Sotrane Publishers’ Position on Using AI to Write a Book

Sotrane Publishers' position on AI in book writing is straightforward: a tool may help an author think, plan and revise, but the book must remain the work of a person who knows what it says and can answer for it. We are interested in tools that genuinely help writers, and we expect the knowledge, judgement and responsibility on every page to be the author's own.

We do not ask authors to avoid AI tools altogether. Used well, they can help organise recordings, spot a repeated paragraph or show what a first-time reader cannot yet follow. Used carelessly, they can replace the very material that makes a book worth reading. After an AI rewrite, a retired teacher's chapter might gain cleaner spelling and a line about how she "embraced the transformative power of education", and lose the question a pupil asked on her first day in class and her admission that she had no good answer. No tool lived that first day, decided what it taught her or can take responsibility for presenting it truthfully.

This page sets out what that means in practice: where we think AI help is welcome, what we expect authors to check, how we think about Zambian voices and local languages, what we ask authors to disclose and what we expect of ourselves as a publisher.

Help is welcome when it helps the writer think

Authors have always had help. A friend reads an early draft, a librarian helps find a source, an editor asks why a scene starts too late and a spellchecker catches an error a tired writer has read past five times. No serious publisher expects a good book to be produced alone.

AI can do some of those supporting jobs. A writer with a folder of interviews might use a tool to group their own questions by theme before planning chapters. Someone preparing a training handbook might ask it to identify terms that have not been explained for a beginner. A novelist might ask for a list of possible inconsistencies in the timeline and then return to the manuscript to decide whether each one is real. An author writing in English as an additional language may use suggestions to make a sentence clearer without surrendering its meaning or the natural rhythm of their voice.

The task makes the difference. "Show me where a reader might be confused" invites the writer to investigate. "Write a moving memoir in my voice" invites a machine to fill a gap with plausible experience. The practical test is whether, after using the tool, the author understands the chapter better and has more control over it, or less.

A lecturer in Lusaka writing for students might ask what questions a new reader would have after a chapter on study habits. If one question reveals that she has assumed every student has reliable internet access, the tool has helped her notice a gap. She still has to decide how the chapter should serve students with different resources, and no prompt can replace listening to the people the book is for.

The reader is owed an accountable author

A book can be beautifully arranged and still contain no knowledge anyone will stand behind. If a health educator, a financial trainer or a historian prints an inaccurate statement, readers cannot question the software that produced it. They will question the name on the cover and the publisher associated with the book.

For that reason, we do not treat a fluent AI answer as a checked source. A date, quotation, study, procedure or figure belongs in a manuscript only after the author has verified it against reliable material, and a convincing reference that cannot be found is not evidence. A tool can help someone list what needs checking, but it cannot make a claim true by repeating it confidently. The US National Institute of Standards and Technology describes confident false output of this kind as "confabulation" in its generative AI risk profile.

Consider an entrepreneur writing a book for small shop owners who asks for help making a chapter on stock records easier to read. The new version adds a statistic about how many Zambian businesses fail because of poor record-keeping. Unless she can find and assess a credible source for that exact claim, it has no place in the book. She can still explain, from her own experience or a clearly marked illustrative case, how missing records can hide losses. A real, narrower explanation is better than an impressive invented number.

Accountability also means being able to answer an editor's ordinary questions. Why is this example here? What happened before it? Where did this claim come from? Why does the character speak this way? If the author cannot explain the material because a tool supplied most of it, the manuscript needs more work, however polished the surface appears.

A tool can help an author think. It cannot take responsibility for what the book tells its readers.

A Zambian book should keep its home in a rewrite

There is nothing provincial about a book that sounds like its writer and respects its setting. An author may write for Zambian readers, for readers elsewhere in Africa or for a wider audience, and clarity for those readers is a worthwhile goal. Flattening every voice into the same anonymous international English is not.

Suppose a novel set in Luanshya includes a grandmother who avoids a direct answer when a child asks about a family dispute. A tool might rewrite the exchange as an orderly explanation of everyone's feelings. The sentences become easier to follow, but the child no longer has to work anything out and the grandmother no longer sounds like herself. The author can edit for clarity while keeping the relationship and the uncertainty that drive the scene.

Local languages call for particular care. A model may offer a translation that looks authoritative to someone who cannot check it, while spelling, register and meaning can all depend on context. A term used in a familiar conversation may sit awkwardly in a formal narrator's voice. Ask someone fluent in the language and familiar with the situation, and if you cannot verify a phrase, do not publish it simply because it gives a page a convenient "African" flavour.

This does not mean protecting every awkward sentence in the name of authenticity. Readers deserve clear, well-edited writing. It means asking what a change keeps and what it removes. Our article on manuscripts that sound like AI offers ways to find the places where smooth prose has replaced a writer's own observations.

Access matters, and shortcuts still have a cost

Writers do not all have the same money, time, training or access to professional support. It would be easy to dismiss AI use from a position where every author could afford months of editorial help, and that would ignore why a first-time writer might try a free or inexpensive tool to get unstuck.

A parent developing a children's story may use one to generate questions about the age of the intended reader, or to make a checklist before reading the draft aloud. A university student turning a dissertation into a book for general readers may ask which technical terms need introducing earlier. These can be sensible steps. They do not mean the tool has judged whether a child enjoys the story, whether a claim is supported or whether the revised chapter still makes sense to a general reader.

An author with a limited budget can still decide carefully where to spend it. Work through the draft yourself, find a critical reader who knows the subject and ask for a sample edit or an appraisal if you can. If the central problem is structure, another pass through a grammar tool may produce tidy pages without fixing the book. If the book is sound and close to print, a proofread has a different purpose. Our guide to AI-assisted editing gives a practical way to test suggestions on a short passage before changing a whole manuscript.

There is no shame in needing help. The mistake is to claim that a chapter has been professionally edited or fact-checked when all that happened was an automated rewrite.

Be open about meaningful assistance

An author who uses AI to generate a few planning questions has used it very differently from one who asks for whole chapters and submits them with little revision. Treating those two activities as identical is unhelpful, and so is pretending the second never happened.

Our position is that authors should be candid with their publisher about substantial AI-generated text or images in a manuscript or cover proposal, and about any tool use that affects sources, translation or other people's material. The purpose is an honest conversation about quality, permissions and what still needs checking. The specific obligations in a publishing agreement or on a distribution platform must be read in their current terms, and we do not set a universal percentage of permitted AI writing.

Publishers have responsibilities too. An author may trust a publisher or editor with an unpublished memoir, interview transcripts or confidential case studies. Sending that material to an external AI service is a separate decision from reading it for editorial work. The people who own the material or are described in it deserve serious consideration, and the author should know what is being proposed. UNESCO's guidance on generative AI stresses the protection of privacy and human agency in education and research, and the same principles apply well to an unpublished manuscript.

A writer can often ask a useful question without uploading an entire book. Remove identifying information where possible, use a short sample and check the tool's current data settings and terms. With sensitive interviews or family histories, even an unnamed passage may identify someone through its details, and sometimes the right choice is to keep the material out of the tool entirely.

We would rather read the author's thinking than a perfect imitation of it

A first draft may repeat itself, and a first-time author may need help finding the centre of a chapter. Those are ordinary problems that can be worked on. A manuscript made of polished statements the author cannot explain is harder to trust, because the difficulty has been hidden rather than resolved.

A book about youth unemployment does not improve because a tool supplies impressive language about innovation. It improves when the author knows which young people the book addresses, what evidence supports each claim and what practical decisions a reader can make after closing it. A family memoir earns attention through observed detail, careful memory and honest limits, and a novel earns it through characters whose choices have consequences. Tools can assist the work around those things, but they cannot supply a real life or a point of view on request.

If you are using AI while preparing a book, keep your original drafts and review every change against them. Verify each fact the tool introduces. Read passages aloud, especially dialogue and anything in a Zambian language. Tell your editor where substantial generated material appears, and be willing to rewrite it if it does not belong. Our article on using AI to write a book shows how small supporting tasks can help without giving the whole book away.

Sotrane's position comes down to this: use a tool where it genuinely helps, and let the author's knowledge, judgement and responsibility remain visible on every page. We will judge a manuscript by what it gives the reader and whether its author can stand behind it, not by a claim that it was written "with AI" or "without AI".

Where would you draw the line between useful writing assistance and letting a tool become the author?