AI Ghostwriting Ethics: Where Is the Line on Honesty?

Ghostwriting has never been dishonest in itself. Presidents, chief executives and celebrity memoirists have used hidden writers openly for a century. AI ghostwriting ethics turns on the same question: does the reader rely on you having written, lived or checked these words? If they do and you let them believe something false, you have crossed the line. Context decides when you should disclose.

That is the thesis of this essay, and I'll defend it against two opposite camps. One says AI is just a tool, so nobody needs to mention it. The other says any uncredited help, human or machine, is a lie. Both get something right. Neither gives a working answer for a student, a speechwriter, a founder with a LinkedIn habit or a novelist with a deadline.

What is ghostwriting, and why was it ever acceptable?

Ghostwriting means one person writes and another person's name goes on the work. The 2024 paper by Fiona Draxler and colleagues in ACM Transactions on Computer-Human Interaction lists the usual places it happens: medicine, autobiographies and political speeches. They also list the usual reasons: lack of time, lack of writing skill, and the commercial value of a famous name on the cover.

Those reasons sound shabby when listed. Yet the practice has been tolerated for a long time in some fields and condemned in others, and that pattern explains most of what follows.

Speeches: the oldest open secret

Warren G. Harding is usually credited as the first US president to hire a full-time speechwriter, Judson Welliver, a former newspaper reporter. H. L. Mencken said after Harding's inaugural address that the president "writes the worst English I have ever encountered." According to a 2008 review in The Baltimore Sun of Robert Schlesinger's history of presidential speechwriters, White House Ghosts, every president except Calvin Coolidge followed Harding's lead.

Nobody calls a State of the Union address dishonest because a team drafted it. The convention is so settled that former White House speechwriters have a club named after Welliver. The audience is not relying on the president having typed the sentences. It is relying on the president meaning them, and being answerable for them afterwards.

Memoirs: your life, someone else's sentences

Celebrity memoirs are the second big category. J. R. Moehringer ghostwrote Phil Knight's Shoe Dog (2016) and Prince Harry's Spare (2023). In 2023 he even wrote an essay for The New Yorker about working with the prince, which tells you how far ghostwriting has moved from hidden to semi-public.

The memoir case shows why ghostwriting can be honest even when the cover names only one person. The material is the subject's own: their childhood, their grief, their version of events. The ghostwriter supplies structure and sentences after hours of interviews. Readers buying a famous person's memoir mostly want the story and the perspective, and those really do come from the named author.

The Pulitzer that tested the convention

John F. Kennedy's Profiles in Courage won the 1957 Pulitzer Prize for Biography. That December, columnist Drew Pearson said on Mike Wallace's television show that the book was ghostwritten, pointing at Kennedy's aide Ted Sorensen. Kennedy hired the lawyer Clark Clifford, produced handwritten notes, and Sorensen signed an affidavit backing Kennedy's authorship. ABC read a retraction on air.

Fifty years later, in his 2008 memoir Counselor, Sorensen wrote that he "did a first draft of most chapters" and "helped choose the words of many of its sentences." Kennedy had thanked Sorensen in the preface, so the help was never completely hidden. The fight was over its size.

This is the most useful historical case for AI questions, because it shows where the tolerance ran out. A speech carries no implied claim of personal craft. A prize for writing does. The more a byline is the thing being judged, the less room there is for an invisible hand.

Where ghostwriting was never acceptable

Two fields have treated ghostwriting as misconduct for a long time. In medical research, papers drafted by undisclosed writers and then signed by academics became a recognised problem; a 2009 article in PLoS Medicine by Peter Gøtzsche and colleagues asked what should be done to tackle ghostwriting in the medical literature. Draxler's team notes that unethical practices like this led to formal authorship criteria in science.

The second field is education. A student who pays someone to write an essay is not using a convention; they are cheating, because the whole point of the essay is to show what the student can do. No one has ever argued that a coursework essay works like a presidential speech.

What made human ghostwriting honest enough?

Looking across those cases, ghostwriting has been accepted when three conditions held, and condemned when one of them broke.

• The audience knew the convention. Voters know about speechwriters. Readers of celebrity memoirs half-expect a collaborator. Nobody reading a medical paper expects a hidden pharma writer, and no teacher expects a hired essay.

• The substance belonged to the named person. The policy in the speech is the president's. The life in the memoir is the prince's. A ghostwritten research paper put an academic's name on someone else's analysis.

• The named person was accountable. They read it, approved it, and would defend it. If the speech contains a false claim, the president owns it, not the speechwriter.

Notice what is missing from that list: who physically produced the sentences. That has rarely been the test. The test has been whether the reader's reasonable assumptions about the text were true.

Ghostwriter vs AI writer: what actually changes?

If the old test still works, you might think AI changes nothing. It changes quite a lot, just not in the way the loudest critics claim. Here is a side-by-side comparison of a hired human ghostwriter and an AI model doing the same job.

The fabrication problem

A good ghostwriter who doesn't know a fact asks you. An AI model may simply fill the gap with something that sounds right. That matters for honesty in a direct way: if you sign text that contains invented quotes or wrong figures, you are now vouching for claims you never made. The accountability condition from the previous section only holds if you actually read and check every line.

Everyone has a ghost now

Eternal Perspective Ministries, a Christian publisher, put the scale point bluntly in a 2026 article: now everyone can have a ghostwriter without finding or paying one. When ghostwriting meant hiring a professional for months, it was limited to people with a public role and a team. Now a student, a junior marketer and a retiree writing a family history can all do it in minutes. The conventions that made speechwriting acceptable were never designed for this.

No one to credit makes hiding easier

The most interesting research here is the "AI Ghostwriter Effect" study by Draxler and colleagues at LMU Munich, the University of Bayreuth and Aalto University. In two studies (30 and 96 participants), people did not consider themselves the owners or authors of AI-generated text, yet they still did not publicly declare the AI's authorship. Participants were more willing to recognise ownership by a supposed human ghostwriter than by an AI one, while the excuses they gave for not crediting either were similar.

In plain terms: people know the AI wrote it, feel it isn't quite theirs, and put their own name on it anyway. With a human ghostwriter there is at least a contract, a person and sometimes an acknowledgement page. With AI, there is no one to object, so the gap between "I didn't write this" and "I'll sign it" widens.

Is AI ghostwriting ethical?

Sometimes, yes. AI drafting can be as ethical as a speechwriter's draft when the same three conditions hold: the audience's assumptions are true, the substance is yours, and you stand behind every line. It becomes unethical when a reader relies on your personal writing, knowledge or verification and you let them believe in something that isn't there.

That position gets attacked from both sides, so the counter-arguments deserve a fair hearing.

The case that AI is just a tool

The strongest version goes like this. Writers have always used help: editors, spellcheckers, thesauruses, colleagues who rewrite a paragraph over coffee. Nobody discloses Grammarly. Even the Authors Guild, no friend of AI companies, says AI spelling and grammar checks are "uniformly considered ethical and professional," as is using AI for research if you don't copy its output.

The Alliance of Independent Authors adds an accountability principle: what matters is that you know what you are doing, made an informed choice and take responsibility for what you publish. On this view, a LinkedIn post drafted by Claude and approved by a founder is no different from one drafted by a comms intern, and nobody expects the intern to be named.

I think this argument is right about low-stakes, conventional writing and wrong as a general rule. It treats all assistance as equal. But the Authors Guild itself warns against treating all AI use as ethically equivalent: fixing commas and generating chapters are different acts. Spellcheck changes how your words look. Generation changes whose words they are.

The case that any uncredited help is a lie

The opposite camp is just as clear. A Psychology Today post from July 2026 argued that putting your name on an article AI wrote "in its entirety (or close to it)" means "you're just plain lying." The Eternal Perspective Ministries article goes further: when whole chapters are written by editors or by AI, they should be credited as co-writers, and many writers refuse to ghostwrite at all because it isn't honest.

This view has the virtue of consistency, and I respect writers who live by it. But it condemns a century of political speeches, most celebrity memoirs and the bulk of corporate communications, including many cases where nobody was deceived. If a convention is known to everyone, invoking it is not a lie. The strict view also gives no guidance for the huge middle ground where AI suggested a structure, drafted a paragraph you then rewrote, or tightened your own prose.

Where I land

The honest question in AI writing ethics is not "did AI touch this?" but "what does my name on this promise, and is the promise true?" A speech promises that the speaker means it. A personal essay promises that the writer's mind produced it. A news report promises that someone checked it. A student essay promises that the student did the work. Disclosure is how you keep the promise true when your process doesn't match what the reader would assume.

Is it dishonest to publish AI written content?

Publishing AI-written content is not dishonest by default. It becomes dishonest when the context implies human authorship and you hide that it isn't. A product description written by AI and published on a shop page deceives no one, because nobody reads a product page for its author. A novel sold as the work of a named human writer is a different matter.

Publishing is where the stakes have become concrete. The University of Oregon's Daily Emerald reported in 2026 that Hachette, one of the Big Five publishers, cancelled the release of a horror novel, Shy Girl, after readers claimed it was AI-generated. The same article noted that many publishers now require authors to disclose AI use, but there is no standard way to vet manuscripts for it.

For authors, the Authors Guild's guidance is specific:

• Disclose substantial AI-generated text publicly. It suggests a note in the front matter of a book or at the bottom of an article, on the grounds that readers have a right to know whether a human wrote what they buy.

• Check your contract. Book contracts usually include a warranty that the manuscript is original, and the Guild warns that including AI-generated text may breach it.

• Disclaim AI text when registering copyright. The Guild says knowingly failing to do this may be treated as fraud on the US Copyright Office and could invalidate the registration.

The Alliance of Independent Authors lists the opposite failure too: claiming a book was written entirely by a human when AI played a substantial role is, in its words, misrepresenting authorship to readers.

Why people stay quiet anyway

Jingchao Fang and Mina Lee at the University of Chicago reviewed the research on why writers do or don't disclose AI use in 2025. The factors they identified include how likely the AI use is to be detected, whether there is a formal penalty for not disclosing, and fear of social costs: readers may rate disclosed AI-assisted writing as lower quality or less sincere. On the other side, disclosure can make a writer look honest, and it spreads when others do it too.

That last point matters more than it seems. Disclosure feels risky partly because it is rare. Each person who discloses plainly and without drama makes it cheaper for the next one.

Why "nobody can tell" is the wrong test

AI detectors are unreliable in both directions, which tempts some people to think undetectable use is fine. A 2023 letter in the Croatian Medical Journal discussed a study in which the ZeroGPT detector flagged most of 200 published research abstracts as containing AI text, and also produced a high rate of false positives on abstracts written before the AI era. If the tools can't tell, the ethics can't rest on getting caught. They have to rest on what you told your reader.

How does authorship ethics change from one context to another?

Authorship ethics is not one rule applied everywhere. It is a set of expectations that differ by genre, and you can usually work out the right level of disclosure by asking what the reader in that genre assumes. Here is how I'd map the common cases.

Two things stand out. First, disclosure expectations rise as the text gets closer to being a performance of the named person's own mind or labour. Second, rules beat ethics in practice: if your school, employer, publisher or platform has a policy, that policy is the floor, not a suggestion.

The personal LinkedIn post is the trickiest row. Executive posts have long been written by comms teams, and many readers know it. But the genre has drifted toward first-person confessions and lessons from failure, which only work if they really are personal. A ghostwritten announcement is fine. A ghostwritten story about your father's death, posted as if you wrote it in a moment of reflection, borrows trust it didn't earn.

A practical rule of thumb for AI ghostwriting ethics

Here is the rule I'd use: disclose AI help whenever a reasonable reader, learning exactly how the text was made, would feel misled. If they would shrug, you don't owe them a note. If they would feel cheated, you do.

To apply it to a real document, work through these steps.

1. Name what your byline promises. Write one sentence: "Readers of this assume that I..." Fill it in honestly. "...mean what it says" is a weak promise. "...wrote this myself as coursework" is a strong one.

2. Check the written rules first. Look for a course AI policy, a journal's authorship rules, a publisher's contract warranty, a platform's content guidelines or an employer's policy. If one exists, follow it even when your own ethics would allow more.

3. Estimate the AI share honestly. Was AI used for spelling, for ideas, for a first draft you rewrote, or for text you pasted in nearly unchanged? These are different, and memory is unreliable after a few editing sessions, so a record helps.

4. Ask the reliance question. Would your reader feel misled if they saw the full history of the document? Picture a specific person: your teacher, your editor, a subscriber who pays for your newsletter.

5. Check the substance is yours. Even with AI drafting, the ideas, experiences, judgements and facts should be ones you hold and have verified. If AI supplied the argument itself, that is closer to authorship than assistance.

6. Size the disclosure to the stakes. A one-line note is enough for most articles. Coursework may need a fuller statement of what you used and how. A book may need a note in the front matter.

7. Keep the evidence. If you ever need to show what was yours, a record made while writing beats a reconstruction made after an accusation.

What a proportionate disclosure looks like

Good disclosures are specific and short. Vague ones ("AI may have been used") sound evasive and tell the reader nothing. Some examples:

• Article: "I drafted this piece myself and used ChatGPT to suggest a structure and tighten two sections. All facts were checked by me."

• Coursework: "Following the course AI policy, I used Claude to generate counter-arguments in section 3, which I then rewrote. The analysis and conclusions are my own."

• Book front matter: "Portions of the first drafts of chapters 4 and 7 were generated with an AI tool and substantially revised by the author."

• Executive post: "Drafted with help from my team and AI tools; the views are mine."

Notice that none of them apologise. A disclosure is a description of method, like a methods section in a paper, not a confession.

What if your ghostwriter is using AI?

This is the angle most discussions leave out. If you pay a human ghostwriter, you may now be getting AI text without knowing it, and the honesty problem lands on you because your name is on the cover.

The Authors Guild recommends that anyone hiring a ghostwriter, co-author, editor or book coach sign a written contract that bars adding more than a trivial amount of AI-generated text without written approval, and requires the helper to disclose their AI use. That makes sense. You can only make honest promises to readers if your collaborators are honest with you.

For ghostwriters, the logic runs in reverse: quietly handing over AI-drafted chapters puts a client at legal and reputational risk they never agreed to.

Common mistakes people make with AI and authorship

Most dishonest AI use isn't villainous. It comes from predictable errors.

Treating editing as a way to wash the text. Draxler's study found that the more influence people had over AI-generated text, the stronger their sense of ownership. That feeling is real but it isn't proof. Changing every tenth word of an AI draft may make it feel like yours without making the thinking yours. Heavy rewriting is different; light polishing is not.

Borrowing a convention from the wrong genre. "CEOs have ghostwriters" is true and irrelevant to a scholarship essay. Before relying on a norm, check that it belongs to the genre you're actually writing in.

Over-disclosing trivia. Declaring that you used spellcheck or asked an AI for a synonym is noise. It can make readers think the AI did more than it did, which is its own kind of inaccuracy.

Disclosing vaguely. "This post used AI" covers everything from a typo fix to a fully generated article. Say which parts and how.

Deciding after the fact. If you only think about disclosure when someone asks, you'll be guessing about your own process. Decide at the start what you'll use AI for and how you'll record it.

Assuming detectors settle the question. Passing a detector says nothing about whether your reader was misled, and failing one says nothing about whether you cheated. Detection and honesty are separate problems.

So where is the line?

The line was never between human help and no help. Harding's speeches, Shoe Dog and Spare all had ghosts, and the people named on them were not liars, because the substance was theirs and they stood behind it. The line sits where a reader's reasonable assumption about who wrote, lived or checked the words is false and the named author lets it stand.

That is the core of AI ghostwriting ethics. AI moves people across the line more easily because it is cheap, always available and has no one to credit. It doesn't move the line itself. A student essay, a personal essay or a novel still promises personal work. A product page or a routine company update still doesn't.

The practical side of this is easier than it sounds if you keep a record while you write. Assist Writer is a browser editor that marks text you paste in as AI (with Ctrl+Shift+V) separately from what you type and shows the human and AI split of the document, so a disclosure becomes a matter of reading the numbers rather than trying to remember. For more on school AI policies, false accusations and how detectors differ from authorship records, browse the Assist Writer blog on honest AI writing.

Frequently asked questions

Is AI ghostwriting ethical?

It can be. Using AI to draft text is about as ethical as using a human speechwriter when three things hold: your readers' assumptions about the text are true, the ideas and experiences are yours, and you have checked and stand behind every line. It becomes dishonest when readers rely on your personal writing, as in coursework, personal essays or novels, and you hide that AI produced it.

Is using AI the same as hiring a ghostwriter?

Not quite. A human ghostwriter interviews you, works from your material, checks facts and signs a contract that usually assigns copyright to you. An AI model knows only what you put in the prompt, can invent facts with confidence, and its purely generated text is not copyrightable, according to the Authors Guild. It is also cheap enough that anyone can use it daily, which strains old ghostwriting conventions.

Do I need to disclose AI use in a LinkedIn post or newsletter?

It depends on what the post promises. A company update or announcement drafted with AI and approved by you rarely needs a note, much like posts written by a comms team. A first-person story about your own experience, sold as a personal reflection, is different: readers assume you wrote it yourself. If they would feel misled learning AI wrote it, add a short disclosure.

Can I copyright a book I wrote with AI help?

Partly. The Authors Guild says AI-generated text is not copyrightable because it lacks human authorship, so it must be disclosed and disclaimed when registering a work with the US Copyright Office. The human-written parts can still be protected. The Guild also warns that knowingly hiding AI text may invalidate a registration and could breach the originality warranty in a book contract.

How should I word an AI disclosure?

Be specific and brief, and describe method rather than apologise. Say which tool you used, for which parts, and what you did with the output, for example: "I used ChatGPT to suggest a structure and tighten two sections; the arguments and fact-checking are mine." Avoid vague lines such as "AI may have been used", which tell readers nothing and can sound evasive.

Did famous people disclose their ghostwriters?

Sometimes, partly. John F. Kennedy thanked Ted Sorensen in the preface of Profiles in Courage, yet Sorensen only acknowledged in his 2008 memoir that he drafted most chapters. Prince Harry's Spare was ghostwritten by J. R. Moehringer, who later wrote about the collaboration in The New Yorker. Presidential speechwriting has been an open convention since Warren Harding hired Judson Welliver.