Why Authors May Need to Prove How Their Books Were Written
By Wendy Keir | EmpowerAi™
Authors have traditionally been asked to submit a finished manuscript. Increasingly, they may also need to retain evidence showing how that manuscript came into existence.
The director-general of the Commonwealth Foundation has warned that writers should be prepared to prove that their work is their own following a dispute surrounding the 2026 Commonwealth Short Story Prize. Several winning stories were accused publicly of having been produced with artificial intelligence, including the overall winner, Jamir Nazir.
The Foundation investigated the allegations and asked the writers to provide evidence of their creative processes. According to its published statements, the organisation reviewed the available material, spoke with the authors and concluded that the winning stories should stand.
The controversy has exposed a problem that publishing has not resolved. AI-generated writing can be difficult to identify reliably, although accusations alone can still damage an author’s reputation, threaten publication and disrupt a career.
This changes the practical meaning of authorship. It may no longer be enough for an author to state that they wrote the work. They may also need to show the route by which the work was developed.
AI detectors are being asked to provide certainty they cannot provide
The allegations surrounding the Commonwealth Prize were partly driven by AI-detection software. One system reportedly classified Nazir’s story as entirely AI-generated. The Foundation did not accept that result as conclusive and has said that current detection methods have a high potential for error, particularly across different languages and literary styles.
This is an important distinction.
AI detectors identify patterns that appear statistically similar to machine-generated text. They do not observe the writing process. They cannot know whether a particular sentence was written by a person, influenced by an editor, translated from another language or shaped by the conventions of a particular literary tradition.
A result may indicate that a piece of writing deserves closer attention. It cannot provide a complete history of how the work was created.
The risk becomes greater when detection is treated as proof. An author may be expected to defend themselves against a conclusion produced by a system whose reasoning they cannot inspect and whose accuracy may vary according to language, genre and writing style.
The Commonwealth case also raises questions about fairness. Writers whose English does not follow dominant publishing patterns may be more vulnerable to being labelled unusual or artificial. Highly polished prose can attract suspicion, while repeated phrases, formal structures or certain grammatical choices may be interpreted differently by a detector than by an experienced human editor.
The technology can create an allegation much more quickly than the author can clear it.
The writing process may become part of the evidence
The Commonwealth Foundation appears to have relied on more than the submitted stories. It reviewed additional evidence and held discussions with the writers before confirming its decision.
This suggests a practical direction for authors.
Drafts may matter. Research files may matter. Notes, outlines, editorial correspondence and tracked changes may matter. An author who can show how an idea developed over several months is in a stronger position than someone who has only a finished document and no visible history behind it.
None of these records proves authorship perfectly. A person could fabricate drafts, while a genuine author may work in a way that leaves very little documentation. Some writers compose directly into a final document, delete earlier versions or develop much of the work privately before writing it down.
Even so, the cumulative record can show a recognisable creative process.
For an expert author, that record may include client observations, teaching materials, previous articles, recorded talks and earlier versions of the framework. The book can be traced back through a body of work rather than appearing suddenly as a polished manuscript.
That history may become part of the value.
AI use needs more precise language
Publishing conversations often divide work into two broad categories: human-written and AI-generated.
In practice, the boundary is much less tidy.
An author may use AI to transcribe an interview, identify repetition, question the structure of a chapter or suggest alternative wording. Another may ask AI to create complete sections and then edit them. A third may provide a detailed argument, examples and source material before using AI to help organise the first draft.
All three have used AI, although the role of the author is different in each case.
A useful policy therefore needs to ask more than whether AI was involved.
Who developed the central argument? Who selected the evidence? Who created the examples? Who checked the claims? Who made the final decisions? Who accepts responsibility for the published work?
These questions provide a clearer picture of authorship than a general declaration that AI was or was not used.
The Commonwealth Foundation has said it is reviewing its processes and will introduce clearer guidance for future competitions. That work will need to address the difference between assistance and substitution, while recognising that the line between them may be difficult to standardise.
Provenance may become part of an author’s professional infrastructure
Most authors do not currently organise their files with a future authorship dispute in mind.
That may need to change.
An author could retain dated outlines, early drafts and research notes. Editorial changes could be tracked rather than overwritten. Sources could be recorded more carefully. Where AI is used, the author could keep a simple record of the purpose it served and the decisions that remained human.
This does not need to become an elaborate surveillance system around the creative process. It is closer to ordinary professional record-keeping.
Expert-led businesses already retain evidence for many other reasons. They document client work, protect intellectual property, record approvals and maintain versions of important materials. The same discipline can be applied to a book, course or framework.
It also helps the author understand the provenance of their own work.
A long project often contains material from many sources and stages. Keeping a clear record makes it easier to distinguish original thinking, external evidence, editorial input and AI assistance. That clarity supports copyright, licensing, future updates and the creation of connected products.
AI Book Companion™ requires the same discipline
An AI Book Companion™ is built from an author’s existing body of work, although the companion produces new language whenever a reader asks a question.
This makes provenance particularly important.
The system needs to know which ideas belong to the author, which language comes directly from the book and which parts of the response are being generated to help the reader interpret the material. It also needs to recognise where the author’s work does not provide an answer.
The build process should therefore leave a clear record.
There should be an approved source library. The author’s frameworks and terminology should be defined. The boundaries of the companion should be documented, including subjects it cannot address and situations where it must direct the reader elsewhere.
This protects both the author and the reader.
The author can see how their work is being represented. The reader is less likely to receive an invented principle presented as though it came from the book. Where a response is generated rather than quoted, the companion can still remain governed by the author’s established ideas.
The provenance of the source material becomes part of the architecture.
The burden should not fall entirely on authors
There is an uncomfortable aspect to asking authors to prove that they are human creators.
A writer may be accused publicly on the basis of an automated score and then expected to produce private notes, drafts and personal correspondence to clear their name. The accusation can spread quickly, while the review process is slower and less visible.
Publishers, competitions and platforms therefore need fair procedures.
An AI-detection result should not be treated as a verdict. Authors should understand what evidence is being considered and have an opportunity to respond. Organisations also need to recognise the risk of bias, particularly when writers from different linguistic and cultural backgrounds are being assessed.
Clear rules introduced before submission would help.
Writers should know what forms of AI assistance are permitted, what needs to be disclosed and what records they may be expected to retain. A policy developed only after an accusation is unlikely to provide consistent protection for either the author or the organisation.
Authorship is becoming more visible because it is less easily assumed
For most of publishing history, the author’s name on the cover was accepted as a sufficient account of origin.
Generative AI has weakened that assumption.
The industry is now trying to establish new forms of evidence. Some of these will involve disclosure statements and publishing contracts. Others may involve technical tools, version histories and clearer documentation of the writing process.
The most useful approach is unlikely to depend on a single AI detector or a simple badge declaring a book human-written.
Provenance is cumulative. It sits in the author’s drafts, research, established body of work, editorial relationships and willingness to take responsibility for the final material.
For expert authors, this may become an advantage. Their ideas often have a history that can be seen through years of practice, teaching and application. A book grounded in that history carries something that cannot be created merely by producing fluent text.
The writing matters. The route by which the thinking was developed may begin to matter just as much.
Three Key Insights
1. AI-detection scores should not be treated as proof of authorship or misconduct.
They identify patterns rather than observing how a piece of writing was created, and their reliability may vary across languages and styles.
2. Authors may need to retain more evidence of their creative process.
Drafts, notes, research, tracked changes and editorial records can help demonstrate how a work developed.
3. Provenance should be built into AI Book Companion™ from the beginning.
The system needs a clear distinction between the author’s established ideas, approved source material and newly generated responses.
Three Questions I’m Thinking About
1. What evidence should an author reasonably be expected to provide when their work is accused of being AI-generated?
2. How can publishers investigate genuine concerns without allowing unreliable detection tools to damage authors unfairly?
3. Will a documented creative process become part of the professional value of a book, particularly for expert authors?
Sources
Commonwealth Foundation — 2026 Commonwealth Short Story Prize Update
https://commonwealthfoundation.com/2026-cw-prize-update/
Commonwealth Foundation — 2026 Commonwealth Short Story Prize
https://commonwealthfoundation.com/commonwealth-short-story-prize-2026/
The Guardian — Short story accused of being AI-written wins overall Commonwealth prize
https://www.theguardian.com/books/2026/jul/01/judges-claims-ai-use-commonwealth-short-story-prize-jamir-nazir
The Guardian — ‘Obvious markers of AI’: doubts raised over winner of short story prize
https://www.theguardian.com/books/2026/may/19/commonwealth-short-story-prize-winner-doubts-ai-artificial-intelligence
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.