AI Can Help You Write a Book—but It Can Also Make Your Book Sound Like Everyone Else’s

AI writing tools can be useful to an author. They can sort a folder of notes, suggest questions a reader might ask and point out where a timeline is unclear. They can also produce smooth, confident paragraphs that could belong to almost any book, and occasionally state facts that are not true. Both outcomes come from the same strength: these tools are very good at producing plausible language, and plausible is not the same as true or personal.

Ask a chatbot for the opening of a memoir about growing up in Kitwe and it may offer something like "In the tapestry of life, every challenge becomes a stepping stone towards destiny." The sentence is fluent and says nothing about Kitwe or about the writer. Compare a remembered detail: after heavy rain, a girl and her brother walk around the deepest puddle on the way to school, because the mud can swallow a shoe and their mother will be furious if they arrive for assembly with soaked socks. That memory is not yet a finished paragraph, but it belongs to one person. A tool can help arrange it or question it. It cannot remember the walk.

Sotrane Publishers has set out its position on AI in book writing: tools are welcome where they help the writer think, and a book must remain the work of a person who knows what it says and can answer for it. This article looks at what that means at the writing desk.

Give the tool a job you can judge

A request such as "Write my book about resilience" asks the software to fill a large empty space with plausible language. The result may sound confident, because confidence is easy to imitate. A narrower request lets the author judge whether the help is any good.

Suppose a retired teacher has recorded several hours of memories about her first posting to a rural school. She might ask a tool to group her own notes by possible themes, show where the timeline is unclear or suggest questions to answer in a scene that is missing. She then decides what happened, what matters and which order serves the reader. If a suggested chapter plan ignores the school and community that shaped the story, she discards it.

A business author might ask for the questions a beginner would still have after reading her explanation of stock records. That can expose a gap, but she still has to answer from her own practice and check the explanation with someone who keeps such records. A novelist might ask for a list of possible consequences of a character's choice, and then decide which one follows from the people already on the page.

The difference lies in the task. "Find the questions I have not answered" keeps the author thinking. "Write a compelling chapter in my voice" can quietly replace both the thinking and the voice. Professional editing also asks hard questions, with a human editor accountable for what they have read, and neither kind of help removes the author's responsibility for the book.

Treat generic prose as a warning sign

Readers recognise certain patterns: "in today's world", "a journey of self-discovery", "unwavering determination" and a closing paragraph declaring the story "a testament to the human spirit". These phrases do not prove that someone used AI, since people wrote vague prose long before chatbots existed. The trouble is that they make one book interchangeable with another.

A Zambian memoir does not become specific by adding "Lusaka", "the Copperbelt" or "Africa" to those phrases. The reader needs a decision, a place with consequences, a relationship or a detail that changes the meaning of a scene. Instead of "My mother taught me resilience", the author might remember her mother walking back to the market after a failed sale because there was still supper to find. Which market, what was being sold and what the child understood at the time are questions only the writer can answer. Never invent the detail because a vivid sentence looks better.

In a practical book, vagueness costs the reader something. "Entrepreneurs should leverage innovative solutions to maximise their potential" sounds purposeful but tells a small trader nothing. An author who has actually managed stock can explain what to write down when goods are bought, what to record when something is sold on credit and which figures to check before ordering more. The example should match the author's experience and the reader's circumstances.

A simple test helps. If a tool produces three paragraphs that could be moved into any other book without changing a word, ask what real material they are standing in for, then return to your notes, interviews, records or observations.

Plausible language is easy to produce. A book needs language that is true, specific and the author's own.

Check every fact the tool supplies

A confident answer can contain a date, a quotation, a book title, a regulation or a research reference that is wrong or does not exist at all. Do not place it in a manuscript because the wording looks scholarly. Trace it to the original source and check what that source actually says. If you cannot find it, remove the claim or write honestly about what you do know.

Consider an author writing a guide for first-year university students in Zambia. A tool supplies a precise figure for dropout rates and names a report as its source. The figure would make a dramatic opening, but the responsible step is to find the report, check its date, definitions and coverage and decide whether it applies to the students the book addresses. If that cannot be done, the book can still describe the problem through verified interviews or the author's own clearly identified experience. It does not need an invented percentage.

The same caution applies to quotations attributed to well-known Zambian figures, the spelling of local place names, publishing procedures and the current rules of online platforms. A sentence shaped like a citation is not a citation. Where a subject changes over time, check current primary sources. Our ISBN guide shows why a publishing detail should be confirmed for the actual book and format rather than copied from a confident summary. A useful working rule is that if you would be embarrassed to show a reader where a claim came from, it is not ready for print.

Keep the language that sounds lived in

Writers sometimes paste an informal draft into a tool and ask it to "make this professional". The result removes repetition and grammatical errors, but it may also remove humour, local phrasing and the way particular people speak. The author sees cleaner sentences and misses what was lost.

A market trader in a novel who jokes with a customer while weighing tomatoes may speak with affection, irritation and a local rhythm all at once. Converting her lines into polished international business English changes the relationship, however correct the grammar becomes. Keep the sense and the context, and edit only where a reader would genuinely be lost. A book set in Zambia does not need to explain Zambia to Zambians as if they were visitors.

The same applies in non-fiction. A pastor's teaching book, a lecturer's study guide and a young poet's collection should not all sound like the same speechwriter. Look for the places where your own vocabulary, opinions and observations remain. If the tool drafts a sentence you would never say and cannot defend, do not keep it simply because it sounds impressive.

Read a passage aloud beside your original notes. Do you recognise yourself in it? Can you explain why each example was chosen? If a reader asks what happened next, can you answer from knowledge rather than asking the tool for another paragraph? Our article on ghostwriting and voice offers a useful comparison, because even a human collaborator must check every page against the author's own account.

Think before uploading someone else's story

An unpublished manuscript may contain phone numbers, school records, private family conversations, interview transcripts or an organisation's confidential material. Pasting the whole file into a third-party tool can disclose information the author was trusted to keep. Before uploading anything, read the tool's current terms and privacy settings and ask whether you have permission to share the material. If the task does not need identifying details, remove them or work with a short anonymised passage.

A memoirist drafting a difficult family scene could ask for help organising its themes without supplying a relative's full name and medical history. A lecturer working from student interviews should consider what was promised to participants before placing raw transcripts in any outside system. Changing names alone may not protect someone who can still be recognised through their town, job and circumstances. Our article on memoir and family privacy explains how to make those choices even when no AI tool is involved. No setting can replace a decision about what you are entitled to share.

Review an AI-assisted chapter before calling it finished

Take one chapter that a tool helped with and mark four kinds of sentence. The first kind reports a fact, names a source or makes a checkable claim, and each of those needs verifying against reliable material. The second describes a person or event you know, and the words should reflect only what you can honestly say. The third is a general statement that could appear in any book, which should be replaced or cut if it adds nothing. The fourth sounds elegant but does not move the story or argument forward, and you should be able to say why it is there.

Then give the chapter to a reader who knows the subject or genre but does not know which parts a tool drafted. Ask where the passage becomes thin, where an example is missing and where a sentence does not sound like you. Do not ask them to "spot the AI". The aim is a stronger and more truthful book.

Keep earlier drafts and any prompts that help you remember how the manuscript developed. If you are working with a publisher, editor or collaborator, tell them where substantial generated material appears and check the exact terms of any agreement rather than guessing. Sotrane's own position asks authors to be candid about substantial AI-generated text or images, so that quality, permissions and checking can be discussed honestly.

AI can untangle notes, propose a structure or reveal a question you missed. Those are worthwhile jobs. The book still needs a person who knows what happened, has something to say, checks the evidence and chooses words that belong to the people and places on its pages.