became much more mixed. They came close to the group of people. You can read it here.

So the way you ask does matter. Ask for many ideas, step by step, and you get a wider list to pick from.

But look at what that study measured. It measured a list of ideas. It did not measure the finished post.

Your idea can be fresh. The post can still come out sounding like everyone else's. The next study shows why.

Whose words make every post sound the same?

Vishakh Padmakumar and He He at New York University ran a careful test. It was shown at ICLR 2024, a big meeting of AI researchers.

People wrote short essays that argue a point. Some wrote alone. Some wrote with an older AI model called GPT-3. Some wrote with InstructGPT, a newer model trained to follow instructions.

With InstructGPT, essays by different people became more alike. With the older GPT-3, they did not.

Then the researchers looked inside each essay. They split it into the sentences the person typed and the sentences the model wrote.

The sameness came from the model's sentences. The sentences people typed themselves stayed just as different from each other as when they wrote alone.

This is the key part. Your own sentences did not get pulled to the middle. Only the model's sentences did.

A second study shows the same thing from another side. Barrett Anderson, Jash Shah and Max Kreminski watched 36 people come up with ideas. ChatGPT gave similar ideas to different people. The people also felt the ideas were less their own.

So the sameness does not come from you. It comes from the sentences the model writes for you.

Founder-led marketing: what to write yourself, and what to hand to the agent

Here is the answer. Write the facts only you have yourself. Let the agent fix the order and the grammar, and nothing else.

This is how to do it for your next post:

  1. Before you open the agent, write three to five lines yourself. What happened. One number from your own work. A customer's exact words.
  2. Give those lines to the agent. Tell it to keep them word for word.
  3. If you need ideas, ask for many. Ask it to think step by step. Pick one yourself. Do not take the first.
  4. Read the draft. Mark every sentence anyone could have written. Change it to one only you could write, or delete it.
  5. Last check. Could another founder post this with their name on it? If yes, it is not done yet.

It takes about five minutes. The agent still does the slow part. You keep the part that makes the post yours.

What these studies cannot tell you

These studies are strong, but they have limits. Here they are.

  • The people wrote stories, essays and answers to tests. Nobody tested launch posts.
  • Computers measured how alike the writing was. Buyers did not.
  • No study here shows that a post that sounds like others sells less.
  • Some tests used older AI models. Newer ones may be a little different.
  • 10.7 percent more alike is a real change, but not a huge one.

So treat this as a good reason to change a habit. Do not treat it as proof that AI writing hurts sales.

Frequently asked questions

Should I stop using AI to write my posts?

No. For most people it makes the writing clearer. Just write your own facts first, and keep them.

Will a different AI model make me sound different?

Not much. Wenger and Kenett found that all 22 models they tested gave answers that bunched together.

What if I give the agent a long prompt about my style?

The studies point at the sentences the model writes. So the safest fix is to keep your own lines word for word.

Is this about Google rankings?

No. It is about a reader who feels they have read your post before.

Where can I read the studies?

When your agent writes a post for you, which part do you still write yourself?