Human-authored book labels are gaining ground fast. At the 78th Frankfurt Book Fair, which ran from 7 to 11 October 2026, programmes from Britain, France and Germany were on show, each promising to tell readers that a book was written by a person rather than generated by artificial intelligence. AFP’s Sam Reeves reported that the push follows “a string of scandals in which authors were suspected of using the technology without admitting it”, from a Goncourt contender in Paris to a $2 million debut that collapsed in New York.
The schemes differ a great deal. The US Authors Guild runs a self-certification mark that now covers around 6,500 titles. A British startup, Books By People, has just had its “organic literature” certification mark approved by the UK Intellectual Property Office. In France, Librinova audits manuscripts and interviews their authors before granting a “human creation” stamp, and in Germany a group of children’s publishers has launched its own label. All of them are answering the same question: how does a reader know who, or what, wrote the book in their hands?
We read the AFP report and then checked its sources: the schemes’ own rules, the Authors Guild’s expansion announcement, WIRED’s interview with Books By People, the BBC, Publishers Weekly and The Independent on the scandals, and the French prize records. This article explains how each human-authored label works, what the numbers show, why detection software cannot carry the weight on its own, and what the trend means for publishers, authors and any business that publishes content.
Table of contents
- What AFP Reported: Human-Authored Labels Take Centre Stage in Frankfurt
- The Scandals Behind the Rush to Human-Authored Labels
- Every Major Human-Authored Book Label Compared
- How Human-Authored Certification Actually Works
- The Numbers: How Far Human-Authored Labels Have Spread
- Why Detection Alone Cannot Carry a Human-Authored Mark
- The Case Against Human-Authored Labels
- Regulation vs Human-Authored Labelling: Where the Law Stands
- What Human-Authored Labels Mean for Publishers, Authors and Businesses
- Human-Authored Book Labels FAQ
- References
What AFP Reported: Human-Authored Labels Take Centre Stage in Frankfurt
The AFP story, published on 9 October and syndicated widely the next day, is short but dense. Its core claim is that “initiatives to label books as written by humans, not artificial intelligence, are gaining ground”, and that several human-authored labelling programmes “were being showcased at the world’s biggest book industry fair in Frankfurt this week, where the challenges posed by AI were hotly debated”.
Frankfurt 2026: a fair dominated by the AI question
The 78th Frankfurter Buchmesse, with the Czech Republic as guest of honour, opened against a tense backdrop. At the opening press conference on 6 October, Sebastian Guggolz, head of the German Publishers and Booksellers Association (Börsenverein des Deutschen Buchhandels), named falling reading skills and the absence of AI regulation as the two themes of the year. The association published a “Frankfurt Appeal” calling for transparency about the works used to train AI, fair licensing and stronger copyright enforcement, which about 120 publishers had signed by the start of the fair.
Guggolz also cited 2025 market figures showing a 30.6% fall in book buying among 10 to 15 year olds. Human-authored labels did not create that anxiety, but they arrived in the middle of it.
Three countries, three different human-authored models
AFP highlighted three approaches. In France, Librinova offers an audited “human creation” label for a small fee. In Britain, Books By People certifies titles for publishers that pay to join. In the United States, the Authors Guild runs a self-certification system based on a signed licence. Germany’s contribution at the fair came from children’s and young adult publishers, who introduced a label stating that their books are “written, designed and reviewed by humans”.
The warning that framed the week
The most quoted line in the AFP piece was a caution rather than a cheer. “There’s no simple ‘with AI’ or ‘without AI’,” Guggolz said at the fair, adding that basic programmes to categorise how much AI went into a book could fall short. “The question has to go deeper. It has to be addressed to each individual and to the entire industry: How do we want to make our practices transparent?” That tension, between a simple human-authored badge and a messy reality, runs through every scheme examined below.
The Scandals Behind the Rush to Human-Authored Labels
Publishing has long relied on an unspoken rule: a book that reaches the shelf was written by the person named on the cover. A run of cases in 2026 has shaken that assumption, and each one strengthened the commercial case for a visible human-authored promise.
Thélyson Orélien and the Goncourt
The most high-profile case, AFP says, is that of Thélyson Orélien, a 38-year-old Haitian-Canadian writer whose debut novel C’était ça ou mourir (“It Was Either That or Die”) was a best-seller and a prize contender. On 21 September an anonymous account on X called Balance ton Claude posted screenshots claiming that the AI detector Pangram had scored passages of the book as AI-written. Librinova’s co-founder Charlotte Allibert told AFP the affair was a “real earthquake” for the French book trade.
Orélien denies using AI and believes the attacks are racially motivated. The novel had been on the Prix Goncourt’s first list of 16 titles, published on 2 September. On 25 September the Académie Goncourt announced it was withdrawing the book from its selection, and the second list of eight, published on 6 October, did not include it. We covered the detector results in detail in our analysis of how reliable AI detectors are: newsrooms that tested the same book got answers ranging from “very likely” human to 100% machine-written.
Shy Girl: the first major-publisher withdrawal
In March, Hachette cancelled the US release of Mia Ballard’s horror novel Shy Girl and withdrew the UK edition. The book had been self-published in February 2025 and released in Britain by Hachette’s Wildfire imprint in November 2025, selling almost 2,000 copies there, according to the BBC. Orbit, the US imprint, had planned to publish it the following month. The New York Times, which first reported the story, said it “appears to be the first commercial novel from a major publishing house to be pulled over evidence of AI use”.
Ballard denied using AI herself. She told the Times that an acquaintance she hired to edit the original self-published version had used it. That defence points to a gap that a simple human-authored badge struggles with: a book passes through many hands, and the author is not the only person who might reach for a chatbot.
Call Me, I’ll Hide the Body: a $2 million deal withdrawn
In July, Jerry Falade, a Nigerian PhD student at Southern Methodist University in Dallas, saw his debut crime novel Call Me, I’ll Hide the Body go to auction. Fourteen US publishers bid, and the reported US figure passed $2 million. On 29 July the manuscript was withdrawn from submission after a journalist ran it through AI detection software and the agents said they could no longer stand behind it. Pangram scored the book as 97% AI-generated, WIRED reports.
Falade denies using AI to write the novel and says detection tools mistook features of Yorùbá oral storytelling for machine writing. In an essay published in September, he counted 18 “not X, but Y” constructions and 46 em dashes in his manuscript, both now treated as AI tells. “It takes months, even years to write a book,” he wrote. “It takes AI-detection software two minutes to ruin your life’s work.”
A prize-winning short story under suspicion
In May, readers alleged that one of the five winning stories of the 2026 Commonwealth Short Story Prize had been generated with AI, after it scored 100% AI on Pangram. The Commonwealth Foundation and Granta, which publishes the winning stories, stood by it and noted that detectors are imperfect, while the Foundation reviewed its selection process. Four cases in six months, across three countries and two languages, explain why a human-authored promise suddenly looks like a product rather than a slogan.
| Date | Event |
|---|---|
| 2023 | France’s Création humaine label is created |
| Jan 2024 | Librinova offers the Création humaine label to its authors for 49 euros |
| Jan 2025 | Authors Guild launches its Human Authored mark in beta for members |
| Jul 2025 | French group Les Nouveaux Éditeurs extends Librinova’s detection tool across its houses |
| 2 Mar 2026 | Authors Guild opens Human Authored to all US-published authors and publishers |
| 10 Mar 2026 | Society of Authors launches its UK scheme at the London Book Fair |
| 20 Mar 2026 | Hachette pulls Shy Girl in the US and UK |
| May 2026 | Commonwealth Short Story Prize winner accused of AI use |
| 29 Jul 2026 | Call Me, I’ll Hide the Body withdrawn from submission |
| 21-25 Sep 2026 | Orélien accused on X; Académie Goncourt withdraws his novel |
| Early Oct 2026 | German children’s publishers launch “Von Menschen geschrieben” label |
| 7 Oct 2026 | UK IPO approves the Books By People certification mark |
| 7-11 Oct 2026 | Frankfurt Book Fair showcases human-authored labels |
Every Major Human-Authored Book Label Compared
There is no single human-authored standard. There are at least six live schemes for books alone, and the BBC counted at least eight broader “human-made” labelling initiatives across industries in March. Here is what each book scheme promises and how it checks.
Authors Guild: Human Authored (United States)
The Authors Guild launched its Human Authored certification in beta for members in January 2025. On 2 March 2026 it opened the mark to all authors with books published in the United States. Members certify free; non-members pay $10 per title and must pass third-party identity verification. Every author signs a licence agreement, and each certified title gets a unique registration number in a public database that anyone can search.
By early March, Publishers Weekly reported, more than 3,000 authors had certified 5,000 titles. AFP puts the total at around 6,500 now. Authors are limited to ten certifications a year without special permission, and publishers can buy certifications in bulk at a discount. The definition is strict on text and loose on tools: the book must be “written by a human and not generated by AI, with the exception of minimal uses such as AI applications that check spelling and grammar or for research”.
Society of Authors: Human Authored (United Kingdom)
On 10 March 2026, novelist Tracy Chevalier launched the Society of Authors’ version of the scheme at the London Book Fair, in partnership with the Authors Guild. Authors register their books on the Human Authored website and download a logo for the back cover. It is free to Society of Authors members and limited to text-based works for now, with an “aspiration to open the scheme to illustrated works and to non-members”.
The society is candid that the human-authored logo is a stopgap. Its chief executive, Anna Ganley, called it “an important sticking plaster to protect and promote human creativity in lieu of AI labelled content in the marketplace”. The scheme’s own site says that “the onus should be on tech companies to label AI-generated content”, and that until that happens this labelling can help.
Books By People: “organic literature” (United Kingdom)
Books By People, co-founded by Esme Dennys and Conrad Young, confirmed on 7 October that the UK Intellectual Property Office had approved its certification mark, a thumbprint-like stamp for book covers. Its standard is called “organic literature”: writing must originate from the author, with limited AI use allowed for tasks such as spelling or grammar checks. Publishers, not authors, join the scheme, paying a flat membership fee plus a sum for each book.
The verification goes further than a promise. WIRED reports that Books By People asks for an authorship declaration, runs each manuscript through proprietary software, and manually examines early drafts and research materials for “indicators of humanity”. Five publishers have signed up, Galley Beggar Press says it intends to adopt the mark, and the startup expects the newly approved mark on its first book by the end of the year. Former Penguin Random House chief executive Markus Dohle has backed it.
Librinova and the Création humaine label (France)
Librinova, a self-publishing company and literary agency, has offered the Création humaine label to its authors since January 2024, at launch for 49 euros. An author signs a contractual commitment that the work was conceived and written by human means. The text is then run through two AI detectors and read by several people, and the author is interviewed by a professional auditor. “None of these steps alone would allow the work to be labelled,” Allibert said at launch.
About 200 works have received the human-authored stamp so far, AFP reports. Allibert says demand is driven by fear as much as pride: “More and more, authors are afraid of being suspected of using AI—self-published authors even more so.” Separately, the French group Les Nouveaux Éditeurs extended Librinova’s detection tool across its houses in July 2025, after a pilot at its La Tribu imprint flagged about 7% of incoming manuscripts as written or substantially rewritten with AI.
Von Menschen geschrieben (Germany)
In early October, Penguin Random House and 16 other members of Germany’s Association of Youth Book Publishers (avj) introduced the label “Von Menschen · geschrieben · gestaltet · geprüft” (written, designed and reviewed by humans). It covers more than text: it promises that authors wrote the words, artists created the illustrations, people selected the layout and design, and publishers checked the content. The publishers say it is meant to help buyers in online shops, where they cannot rely on a bookseller’s curation.
Other human-made badges
Several publishers and companies have gone their own way. Faber and Faber put a “Human Written” stamp on some books, including Sarah Hall’s novel Helm at her request, though the BBC noted Faber had not said how it audits the claim. In Australia, Proudly Human, led by former chief scientist Alan Finkel, runs checks at every stage of publication. Free downloadable badges such as Not By AI sit at the other extreme, with little or no checking.
| Scheme | Country | Who applies | Cost | How it checks | Reported scale |
|---|---|---|---|---|---|
| Human Authored (Authors Guild) | US | Authors, publishers in bulk | Free for members, $10 per title otherwise | Signed licence, identity check, public registry | About 6,500 titles |
| Human Authored (Society of Authors) | UK | Society members | Free for members | Self-declaration and register | Not published |
| Books By People | UK | Publishers | Membership fee plus per-book fee | Declaration, software, review of drafts and research | Five publishers |
| Création humaine (Librinova) | France | Authors | Small fee (49 euros at launch) | Commitment, two detectors, readers, auditor interview | About 200 works |
| Von Menschen geschrieben (avj) | Germany | Member publishers | Not published | Publisher commitment to human text, art and editing | 17 publishers |
| Proudly Human | Australia | Publishers and creators | Not published | Checks at each stage of publication | Not published |
How Human-Authored Certification Actually Works
Strip away the logos and there are only three ways to back a human-authored claim. A scheme can trust the author and punish lies, audit the finished text, or examine how the book was made. Most real schemes mix them.
Model one: self-certification backed by law
The Authors Guild model relies on a contract rather than a test. The Guild does not vet books before publication. Instead, an author who lies is, in the Guild’s words, “liable for trademark infringement, and also potentially for consumer fraud”, and “could be forced to pay significant damages”. The deterrent is legal exposure, and the public registry makes every claim checkable after the fact.
This is cheap and fast, which is why it has scaled to thousands of titles. Its weakness is obvious: it catches only those who are caught some other way. WIRED describes such schemes as operating “under an honesty system”.
Model two: audit, detection and interview
Librinova’s process stacks several imperfect checks on the theory that together they are stronger. Two AI detectors give a statistical signal, human readers give a literary one, and an auditor’s interview tests whether the author can talk convincingly about their own choices. The cost is time and money, which limits scale: about 200 works in nearly three years.
Model three: the document trail
Books By People leans on evidence of process: drafts, redrafts, research notes and editorial correspondence. “We realised pretty early on that, as AI models get better, getting into that race is never going to be sustainable,” Dennys told WIRED. “I wouldn’t call it AI detection.” Where evidence is ambiguous, the company says it errs on the side of the author: “The bedrock of our organisation is to avoid [writers] being unjustly accused.”
| Model | Example | Strength | Weakness |
|---|---|---|---|
| Self-certification | Authors Guild, Society of Authors | Cheap, fast, scales to thousands of titles | Relies on honesty and after-the-fact enforcement |
| Audit and interview | Librinova’s Création humaine | Several independent signals | Slow, costly, still uses detectors |
| Document trail | Books By People | Harder to fake a years-long drafting history | Burdens authors with record-keeping, costs publishers fees |
| Publisher pledge | German avj label, Faber stamp | Covers text, art and editing together | Verification method not published |
What counts as “minimal” AI use?
Every scheme allows some AI, and that is where the definitions start to blur. The Authors Guild permits grammar checkers and editing software, research, and AI brainstorming or outlining “as long as the AI use is merely preparatory and no actual AI-generated text appears in your work”. Its human-authored mark applies only to the body of the book, so an AI-built index or table of contents does not disqualify it. Translations and audiobook narrations need separate certification.
Books By People allows limited AI for spelling and grammar. The German label goes furthest, covering illustrations and design as well as words. AFP notes the worry that “minimal use” could be “open to interpretation”, and it is easy to see why: a grammar tool that rewrites a whole paragraph is doing something very different from one that fixes a comma.
| AI use | Authors Guild | Books By People | German avj label |
|---|---|---|---|
| Spelling and grammar checks | Allowed | Allowed (limited) | Not specified |
| Research | Allowed | Not specified | Not specified |
| Brainstorming and outlining | Allowed if no AI text appears | Not specified | Not specified |
| AI-generated prose in the body | Not allowed | Not allowed | Not allowed |
| AI illustrations or design | Outside scope (text only) | Not specified | Not allowed |
| AI index or table of contents | Allowed (body only is certified) | Not specified | Not specified |
The Numbers: How Far Human-Authored Labels Have Spread
The Authors Guild is the only scheme that publishes enough data to measure growth. Its figures suggest steady, not explosive, uptake.
About 214 new certifications a month is meaningful for a membership body, but small against the size of the market. Amazon alone had to cap self-publishers at three new titles a day in 2023 to slow a flood of AI-generated books. The human-authored registries are a curated island in a much bigger sea.
Uptake by scheme
The other schemes report different units, which makes direct comparison misleading. Librinova counts about 200 labelled works. Books By People counts five member publishers rather than titles. The German label counts 17 publishers. The Society of Authors has not published a figure. What the numbers do show is that the self-certification model has scaled roughly 30 times further than the audited one, which follows directly from cost and effort.
Fear as a driver
The demand side is telling. Allibert says authors want the human-authored stamp because they are afraid of being accused, not just because they want to be celebrated. Dennys reports “a lot more interest from people all over the world” as “anxiety around undisclosed AI use has increased in recent months”. The label is becoming insurance against suspicion, which is a different product from a mark of quality.
Why Detection Alone Cannot Carry a Human-Authored Mark
Every scandal above began with an AI detector score. That makes the reliability of detectors central to whether any human-authored scheme can be trusted, and whether a scheme can protect innocent authors rather than expose them.
One book, many verdicts
The Orélien case is the clearest natural experiment so far. Different testers ran the same novel through detectors and got very different answers. Even the strongest tools report a share of text that looks machine-like, not proof of who typed it.
Small edits, big swings
In the same affair, a poet showed that changing about 20 words in a passage flipped a Pangram result from 100% AI to 100% human. Detectors are statistical models built with natural language processing, which you can read about in this overview of natural language processing. They estimate how predictable text looks, and light editing can move that estimate a long way in either direction.
Pangram, which WIRED describes as the leader in the field, claims 99.7% accuracy, and independent studies have found its error rates very low on long English text. We looked at that record in our piece on whether Pangram deserves its gold-standard reputation. But a low average error rate still produces false accusations at scale.
The base-rate arithmetic
Suppose a human-authored scheme screened 10,000 manuscripts that were all genuinely written by people. The number wrongly flagged depends entirely on the false-positive rate, and a rate that sounds tiny is not zero.
Bias and the burden of proof
Falade’s case adds a sharper worry. If detectors read certain non-native or culturally distinct styles as machine-like, then routine screening would fall hardest on writers who already face extra scrutiny. Orélien and Falade have each said racism played a part in how they were treated. Tom West, chief executive of Publishers’ Licensing Services, warned WIRED: “We need to be quite careful that this doesn’t end up becoming a witch hunt.”
That is why the more careful human-authored schemes treat a detector score as one signal among several, and why Books By People avoids calling its process detection at all.
The Case Against Human-Authored Labels
AFP reported that “some in the book trade view the idea of labels with scepticism”. The objections are serious and they come from inside the industry, not just from AI companies.
The burden points the wrong way
West’s central argument is that labelling human work gets the problem backwards. “Authors are being constantly quizzed about the origins of their ideas and their work. I think that’s something we need to resist,” he told WIRED. “The default should be that a book is human-authored.” On this view, it is the AI-generated book that should carry the label, and platforms and publishers should be the ones to declare it.
Even the Society of Authors partly agrees, calling its own scheme a “sticking plaster” until tech companies label AI content. A human-authored badge that becomes expected could turn its absence into an accusation.
The grey zone is wide
Guggolz’s point that there is “no simple ‘with AI’ or ‘without AI'” is backed by experts outside publishing. AI researcher Sasha Luccioni told the BBC that AI is “so ubiquitous and so integrated into different platforms and services, that it’s truly complicated to establish what ‘AI free’ means”, and argued for a spectrum rather than a binary. Spell checkers, translation aids, research assistants and editing suites increasingly run on a large language model of some kind, so the line between assistance and authorship keeps moving.
Easy to game, hard to police
AFP notes “concerns it could be too easy to circumvent checks in the fledgling initiatives and pass off AI-produced books as human-written”. Self-certification depends on honesty; audits depend on detectors that can be fooled by light editing; document trails can, in principle, be fabricated. No scheme claims to be foolproof, and none publishes data on how many applications it has rejected or how many certified books were later found to involve AI.
Too many badges
Competing human-authored labels may confuse the very readers they are meant to reassure. Consumer expert Dr Amna Khan of Manchester Metropolitan University told the BBC that “competing definitions of what is ‘human made’ are confusing consumers” and that “a universal definition is essential”. The book trade now has two schemes under one name, Human Authored, run on different rules in two countries, plus several more marks.
| Argument for labels | Argument against labels |
|---|---|
| Gives readers a choice and a way to find human writing | Puts the burden on honest authors instead of AI producers |
| Protects authors from suspicion with a recorded commitment | Absence of a label may start to look like guilt |
| Creates legal consequences for false claims | Self-certification is easy to game |
| Signals publisher values, like a “no cheating” policy | “Minimal AI use” is open to interpretation |
| Fills a gap while regulation lags | Competing marks confuse readers |
The case for doing it anyway
Defenders accept the flaws and argue the signal matters more than perfect enforcement. Science fiction author Ted Chiang told WIRED that “it’s just as important for publishers to have a ‘no cheating’ policy as it is for universities, even if there’s no perfect method for enforcement. It’s a statement of values.” He also explained why readers care: “The knowledge of how something was created cannot be separated from your experience.”
Regulation vs Human-Authored Labelling: Where the Law Stands
Private human-authored labels exist partly because the law does not yet require AI-generated books to say so. The rules that do exist are narrow.
The EU AI Act’s transparency rules
Article 50 of the EU AI Act became applicable on 2 August 2026. It requires providers of generative AI systems to mark synthetic output in a machine-readable way, and requires deployers to disclose AI-generated text published to inform the public on matters of public interest, unless it has had human review and someone holds editorial responsibility. We explained the provider and deployer duties in our guide to the EU’s compulsory AI content labels.
A novel is rarely “text published to inform the public on matters of public interest”, so most fiction falls outside the deployer duty. Machine-readable marks also help little once text has been copied, edited or printed. That leaves a gap that human-authored labelling is trying to fill from the other side.
Amazon’s disclosure rule
Amazon’s Kindle Direct Publishing requires authors to tell it when content is AI-generated, meaning text, images or translations created by an AI tool, even if they edited it afterwards. AI-assisted content, where the author created the work and used AI to edit or brainstorm, does not need to be disclosed. KDP’s guidelines do not say that the disclosure is shown to readers.
Copyright, training and certification marks
The labels sit alongside a wider fight over training data. Anthropic’s $1.5 billion settlement with book authors in 2025 showed the scale of money at stake, and the Frankfurt Appeal asks the EU for disclosure of the works used to train AI, as we set out in our explainer on training AI models on copyrighted books. Legally, the strongest human-authored labels are certification marks: the owner sets the rules, licenses the mark and can sue misusers, which is exactly what the UK IPO approval gives Books By People.
| Rule or tool | What it requires | Effect on books |
|---|---|---|
| EU AI Act, Article 50 | Machine-readable marking by providers; disclosure of some AI text by deployers | Most fiction is outside the text disclosure duty |
| Amazon KDP guidelines | Disclose AI-generated content to Amazon | AI-assisted work need not be disclosed |
| UK certification marks | IPO-approved rules for who may use a mark | Lets a scheme owner enforce its human-authored claim |
| Authors Guild licence | Signed agreement for each certified title | Misuse treated as trademark infringement and possible fraud |
| Frankfurt Appeal | Calls for disclosure of the works used to train AI, and licensing | Campaign only; no legal force |
What Human-Authored Labels Mean for Publishers, Authors and Businesses
Whatever happens to individual schemes, the direction is clear: provenance is becoming part of what a book, or any piece of content, is expected to carry. The practical lessons reach well beyond literary fiction.
For publishers
Publishers that want a human-authored mark, or simply want to defend their list, need an AI policy that covers editors, translators and designers as well as authors. Shy Girl showed that a freelance editor’s AI use can sink a book. Contracts should say what AI use is permitted at each stage, and production workflows should keep version histories that could support a certification or a dispute.
For authors
Every scheme that goes beyond self-declaration asks for evidence of process. Authors can protect themselves by keeping dated drafts, research notes and editorial correspondence, and by turning on version history in their writing software. Falade’s earliest defence was text messages from 2021 that predated ChatGPT. Records made at the time are far more persuasive than explanations made after an accusation.
For any business that publishes content
The same pressures are reaching marketing copy, reports, journalism and customer communications. A business that publishes in the EU already has Article 50 duties for some AI text. More broadly, customers increasingly want to know how content was made, and a clear disclosure policy is cheaper than a reputational crisis. Our AI strategy team helps organisations write AI-use and disclosure policies that hold up to that scrutiny.
| Who | Action | Why it matters |
|---|---|---|
| Publishers | Write an AI policy covering authors, editors, translators and designers | Third-party AI use can trigger a withdrawal |
| Publishers | Keep manuscript version histories | Supports certification and disputes |
| Authors | Save dated drafts, notes and correspondence | Evidence of process beats detector scores |
| Authors | Read each scheme’s rules on permitted AI use | “Minimal use” differs between schemes |
| Businesses | Publish an AI content disclosure policy | Meets customer expectations and EU duties |
| Everyone | Never treat a detector score as proof | False positives and bias are documented |
Watch the next twelve months
Three signals will show whether human-authored labelling sticks. The first is whether a major retailer or platform starts displaying the marks in search results, which would turn them from cover decoration into a buying filter. The second is whether any scheme publishes rejection and revocation data. The third is whether regulators move the burden onto AI-generated books, as the Society of Authors and Tom West both want, which would make human-authored badges far less necessary.
Human-Authored Book Labels FAQ
What does a human-authored label on a book mean?
It means the author, and sometimes the publisher, has committed that the text was written by a person rather than generated by AI. Depending on the scheme, that commitment may be a signed licence, an audited process with an interview, or a review of drafts and research materials.
Can a book that used AI spell checkers still be certified?
Yes, under the main schemes. The Authors Guild allows grammar and spelling tools, research and AI brainstorming as long as no AI-generated text appears in the book. Books By People allows limited AI use for spelling and grammar checks.
How much does Human Authored certification cost?
The Authors Guild’s mark is free for its members and $10 per title for other US-published authors. The Society of Authors’ version is free for its members. Librinova’s label cost 49 euros at launch, and Books By People charges publishers a membership fee plus a per-book fee.
Are human-authored labels legally enforceable?
The stronger ones are. The Authors Guild treats misuse as trademark infringement and possible consumer fraud, and Books By People’s mark has been approved by the UK Intellectual Property Office as a certification mark, which lets its owner set the rules and act against misuse.
Do AI detectors prove a book was written by AI?
No. Detectors estimate how machine-like a text looks. Results vary between tools, light edits can flip a score, and even a low false-positive rate produces some wrongful flags at scale. That is why the more careful human-authored schemes combine detection with interviews or evidence of how the book was written.
References
‘Human-authored’ book labels gain ground amid AI scandals (AFP via Tech Xplore)
Your Next Great Read Might Be Certified ‘Organic’ (WIRED)
Human Authored certification expands to all authors (Authors Guild)
Human Authored FAQ (Authors Guild)
Human Authored scheme (Society of Authors)
Books By People: organic literature certification
Von Menschen geschrieben, gestaltet, geprüft (Penguin Random House Germany)
Les Nouveaux Éditeurs deploy Librinova’s Création humaine tool (Livres Hebdo)
Shy Girl by Mia Ballard: horror novel pulled by publishers over alleged AI use (BBC News)
Is this product ‘human made’? The race to establish an AI-free logo (BBC News)
After the cancellation of Call Me, I’ll Hide the Body, what comes next? (Publishers Weekly)
I lost my $2m book deal because people mistook my African dialect for AI (The Independent)
La deuxième sélection du prix Goncourt 2026 (Livres Hebdo)
Frankfurter Buchmesse warns of falling reading skills and AI (hessenschau)
Content guidelines: artificial intelligence content (Amazon KDP)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
