Skip to content 
This Is Not Another “AI Good or AI Bad” Article
Imagine I give you two short stories.
I tell you that the first was written by a human who spent weeks developing the characters, wrestling with the language, drinking unreasonable quantities of whisky, and periodically threatening Chapter Seven with physical violence.
I tell you the second was generated by AI.
Which one do you expect to feel more authentic?
Now imagine I haven’t actually given you two stories at all. I’ve given you the same story twice. Suddenly we have a much more interesting psychological problem.
Recent research suggests that what readers believe about authorship can influence how they judge writing, sometimes independently of the words sitting directly in front of them. Tell people AI was involved and perceptions of authenticity, trustworthiness, literary merit, and even the writer themselves can change.
Which creates a peculiar challenge for writers in 2026.
We are no longer writing only for readers. We are writing for readers who may also be wondering who—or what—wrote the thing they’re reading.
This Is Not Another “AI Good or AI Bad” Article
Before the internet reaches for its ceremonial pitchforks, this article is not an argument for or against AI. We have surely suffered enough of those to last several geological periods.
AI exists. Writers are using it. Editors are using it. Publishers are trying to decide what exactly constitutes acceptable use, while everybody else argues about em dashes. The psychologically interesting question is different.
What happens to the reading experience once we believe AI was involved?
Because increasingly, it appears that we don’t evaluate writing purely on its content. We also evaluate the story we have been told about how that content came into existence. We can’t help ourselves.
The AI Penalty Is Now Measurable
A fascinating 2026 study by Siavosh Sahebi, Paul Formosa, and Sarah Bankins examined how people responded to written communication they were told was either human-written, AI-assisted, or fully AI-generated.
The researchers found what they called an AI penalty. Communication involving AI was generally perceived as less trustworthy, less authentic, and less useful than communication presented as entirely human-authored. The penalty became stronger as the amount of stated AI involvement increased.
Then came the wonderfully human part. Participants also believed that people should disclose their use of AI. Fair enough. Except that when AI involvement was disclosed, they penalised the communication for it.
The researchers called this the disclosure paradox.
In other words, we appear capable of simultaneously holding the beliefs “You should tell me if you used AI” and “I will think less of this if you tell me you used AI.”
Human psychology remains committed to keeping things simple.
And Creative Writing May Be Particularly Vulnerable
This effect becomes even more interesting when we move specifically into creative writing.
A recent series of 16 preregistered experiments involving more than 27,000 participants found that creative writing received lower evaluations when people believed it had been produced by AI, or with AI assistance, rather than by a human author alone.
Crucially, perceived authenticity helped explain the difference. When we read a novel, memoir, poem, essay, or emotionally personal piece, we are not engaging only with sentences. Somewhere behind the text, we imagine another mind. A person, with all their foibles, life experiences, and wisdom, sharing it with us.
When readers believe that presence has been replaced, or even partly outsourced, the emotional meaning of the writing can change.
Same Story, Different Label
Another experiment makes the problem wonderfully clear. Researchers Martin Abel and Reed Johnson gave participants the same short story but randomly described it either as AI-generated or as representative of a human author’s writing.
The words did not suddenly become worse when the AI label appeared. They were the same words. Yet the AI-labelled version received lower subjective evaluations, particularly for qualities including authenticity, atmosphere, and literary merit.
Interestingly, the effect did not substantially change how much time people actually spent reading or how much they were willing to pay or work to continue reading.
What people say they value and what they actually do are not always identical.
Anyone who has bought a book because of its magnificent cover and then left it unread beside the bed for eighteen months will recognise the general principle. Still, the perception effect is significant. Once AI entered the story of authorship, readers began judging the writing through a different psychological lens.
Why Does the Label Matter So Much?
Part of the answer appears to be that humans don’t judge creative work purely by output. We care about origin, intention, effort, and agency. A hand-knitted scarf means something different from an identical scarf produced by a machine because part of its value lies in knowing someone spent six evenings making it while quietly regretting the entire project.
The same can happen with art.
A paragraph isn’t merely a collection of correctly arranged words. To many readers, it represents an act of observation, imagination, emotional processing, and choice. We want to believe there was somebody on the other side of it.
That becomes particularly important when the writing asks us to feel something. If an author tells us about grief, love, shame, trauma, joy, parenthood, mortality, or the magnificent absurdity of being human, part of our response comes from the sense that another human being understands the experience.
Once AI authorship is suspected, the reader may begin asking a different question.
Did somebody actually mean this?
And once that question appears, emotional trust can wobble.
We Also Bring Confirmation Bias to the Investigation
There is another psychological complication. Once somebody decides a text is AI-generated, they start looking for proof. Suddenly the em dash looks suspicious. The paragraph structure is suspicious. The word delve has apparently entered witness protection. Three adjectives in a row? Machine. A neat summary? Machine. A metaphor involving a tapestry? Straight to jail.
This is confirmation bias doing what confirmation bias does rather well. Once we form a belief, we become more attentive to evidence that appears to support it. The problem is that many supposed “AI tells” are also perfectly ordinary features of human writing. AI learned them from human language in the first place.
A polished article may sound like AI because AI was trained on polished articles. That is roughly equivalent to accusing somebody of being a duck because ducks also have legs.
Can Readers Actually Tell?
The evidence here is more complicated than social-media confidence would suggest.
Research published in 2026 found that ordinary readers could identify human and AI-generated narrative writing at rates better than chance, but they were far from perfectly accurate. Interestingly, those readers also preferred texts they believed had been written by humans. Their perception of authorship rated alongside the actual authorship.
Other research has found that experienced, frequent users of language models can become surprisingly good at detecting AI-generated nonfiction under controlled conditions, while other studies involving reviewers have found considerable uncertainty.
So, the sensible conclusion is not that “nobody can detect AI.” Nor is it that Brian from Facebook can reliably identify ChatGPT because he has developed a spiritual relationship with punctuation. The evidence suggests that detection ability varies enormously according to the reader, genre, model, editing, context, and amount of AI involvement.
Which means suspicion is not proof.
The Authenticity Problem for Genuine Human Writers
This creates a new and rather unpleasant problem for writers whose work genuinely is human.
You may spend days refining a passage until it is beautifully structured and then discover that its very polish makes somebody suspicious.
We are beginning to see writers changing perfectly legitimate stylistic habits because they fear being labelled AI. They remove em dashes. They deliberately make sentences less polished. They worry about using phrases that have appeared on online lists of “AI words.”
At that point, something has gone badly wrong. A writer should not have to make their work worse in order to prove that a human wrote it. Nor should human style become defined by intentional imperfection. Our goal should not be to produce writing sufficiently untidy that nobody could possibly suspect technological involvement.
The goal is to preserve something far more important:
individual voice.
What Human Voice Actually Looks Like
Human voice is not simply quirky punctuation or occasionally breaking a grammar rule. It is the peculiar analogy nobody else would have made because nobody else has your history, humour, obsessions, irritations, memories, vocabulary, cultural background, and slightly unhealthy relationship with a particular childhood incident.
If three people describe the same funeral, they will notice entirely different things. One remembers the flowers. Another remembers the widow’s hands. Another remembers somebody’s phone ringing during the prayer and the entire congregation developing a sudden professional interest in the carpet.
That selection is voice. AI can imitate stylistic surfaces extremely well. What writers need to protect is the thinking underneath them.
Use AI as a Tool, Not a Substitute for Having Something to Say
This is where balanced AI use becomes important.
If you use AI to brainstorm twenty possible chapter titles, organise research notes, test whether an explanation is clear, identify repetition, or ask what objections a reader might raise, you are using a tool around your thinking.
If you repeatedly ask it to decide what you believe, invent your personal stories, manufacture your emotional insights, and generate the actual intellectual substance of your work, something else begins happening.
The writing may still be competent. But increasingly, the machine is not helping you express your thinking. It is supplying the thinking. And readers appear remarkably sensitive to the question of whether a human mind is genuinely present behind creative communication.
So, give your own brain first refusal.
- Think before you prompt.
- Draft before you polish.
- Have an opinion before asking AI to improve it.
- Bring the raw material of being human to the page first.
Then decide where technology can usefully help.
Don’t Let AI Sand Off the Interesting Bits
One of the most common problems I see with heavily AI-assisted writing isn’t that it is technically poor. Quite the opposite. It is frequently too smooth. Every paragraph is nicely proportioned. Every transition is helpful. Everything is balanced, civilised, and suspiciously well behaved.
Which can leave the writing feeling like a hotel room. Perfectly pleasant. Absolutely nothing in it belongs to anybody.
When using AI to edit, writers therefore need to watch what gets removed. The odd sentence rhythms, dry jokes, regional expressions, unexpected metaphors, tiny digressions, and emotional rough edges may be precisely what give the writing personality.
Editing should improve communication. It should not laminate it.
If You Use AI, Keep Evidence of Your Process
There is also a practical reality writers can no longer ignore.
Given recent disputes around AI authorship, I think professional writers should start treating creative provenance as part of ordinary manuscript management.
- Keep drafts.
- Retain version history.
- Save your outlines, research notes, handwritten ideas, editorial correspondence, beta-reader feedback, and Track Changes files. If you use AI materially, keep a simple record of what it was used for.
Not because writers should need to present a police dossier every time they use an em dash. But because a visible creative trail can demonstrate something an AI detector cannot:
how the work actually developed.
At Harvard Ink, we’ve now created a Human Authorship & Manuscript Provenance Pack specifically to help authors document this process. It records manuscript versions, source material, editorial involvement, AI use where applicable, and other evidence showing how a book developed from idea to finished manuscript. If you would like a copy, visit our website to download it (hopefully my web support pops it up today for us), or drop us an email at info@harvardink.com.
We are not demonising AI, but there is a need to protect authorship.
Transparency Needs Nuance Too
The disclosure paradox creates an ethical problem.
If writers know that admitting any AI assistance may cause readers to judge their work more harshly, they have an incentive to hide it. That is not a healthy direction. But equally, describing every technological interaction as equivalent makes little sense.
Using AI to ask whether Chapter Four repeats Chapter Two is not equivalent to asking it to write Chapter Four. Using transcription software is not the same as generating a memoir. Using an AI tool to brainstorm ten alternative subtitles is not equivalent to outsourcing the argument of an entire nonfiction book.
We need a more mature vocabulary around AI assistance; one based on degree, purpose, and authorship, not simply “AI touched this” versus “AI never entered the building.” Writers should also check the AI policies of agents, competitions, publishers, universities, and clients before submitting work because standards continue to vary.
What This Means for Editors
I think human editing actually becomes more valuable in this environment, not less.
Writers increasingly need someone who can improve a manuscript without making it sound like everybody else’s manuscript. A good editor does not impose a generic “professional” voice. They strengthen your voice. They identify what is unclear, repetitive, structurally weak, or ineffective while protecting the rhythms, humour, emotional texture, cultural specificity, and individual observations that make the writing recognisably yours.
Readers are becoming more alert to generic language. Perfectly polished blandness is not the goal. Clear, compelling, unmistakably human writing is. The emerging psychology around AI writing reveals something rather encouraging beneath all the controversy.
Readers still care deeply about human authorship. They care about authenticity, intention, effort, originality, and the sense that another mind genuinely exists behind the words.
Research now suggests that simply believing AI was involved can change how people judge identical or comparable writing, affecting perceived authenticity, trust, and literary value. Yet readers are not flawless detectors of AI authorship, and their assumptions can themselves become biases.
That leaves writers with a more interesting challenge than merely proving they are human.
We need to write like ourselves.
Use technology where it genuinely helps. Keep thinking. Keep making difficult creative decisions. Keep the strange metaphors. Keep the memories nobody else possesses. Keep evidence of your process where necessary, and don’t allow the growing suspicion around AI to frighten you into sanding every individual fingerprint off your prose.
Because perhaps the most valuable thing a writer can offer in an age of infinitely generated words is no longer simply good writing.
It is evidence of a mind worth listening to.
At Harvard Ink, we help fiction and nonfiction authors strengthen their manuscripts without stripping away the thing that increasingly matters most: their individual human voice. Our developmental editing, line and copyediting, writing coaching, ghostwriting, and publishing support focus on clarity, structure, emotional connection, and the psychology of how readers actually experience a book.
We also offer a Human Authorship & Manuscript Provenance Pack to help writers document the development of their work, including drafts, editorial history, research, and responsible AI use where applicable.
If you’re using AI and wondering where assistance ends and authorship begins, or if your manuscript simply needs professional human eyes that won’t turn it into polished beige, we can help.
Explore Harvard Ink’s services and current special offers:
Thank you for reading Part 17 of our series. To read our other posts on psychology for authors and writers, they’re right here:
Read Part XVI of the series here.
Research References
Sahebi, S., Formosa, P., & Bankins, S. (2026). The AI penalty and disclosure paradox: Trust, authenticity and knowledge uptake in AI-mediated communication. Computers in Human Behavior: Artificial Humans, 8, 100304.
Abel, M., & Johnson, R. (2026). AI Authorship Labels Reduce Evaluations but Not Costly Engagement with Creative Writing.
The invisible author: Citizen sociolinguistic perspectives on identifying human and AI-generated narrative texts. Social Sciences & Humanities Open, 13 (2026), 102646.
See also recent experimental research in the Journal of Experimental Psychology: General examining the effects of perceived AI authorship on evaluations of creative writing across 16 preregistered experiments.
