NEW RESEARCH: AI vs NARRATIVE STRUCTURE

Have you ever wondered why some stories feel more human than others?

NEW RESEARCH from University of Maryland analysed over 61,000 stories written by humans and five leading AI models, extracting 304 narrative features relating to plot, character, time, agency, revelation, structure, and storytelling decisions.

The study explored the nuances between stories written by AI vs stories written by humans.

Their key finding is fascinating: The biggest difference between human and AI storytelling is not style or syntax. It’s narrative structure. 

Most people think AI writing is detectable because of words, phrases, em dashes, clichés, or sentence structure – but this paper discovered that even when you remove those stylistic clues, AI stories are still easy to identify because they make different storytelling decisions.

Narrative features alone identified human versus AI stories with 93.2% accuracy.  

In other words: How a story is constructed matters more than how it is written.

The study explores one of the biggest weakness of AI storytelling: AI over-explains.

The researchers found AI stories:

• Explain their themes too explicitly.

• Spell out lessons.

• Tie everything together neatly.

• Resolve conflicts cleanly.

• Moralise more often.

• Remove ambiguity.   

Humans on the other hand are more comfortable leaving gaps. In my experience, confident presenters who trust their audience allow for uncertainty, and they leave room for interpretation. They're happy thinking on their feet and engaging with their audience - not relying on everything being perfectly packaged beforehand.

In a masterclass I watched with Aaron Sorkin, he once challenged the class to not treat their audience like idiots. Basically, allow them to figure out somethings for themselves. The reward (and the payoff) is great when the audience works some things out for themselves. Ernest Hemingway had a whole philosophy around it with his “Iceberg Theory”. 

The audience should connect the dots themselves, but AI wants to explain everything and draw the dots for them.

Human stories are messier

AI likes:

• Straight lines and linear stories.

• Clear cause and effect.

• Single-track narratives.

• Few subplots.

• Predictable resolutions.

Humans prefer:

• Flashbacks.

• Time jumps.

• Multiple threads..

• Ambiguous endings.

• Contradictions.

• Loose ends  .

Even though this research focused on works of fiction, we can think about these findings in the context of business storytelling and executive comms. Too many business presentations are becoming AI-like because they are too linear, too neat, and too rational.

The most memorable human leaders often do the opposite. I’m thinking of business leaders like Indra Nooyi, Steve Jobs, Jensen Huang, Fei-Fei Li, even Warren Buffett. They wander a little, they tell side stories, they create tension, they talk about obstacles and challenges. And they ask good (rhetorical) questions. They show empathy. They leave space for audiences to think...

Human stories contain more ambiguity because people are complicated. We often want two things at the same time, hold conflicting beliefs, or act emotionally rather than logically.

For example: a business leader might say they want innovation, but also fear the risk that comes with change. That tension creates ambiguity, and it’s often where the most interesting and human stories begin, but AI characters in the study tended to be cleaner and more predictable.

My advice for business storytellers: Stop trying to make leaders look perfect.

The most compelling executive stories often involve doubt, failure, contradiction, uncertainty, learning. Audiences trust imperfect humans. Imperfect stories are more believable - you don't want the perfectly written AI post, you want the one I wrote with typos, bad grammer ;) and punctuation in the wrong place.

Another important insight I picked up from the research paper was that human storytellers talk to the audience. We’re much more likely to ‘break the fourth wall’ by addressing the reader directly. Referring to shared experiences.

The very best keynote speakers don’t just deliver information, they create a relationship.

“Our jobs [as performers] is to seek truth, build trust [with our audience], and perform a service.” -Yo-Yo Ma

AI might say, “Here’s what I want to tell you.”

A good business storyteller is more likely to say, “Let’s think about this together.”

The paper concludes that AI tends to converge toward a shared narrative centre while humans explore the edges.

Great storytelling is partly about navigating to places that most people don’t go.

Novel combinations and unexpected bearings, or taking the audience on a different route but though familiar territory.

For example: many AI presentations start with something like, “AI is changing the world – faster than any of us want or expect...”

Everyone expects that.

What about, “The most valuable skill in the age of AI might be listening.”

Same destination. Different route.

Or what about a dry topic like cyber security? “The threat landscape is evolving.”

No.

How about “The most dangerous day for a security team isn’t the day they’re attacked. It’s the day they think they’re safe. Cybersecurity failures rarely begin with a hacker. They begin with overconfidence...”

So at the end of the day, what might this study mean for business storytellers?

I'm glad you asked.

My main takeaway's from this study:

• Stop over-explaining.

• Trust your audience more.

• Leave room for interpretation.

• Not every lesson needs to be stated.

• Use more ambiguity.

• Real life is messy.

• Show contradictions.

• Perfect heroes are forgettable.

• Talk with audiences, not at them.

• Invite participation.

• Use more narrative complexity.

• Flashbacks, callbacks, parallel stories, tension.

• Resist tidy endings.

• Great stories often leave something unresolved.

• Aim for originality of thought, not originality of wording.

• The structure matters more than the sentences.

I can’t help but keep coming back to the idea that AI is becoming increasingly good at writing. Better (probably) in 80% of day-to-day business use cases.

But humans are still better at wondering. And we demonstate empathy in ways which machines arguably will never be able to do.

The paper suggests that what makes human storytelling special isn’t vocabulary or grammar - it’s our willingness to embrace uncertainty, contradiction, ambiguity, and complexity (even if we don’t know the answers).

--

Thoughts?

So 41% of self-published books on Amazon in 2025 were written by AI and most of the authors DIDN'T self-disclose? 🤯

That's up 20% year-on-year from 2024 and it's only going to get worse. (Sample of 14,000 books analysed).

In some tests, only 3% of GPT 5.4 written scripts were accurately flagged as being generated by AI. (GPT 5.4 was fine-tuned to mimic human writing more accurately. Previously 97% of scripts were identified as being generated by AI. That's an insane improvement in the model.)

➡️ But if you analyse the STORY STRUCTURE and not the WORDS, 93.2% of scripts can be identified as being written by AI.

Takeaway: How a story is structured matters more than how it is written?

⬇️

📝 Just because you can doesn't mean that you should.

--

It's a fascinating study worth reading. The research is only 10 pages long (the other 20 pages are sources). You can either read it online here or download the PDF here.