The Amarna Mysteries

How The Amarna Mysteries Are Made

TL;DR: In brief

The Amarna Mysteries is AI-assisted and human-authored: a human-directed work incorporating AI-generated textual material and substantial human creative authorship. The author built the series’ architecture; Anthropic’s Claude generated the first-draft prose from it, chapter by chapter, under their direction; the text was then revised over many passes to the lock, and every sentence in the published book was read, considered, revised where necessary, and approved by the author.

The Amarna Mysteries is written with AI assistance. The questions that matter to a serious reader, a journalist, or a publisher are who was in charge, what the AI actually did, and how quality was controlled.

The answer is a human author running a verification-first authorship and editorial process, one that treats every AI output as a candidate, never a verdict. This page describes that process as it made The Poisoner’s Throne, the first novel, locked for publication in November 2026, and the model behind it.

Readers will weigh the role of AI in fiction for themselves, as they should. This page exists so that judgment can rest on what was actually done, not on assumption in either direction.

AI-Assisted, Human-Authored: The Project Director Model

The Amarna Mysteries is AI-assisted and human-authored: a human-directed work incorporating AI-generated textual material and substantial human creative authorship. “Human-authored” describes the work as a whole — its conception, direction, revision, arrangement, and final form — not a sentence-by-sentence division of the prose.

AI systems generated the first-draft prose and took part in revision, continuity testing, historical checking, prose review, and the comparison of alternatives. That does not mean the books were autonomously written by AI. It means AI-generated material entered a human-directed authorship and editorial process, described step by step below.

I work under what I think of as a Project Director model. The human author remains the creative, editorial, and canon authority. AI systems are not treated as co-authors, ghostwriters, or autonomous decision-makers. They are specialist assistants. Their outputs are candidates, not verdicts.

The Project Director model was not invented for this project. I came to the series after forty years of cross-disciplinary work spanning economics, IT, telecommunications, and project management, with university-level study in archaeology. Directing specialist contributors under written procedure is what I have done for a career; here, the specialists happen to be AI systems.

The center of the system is not a single chat session. It is the project documentation: the Series Bible, change logs, operating records, chapter files, and version history. Established by the author and now dynamically maintained by AI tools as a living reference, the Series Bible tracks the canon: names, dates, relationships, chronology, historical interpretation, language rules, and settled creative decisions. Even when AI tools log and cross-reference these records, human judgment governs what becomes permanent canon.

This matters because a long historical mystery series cannot be safely governed by memory, enthusiasm, or a single AI conversation. A detail in one chapter may affect a relationship, a timeline point, a historical claim, a clue, or a reader payoff several books later.

The workflow therefore separates kinds of assistance. Drafting support is different from continuity review. Historical plausibility is different from prose texture. Structural challenge is different from canon authority. No single tool or model is treated as sufficient.

AI can be useful, but it is not infallible. It can miss real problems. It can also invent problems that are not in the text. For that reason, AI output is treated as raw material for human judgment. Findings are checked. Drafts are revised. Suggestions are rejected when they do not serve the book.

The final manuscript reflects human selection, rewriting, arrangement, verification, and approval. I take responsibility for the final expressive form of the work.

The purpose of the system is not to make AI sound authoritative. It is to make AI interruptible, checkable, and subordinate to human judgment.

How The Poisoner’s Throne Was Made

1. Architecture, by the author. Before any prose existed, I set the premise, the plot, and the characters; wrote the first Series Bible, the canonical record of every name, date, relationship, and settled decision; outlined the chapters; and fixed the rules of voice and period language, including a list of forbidden modern terms the prose may not use. The historical interpretation is mine too: which reading of the evidence the series follows where scholars disagree, and the series’ stated divergences from mainstream Egyptology.

2. Drafting. Anthropic’s Claude generated the first-draft prose, chapter by chapter, from those materials and under my direction.

3. Revision, to the lock. The draft was then revised over many editing passes. Until the end of September 2026 these ran under a written operating procedure in which separate AI systems each had a limited role: a project manager, a line editor, a closed scholarly corpus, and a continuity reviewer, each unable to stand in for another. From 30 September 2026, Claude Code, Anthropic’s coding agent, served as editor in the project’s repository, working only under my rulings: a narration fix was proposed with its evidence, tested with AI cold readers who saw the text without explanation, and executed only on my ruling; dialogue changed only on my word. The passes included forbidden-term and style censuses, Chicago Manual of Style checks, spelling and sense checks against Merriam-Webster, historical cross-checks against Britannica, and continuity checks against the Series Bible. I read the whole book in Word and edited it directly: some five hundred changes across twenty-four of its twenty-five chapters, in my own wording or in wording I chose from suggestions, landed as I made them. I then made my own hand pass of the front and back matter before it was locked. Every sentence in the published manuscript has been read, considered, revised where necessary, and approved by me.

Every outside report was treated as a claim to be checked against the text before anything changed: AutoCrit’s prose metrics, its AI reader personas and its Story Analyzer; reads by Sudowrite, Gemini, Psyncr and locally run models; and research answers from Logically.

The Tools and What Each Did

  • Claude (Anthropic): the first-draft prose; project management in the passes before 30 September 2026; from then, as Claude Code, the editor under the author’s rulings, with AI cold readers and an independent AI checker on each landing that changed the text.
  • NotebookLM (Google): before 30 September 2026, a closed scholarly corpus for historical plausibility and a continuity reviewer, each working against the author’s sources and the Series Bible.
  • ChatGPT (OpenAI): line editing in the passes before 30 September 2026.
  • Narrova (Subtxt): story-structure analysis in the early passes.
  • Logically: research against a closed, author-curated corpus of Egyptology.
  • AutoCrit: prose metrics, AI reader personas, and its Story Analyzer, all read as claims under test.
  • Sudowrite, Gemini, Psyncr, and locally run models: outside reads, treated the same way.
  • Atticus: formatting for print and ebook.
  • The cover: designed by the author; the code that renders it (SVG and HTML) was generated by Claude under his direction.

The Verification Discipline

AI checking tools are unreliable in both directions: they miss real errors, and they fabricate convincing ones — complete with quoted text that exists nowhere in the manuscript. This project documented those failure modes as they occurred and built the process around them.

Every finding from every tool was verified against the actual text before any action was taken. A literal text search is the ground truth; the tool’s verdict is only ever a candidate. Fact-level changes — names, dates, ages, historical claims — were never applied automatically; each one was a human decision. And the final gate was not a tool at all: the author’s complete read of the book before it was locked.

The methodology assumes AI tools will sometimes be wrong, and is built so that their errors are caught before they reach the reader. That is the difference between using AI and being used by it.

Historical Grounding

The series is built against the archaeological and textual record of the late 18th Dynasty, with research run against author-curated scholarly sources. Where the story deliberately departs from mainstream Egyptology, those departures are documented decisions, not accidents.

The scholarship that provides the foundation is surveyed in The Amarna Period: Egypt’s Most Volatile Decade and acknowledged in the front matter of each book as it publishes.

What This Buys the Reader

The point of this machinery is not process for its own sake. It is a series where the geography, chronology, medicine, and politics of the late 18th Dynasty hold up under a knowledgeable reader’s scrutiny; where every departure from mainstream Egyptology is a deliberate, documented choice; and where a mystery arc across several novels stays internally consistent down to the smallest planted clue. The methodology exists so the story can be trusted — and so the reader never has to think about it at all.

Where Human Judgment Governs

Characters, plot, and voice are the author’s: he set them, and any AI suggestion entered the books only on his decision. The choice of which scholarly interpretation to follow when the evidence is disputed is a human decision. What Nefertiti would say, think, or do in a given scene is a human decision. Every scene, chapter, and line of dialogue in the published books reflects human authorial judgment.

Drafting, revision, and canon-tracking were iterative, not divisible into human-written and AI-written components, and AI tools took part at multiple stages. The author does not assign sentence-by-sentence or entry-by-entry provenance to the manuscripts or the project records. The claim described below is made for the published novels only.

The copyright claim therefore does not rest on identifying passages as wholly human-written. The author’s human creative contributions to the work include the conception and overall narrative design of the series; the development of plots and subplots; the characterization and arcs of all named characters; choices of historical interpretation; the integration of historical and archaeological research into the fictional narrative; chapter architecture and scene sequencing; substantive creative revision and restructuring throughout the manuscripts; and the creative selection, coordination, and arrangement of textual material throughout the work.

These descriptions document the human creative process; they are not, in themselves, a list of protectable material. Copyright is claimed to the extent those contributions are embodied in protectable human expression, in sufficiently creative human modifications and revisions, or in protectable creative selection, coordination, and arrangement. Ideas, historical facts, and AI-generated expression considered independently are not claimed as protectable authorship.

Copyright Registration

In November 2026 the author applied to the U.S. Copyright Office to register The Poisoner’s Throne in his own name. The application discloses that the work contains AI-generated text and excludes from the claim “text generated by artificial intelligence, intermingled throughout the work and considered independently of the author’s original expression, creative modifications and revisions, and selection, coordination, and arrangement.” The claim covers the author’s original human text, his creative modifications of AI-generated text, and his selection and arrangement of the text.

This follows the U.S. Copyright Office’s guidance, including its January 2025 report on copyrightability and AI; whether a claim is registered is for the Office to decide, not this page. The cover is not part of the claim.

Publishing and Disclosure

The Poisoner’s Throne is published by The WordFish Group GmbH, Basel, under exclusive license from the author. The book’s copyright page states the AI use, and the same use is declared to its distributor, PublishDrive, for the stores it supplies.

The Book’s Own Statements

The copyright page of The Poisoner’s Throne carries these words:

“This novel was written with the assistance of artificial intelligence tools at every stage of production, including the generation of initial drafts as well as historical verification, structural analysis, and editorial revision. The author served as creative director throughout. Every sentence in the published manuscript has been read, considered, revised where necessary, and approved by the author. Full methodology disclosure is available at theamarnamysteries.com.”

“Cover design by A.J. Tilke; SVG and HTML implementation generated by Claude (Anthropic) under the author’s direction.”

And its historical-fiction disclaimer, which every book in the series carries:

“The characters in this novel are based on historical figures of Egypt’s Eighteenth Dynasty. While the setting, people, and broad political circumstances are drawn from the archaeological and textual record, their private thoughts, dialogue, and specific actions within these pages are imagined. Where the historical record is silent, the author has listened carefully before speaking.”

A Working Method

This approach was developed through practice on The Poisoner’s Throne and continues across the series: five novels so far, four historical mysteries and The Twelve Hours of Night, which follows its dead into the afterlife; the series is not closed. AI tools, used within a disciplined framework and directed by experienced human judgment, contribute meaningfully to the production of serious historical fiction. The work itself is the evidence: The Amarna Mysteries.

The author is responsible for the content of the published novels and for the accuracy of this methodology disclosure. Questions regarding the methodology may be directed to aj@theamarnamysteries.com.

A.J. Tilke is the author of The Amarna Mysteries, four historical mysteries set in ancient Egypt’s 18th Dynasty and a fifth novel that follows its dead into the afterlife. The Poisoner’s Throne (Book 1) is planned for the fourth quarter of 2026, followed by The Restoration Trilogy — The Hittite Reckoning, The Restoration Murders, and The Dakhamunzu Affair — in the second quarter of 2027. The Twelve Hours of Night will follow later.

In the novels

The Poisoner's Throne — cover

The history on this page is the ground beneath The Poisoner’s Throne, Book One of The Amarna Mysteries.