Readers Preferred AI-Written Stories, but Still Valued Human Authorship
By Wendy Keir | EmpowerAi™
A new study has found that readers rated short stories written by ChatGPT as more absorbing and higher quality than comparable stories written by people.
I suspect that finding will attract most of the attention because it feeds directly into the larger argument about whether AI can be creative.
The more useful part of the research sits slightly beneath that headline.
Readers often preferred the AI-generated stories when they judged the text itself. At the same time, they gave higher ratings to stories they believed had been written by a person, even when the information they had been given about the author was false. (The Guardian)
The writing and the origin of the writing were influencing the reader in different ways.
That matters for authors because publishing is increasingly dealing with two separate questions. Can AI produce work that people enjoy reading? Does the reader still care who created it?
This study suggests the answer to both may be yes.
The readers did not reliably recognise AI writing
The research was published in the journal Judgment and Decision Making.
In the first experiment, 1,682 adults read one of six short stories. Three had been written by people and three had been generated by ChatGPT around similar themes. The participants rated the story for qualities including engagement, absorption and overall quality. (The Guardian)
The researchers also told participants whether the story had supposedly been written by a person or by AI. That information was not always accurate.
When the text was assessed on its own qualities, the AI-generated stories performed well. Participants rated them as more absorbing and higher quality than the human-written stories. When participants believed a story had been written by a person, however, its rating increased. (The Guardian)
Two further experiments asked 905 adults to compare stories and identify which had been written by a person.
In one experiment, only around 40% identified the origin correctly. In the other, the figure was 52%. The researchers concluded that there was no general evidence that participants could reliably distinguish human-written stories from AI-generated ones. (The Guardian)
People who described themselves as experienced fiction readers were not noticeably better at identifying the source. Participants with more experience of AI performed somewhat better, suggesting that repeated exposure may help people recognise patterns in generated writing. (The Guardian)
This creates an awkward position for publishing.
Readers may believe they will recognise AI writing when they see it. The evidence suggests that confidence may be misplaced.
Readability is only one part of what a reader values
One explanation offered by the researchers is that the AI stories were simpler and easier to digest.
That should not be dismissed. Clear writing matters. A story that the reader can follow and remain engaged with is doing something important.
AI systems are particularly good at producing language that moves smoothly. They can maintain a recognisable structure, avoid some of the friction found in less polished drafts and generate prose that feels immediately accessible.
Those qualities can produce a positive reading experience, especially within a short experiment.
A book usually asks more of a reader.
It may develop an idea over several hundred pages. It may contain ambiguity, risk, discomfort or a voice that takes time to understand. Fiction may deliberately resist easy interpretation. An expert book may require the reader to stay with a difficult argument before its meaning becomes clear.
Some of the qualities that make writing distinct can also make it less immediately digestible.
This means the study should not be interpreted as proof that AI is now a better author than a person. It shows that, under the conditions tested, participants responded well to particular AI-generated short stories.
That is still significant. It tells us that fluency and reader engagement can no longer be treated as reliable evidence that a person wrote the work.
The label changed how the same kind of work was experienced
The participants’ response to the author label may be more important for the future of books.
They tended to give higher ratings to stories they believed had been written by a person. This happened even where the label did not reflect the true origin of the work. (The Guardian)
The reader was responding to an idea about authorship.
A human-written story carries an implied relationship. Someone chose to write it. They drew on memory, imagination, experience and observation. They made creative decisions and accepted responsibility for the result.
The reader may never meet the author, although there is still an awareness that another person created the work.
That awareness affects how the text is received.
A sentence about grief feels different when the reader believes it emerged from a human attempt to understand grief. A description of fear may carry a different weight when it is connected to a writer who has experienced fear, witnessed it or spent time imagining it.
The words may look similar on the page. Their perceived provenance changes their meaning.
This does not prove that every reader will reject AI-generated books. It suggests that many readers care about more than the functional quality of the prose.
Human authorship may need to become more visible
Until recently, a human author did not need to emphasise that they were human.
The name on the cover carried that assumption.
As AI-generated books become easier to produce and harder to identify, authors may need to give readers more evidence of the person and the thinking behind the work.
For a novelist, that may include discussion of the creative process, early drafts, influences and the decisions that shaped the story.
For an expert author, the provenance may be even clearer.
The book may have emerged from years of client work, research, teaching and repeated application. The author may have developed a framework because existing approaches failed to explain what they were seeing. Case studies and earlier articles may show how the thinking changed over time.
This history is difficult to capture through a label that simply says “human-written”.
The value sits in the connection between the person, the work and the experience from which it developed.
Authors may increasingly need to make that connection visible through their wider body of work, their source material and the way they speak about the development of their ideas.
Expert knowledge cannot be reduced to fluent sentences
The study also matters for coaches, consultants, therapists and other expert-led businesses.
AI can produce a polished article, a plausible framework or an apparently confident explanation. The language may be clear enough that the reader rates it highly.
That does not establish where the thinking came from.
An expert’s authority usually develops through accumulated judgment. They have seen what happens when a method is applied to different people and circumstances. They understand the exceptions, the limitations and the point at which a general principle stops being useful.
A language model can express a principle without holding that experience.
This is why expert-led businesses need to be careful about allowing the quality of the generated text to stand in for the quality of the underlying knowledge.
An article may read well while containing a framework that has never been tested. A course may sound coherent while combining ideas from unrelated sources. An AI agent may speak confidently while moving beyond the expert’s actual methodology.
Fluency can make weak thinking harder to notice.
The practical response is to make the source and authority behind the material clearer. Businesses need defined intellectual property, approved source material and a visible line of responsibility for what is published or delivered through AI.
AI can improve the reader’s experience without becoming the author
The study does not require authors to reject AI.
It does suggest that the role of AI needs to be described more precisely.
An author may use AI to test whether an explanation is clear, identify repetition or examine how different readers might interpret a section. They may use it to organise research, question the structure or improve access to the finished work.
The author still needs to decide what the work means.
They remain responsible for the argument, the creative choices, the evidence and the final result.
The distinction becomes blurred when generated text is presented under a human name without any clear account of how the work was produced. Readers may enjoy the text while believing they are entering into a relationship with a human author that does not really exist.
Transparency matters because the study shows that information about authorship changes the reading experience.
The label is not an administrative detail added after publication. It becomes part of how the reader understands the work.
AI Book Companion™ depends on visible intellectual provenance
An AI Book Companion™ produces generated language.
A reader asks a question and the system creates a response based on the author’s book, frameworks and approved supporting material.
The response does not need to be a quotation from the book to remain connected to the author. It does need to stay inside the author’s established body of knowledge.
That requires visible provenance.
The reader should know that they are speaking with an AI system. They should also know whose thinking governs the conversation, which material the companion has been built around and where its authority ends.
The companion may make an explanation easier to understand. It may ask a reflective question or help the reader apply a framework to their own situation.
It should not invent a new principle and quietly attribute it to the author.
This is where AI can contribute something useful without replacing the human relationship behind the book.
The technology can provide accessibility, continuity and interaction. The author provides the source, judgment, meaning and boundaries.
The study’s findings help explain why both parts matter.
Readers may prefer the clarity of AI-generated language. They may also place greater value on knowing that a person stands behind the work.
An AI Book Companion™ can hold those elements together when its role is clear.
Detection is unlikely to solve the authorship question
The participants’ difficulty identifying AI stories also raises doubts about the reliance on AI-detection tools.
Readers could not consistently tell which stories came from ChatGPT. Experience with fiction did not make them substantially more accurate. (The Guardian)
Automated detection systems face similar limitations. They analyse statistical characteristics of the text rather than observing the writing process.
A result may suggest that a document resembles generated writing. It cannot provide a complete account of how the ideas were developed, who made the decisions or what role AI played.
As generated writing improves, detection may become even less useful as the main mechanism for establishing authorship.
Provenance provides a more practical direction.
Authors can retain drafts, notes and research. Publishers can require clearer disclosure. Expert businesses can document their source material and approval process. AI systems can identify themselves and explain the knowledge base governing their responses.
None of these measures creates perfect proof.
Together, they provide a more meaningful account of authorship than asking whether a detector believes the sentences look human.
The value of the author is moving beyond the production of text
AI can now generate readable and enjoyable stories. This study gives us stronger evidence that readers may respond positively to the result.
The consequence is not necessarily that human authors become less valuable.
Their value may become easier to see in a different place.
The author is the person who develops the perspective, chooses what matters and takes responsibility for the work. They bring an intellectual and creative history that exists beyond the finished sentences.
For expert authors, the book represents developed judgment rather than simply an ability to arrange words fluently.
AI may help readers access that judgment. It may help the author express it more clearly or allow the book to continue through an interactive companion.
The human provenance still matters because readers appear to care where the work came from, even when they cannot identify that origin from the prose alone.
Three Key Insights
1. Readers may enjoy AI-generated writing without recognising its origin.
Participants rated the AI stories highly and generally struggled to distinguish them from stories written by people.
2. Human authorship affects how readers value the work.
Stories believed to have been written by a person received higher ratings, even when the label was inaccurate.
3. AI Book Companion™ needs to preserve the author’s intellectual provenance.
Generated responses can improve accessibility and engagement while remaining clearly governed by the author’s ideas, approved material and boundaries.
Three Questions I’m Thinking About
1. Will readers begin to judge books partly through the visible history of the author’s thinking rather than through the finished prose alone?
2. How should publishers explain different levels of AI involvement without reducing authorship to an overly simple human-or-machine label?
3. Could the strongest use of AI in publishing be to improve the experience around a human author’s work rather than generating a replacement for it?
Sources
The Guardian — AI-generated stories rated better quality than human-written ones, study finds
Open Science Framework — Stories and research materials used in the study
EmpowerAi™ | Real AI. Real Impact. Smarter Business Decisions.
AI Book Companion™ helps expert authors transform a completed book into a governed, interactive knowledge asset built around their own ideas, methodology and boundaries.