Discover insights on psychology, emotions, and human connection – exploring the secret benefits of mindful living.

There is a strange new question hiding inside almost everything we read online:
Was this written by a person?
A few years ago, most of us probably never considered it. A story, an email, a dating profile, a heartfelt message, a product review — we assumed that somewhere behind the words was another human being.
That assumption is disappearing.
Generative AI can now produce essays, stories, messages, jokes, apologies, dating-app introductions, and deeply personal-sounding reflections in seconds. And while people often believe they can recognize AI writing when they see it, new research suggests that confidence may be misplaced.
A study published in August 2026 in Judgment and Decision Making asked a deceptively simple question: Can people tell the difference between a fictional story written by a human and one generated by AI?
The answer was uncomfortable.
For the most part, they could not.
And perhaps even more interestingly, people sometimes preferred the AI-generated stories — while simultaneously judging stories more favorably when they believed a human had written them.
That contradiction tells us something important not only about artificial intelligence, but about ourselves.
The Experiment: Human Story or ChatGPT?
Researchers Sydney Sears and Deena Skolnick Weisberg of Villanova University conducted three studies involving more than 2,500 adults.
In the first study, 1,682 participants from the United States read a roughly 1,000-word fictional story.
Some stories had been written by published human authors. Others had been generated using ChatGPT.
But there was another twist.
Participants were also told who supposedly wrote the story. Some received accurate information. Others were deliberately told that an AI-generated story had been written by a human, or that a human-written story had been generated by ChatGPT.
This allowed the researchers to separate two things that are usually mixed together:
the quality of the writing itself and our beliefs about who created it.
The results were striking.
Participants rated the AI-generated stories as more absorbing and, overall, higher in quality than the human-written stories.
Yet when people were told that a story had been written by a human, they tended to evaluate it more positively — regardless of who had actually written it.
In other words, people appeared to have two beliefs operating at the same time.
The first was:
“I like this story.”
The second was:
“I probably like it more if a human wrote it.”
That difference may become increasingly important as AI moves deeper into everyday communication.
When People Had to Guess, They Struggled
The researchers then made the task more direct.
In Studies 2 and 3, 905 participants were shown two stories: one written by a human and one generated by ChatGPT.
They knew that one was human and one was AI.
Their job was simply to identify which was which.
This setup should theoretically make the task easier. Participants were not trying to determine whether a random piece of text somewhere on the internet might be AI-generated. They knew in advance that exactly one of the two stories came from AI.
Still, they struggled.
In Study 2, only about 39 percent of participants correctly identified the stories — actually performing below what would be expected from random guessing.
In Study 3, approximately 52 percent answered correctly, essentially no better than chance.
The researchers concluded that across the two experiments there was no general evidence that readers could reliably distinguish the human-written stories from the AI-generated ones.
Perhaps even more revealing was confidence.
People who felt more confident about their answer were not necessarily more accurate.
That may be one of the more psychologically important findings of the research.
We do not simply struggle to recognize AI.
We may struggle without realizing that we are struggling.
What Do We Think AI Writing Is Supposed to Sound Like?
Ask someone how they recognize AI-generated text and they will often give you a list.
It is too polished.
Too organized.
Too predictable.
Too emotionally neutral.
Too repetitive.
Too clean.
Or perhaps the opposite: unnecessarily dramatic, overly descriptive, and suspiciously enthusiastic.
These clues can sometimes be useful. AI systems do develop recognizable patterns.
But the Villanova research suggests that people may also be relying on stereotypes about machine writing that are becoming outdated.
In Study 2, for example, participants reported using features such as how easy the story was to understand when deciding whether it had been generated by AI.
But relying on some of these intuitions was associated with worse performance.
The problem may be simple.
We have built a mental image of what “computer writing” looks like.
Meanwhile, the computers changed.
Modern language models are trained on enormous amounts of human-produced language. They can imitate narrative structure, emotional pacing, dialogue, symbolism, uncertainty, humor, and stylistic variation.
The old picture of machine-generated text — robotic sentences assembled by something that obviously does not understand human language — is becoming less useful.
There Is Another Bias Hiding in the Results
The most fascinating part of the study may not be that people failed to identify AI.
It may be that the label changed the experience of reading.
A story described as human-written was judged more favorably than the same kind of material described as AI-generated.
Psychologically, that makes sense.
When we believe another human created something, we do not experience only the words.
We imagine an intention behind them.
Someone felt something.
Someone remembered something.
Someone struggled to find the right sentence.
Someone decided that this particular experience was worth sharing.
The imagined person behind the text becomes part of the text itself.
If we believe a machine produced those same words, some of that invisible human context disappears.
The sentences may be identical.
Our relationship to them is not.
This helps explain why debates about AI creativity are about more than technical quality.
People are not only asking:
“Is this good?”
They are also asking:
“Does it mean the same thing if no human experienced what is being described?”
Those are two very different questions.
Does This Mean AI Is More Creative Than Humans?
No.
That would go far beyond what this research shows.
The experiment involved a small collection of short fictional stories under specific conditions.
The researchers used three published human-written stories and created corresponding AI stories using detailed prompts that described important themes, settings, perspectives, and narrative elements.
For example, the AI was not simply told, “Write a brilliant story.”
It received fairly detailed instructions about what kind of story to produce.
The research therefore does not demonstrate that AI is universally better at fiction, that it can replace writers, or that human creativity has somehow become obsolete.
Nor does the study tell us whether readers would make the same judgments about novels, poetry, journalism, memoirs, personal letters, or other forms of writing.
The researchers themselves acknowledge limitations in how broadly the results should be interpreted. The experiments also were not preregistered, something worth keeping in mind when evaluating the evidence.
But the study does demonstrate something narrower — and still significant.
Under these conditions, AI-generated fiction could be good enough that ordinary readers generally could not reliably identify its origin.
That threshold matters.
Familiarity With Literature Didn’t Help Much
One might expect people who read a great deal of fiction to be particularly good at spotting machine-generated prose.
According to the study, that was not the case.
Self-reported expertise with fictional literature did not reliably predict better identification.
Experience with AI, however, showed some relationship with greater accuracy.
People who were more familiar with AI systems were somewhat better at identifying AI-generated writing.
This raises an interesting possibility.
Learning to recognize AI may have less to do with becoming a sophisticated literary critic and more to do with becoming familiar with how language models behave.
Even then, the advantage was not enough to turn participants into reliable detectors.
For everyday readers, intuition remains shaky.
Why This Matters Beyond Fiction
At first glance, this seems like a study about short stories.
It is really about something much larger.
Increasingly, written language is how we build relationships with people we have never met.
We meet through dating apps.
We work through Slack and email.
We maintain friendships through text messages.
We argue through screens.
We apologize through screens.
We flirt through screens.
We share grief, loneliness, attraction, insecurity, and hope through screens.
And now AI can participate in all of those conversations.
Someone can ask an AI system to rewrite an angry message before sending it to a spouse.
A nervous dater can have AI compose a first message.
A manager can generate a compassionate email to an employee.
A person who does not know how to apologize can ask a chatbot to write the apology.
The recipient may never know.
That creates a psychological puzzle that society has barely begun to discuss.
Suppose someone receives a message that makes them feel genuinely understood.
Later, they discover that the sender used AI to write it.
Was the connection less real?
There is no simple answer.
Perhaps the sender genuinely cared but needed help finding words.
Perhaps AI allowed someone with poor communication skills to express something they had always struggled to say.
Or perhaps the sender outsourced emotional effort entirely.
The text alone may not reveal the difference.
Intent becomes increasingly important.
We May Need a New Definition of Authenticity
For most of modern history, authorship was relatively straightforward.
Someone wrote something.
Today, authorship is becoming a spectrum.
One person may write every word themselves.
Another may write a draft and ask AI to improve the grammar.
Another may describe what they want to say and let AI compose it.
Another may generate twenty versions, combine several of them, and rewrite the result.
And another may simply press a button and send whatever appears.
Calling all of these situations either “human-written” or “AI-written” may eventually become inadequate.
The more useful question may be:
How much human intention is behind the words?
That distinction matters particularly in relationships.
We often care less about literary originality than about whether the person communicating with us genuinely means what is being said.
A perfectly written apology means very little if the person sending it feels no remorse.
A simple, awkward sentence can mean everything if it is sincere.
AI does not eliminate that difference.
If anything, it may make the difference harder to see.
The Human Label Still Matters
There is something almost paradoxical about the Cambridge study.
Participants could not reliably recognize which stories came from AI.
They sometimes preferred the AI-generated work.
Yet they still valued stories more when they believed a human had written them.
That suggests that people are not judging creative work purely as information.
We care about origin.
We care about intention.
We care about the mind we imagine on the other side.
And perhaps this is why AI-generated language feels so psychologically complicated.
The technology is becoming increasingly capable of reproducing the external signals of human communication.
But communication has never been only about signals.
It is also about our belief that another consciousness is trying to reach us.
As AI becomes better at producing words that sound human, recognizing whether a machine helped create a message may become increasingly difficult.
The more important challenge may be deciding when that distinction matters.
Because in the years ahead, the question may no longer be simply:
“Was this written by AI?”
It may become:
“Who meant it?”
Source
The research discussed in this article is “Bot or not: Can people tell the difference between stories written by a human or by an AI system?” by Sydney Sears and Deena Skolnick Weisberg, published online by Cambridge University Press in Judgment and Decision Making on August 5, 2026. DOI: 10.1017/jdm.2026.10042.