Blueprint Case Study: A Non-Fiction Book, Every Page Checked

Blueprint Case Study 2

One non-fiction book,
every page of the Blueprint.

We took a raw AI draft of a 44-page field guide through all 30 pages of THE 2026 AUTHOR BLUEPRINT, ticked each one honestly, and published what went wrong as well as what worked.

The book is The Attention Sovereignty Protocol, a short guide for people who buy books and no longer finish them. It started as a single production prompt and an AI first draft. The draft read well. Checked properly, it undersold its own promise with the wrong reading-speed maths, leaned on pop neuroscience, and pointed its QR code at a page that did not exist.

Below is the page-by-page scorecard, what the human passes caught, the finished pages, and the upload files. It is a proof edition shared as an example: not for sale, and never launched, so there are no sales results to report.

Blueprint
30 of 30 pages checked
Every page ticked, adapted or marked not applicable, with the reason.
Edit pass
73 logged changes
Five rounds, each change recorded against the untouched AI draft.
Print
44-page KDP interior
6 x 9 in, black and white, fonts embedded, text inside the margins.
Ebook
0 EpubCheck errors
The EPUB passes the W3C validator with no errors or warnings.
Page by page

The Blueprint scorecard

The statuses are strict on purpose. “Done late” means the step was done after drafting instead of before, which is the wrong order, and the scorecard says so.

PageBlueprint stepStatusWhat happened
1Cover (of the manual)Not applicableThe manual's own cover
2Execution ContractNot applicableA personal commitment for a human author
3The 2026 MarketDoneMarket facts verified for this book
4Tech StackAdaptedCover and diagrams drawn as code instead of Canva
5Deep Work EngineNot applicableBuilt in one AI-driven session
6Silo StrategyDoneBook, live bridge link, 6 social posts
7Second BrainDone lateBuilt; NotebookLM upload not done
8C-I-S-O promptsDone lateMaster prompt rewritten in C-I-S-O
9Income StackDoneNorth Star filled in; affiliates declined
1030-Day RoadmapAdaptedPlanned vs. actual recorded
11Market forensics & review mineDone late32 real review quotes; changed the book
12AI market analystDone lateVerdict: crowded but open
13Blue Ocean hooksDoneTwo options drafted; title kept
14AI Voice LabDone lateMeasured against the author's own writing
15Revenue EngineDoneBridge live; backend mapped
16Outline (Rule of 10)Adapted14 modules, same 3-part structure
17Book BibleDone lateBuilt, then used to check every number
18Vomit draftDoneGenuine, untouched AI draft
19Chapter 1 hookDonePain Mirror; the loop closes in Module 14
20Human-in-the-loop editDone73 logged edits over 5 rounds
21Human-in-the-loop surgeryDoneScrub list, drift audit, 25-word rule
22Bento grid designDoneScannability audit passed
23Formatting standardsDone6×9 interior, EPUB, phone-safe tables
24Cover briefDonePasses the 1-inch thumbnail test
25Legal shield & exportDoneAI disclosure, CMYK cover, EpubCheck clean
26Review enginePreparedStreet-team email ready
27Amazon Ads 101PreparedKeywords and ad copy ready
28Audio assetPreparedReadability check done; no audio made
29Launch dayPreparedHour-by-hour plan
30Beyond the bookPreparedEcosystem map

13 done · 6 done late (after drafting) · 3 adapted · 5 prepared, not run · 3 not applicable. None skipped silently.

Before and after

What the human passes caught

Module 4 · maths
The draft undersold the book
AI draft
“somewhere between seven and ten books a year”

Final
“close to ten books” at a careful pace, about 18 at full speed, using the measured adult average of about 238 words a minute (Brysbaert, 2019).
Module 9 · accuracy
Pop neuroscience out
AI draft
“Re-sensitize the dopamine baseline.”

Final
“Rebuild your tolerance for slow input.” Brain claims were replaced with named research readers can check.
Module 14 · links
A QR code to a dead page
The draft pointed its QR at a URL that returned a 404. A printed dead link can't be fixed later, so it now points to the live Resource Vault.
Page 14 · voice
It didn't sound like the author
Measured against the author's own writing: 10.5 contractions per 1,000 words against the author's 27.7. Now 17.2, with full forms kept where they carry emphasis.
Page 24 · cover
The first cover looked AI-made
A navy cover with a glowing beam and floating app tiles. The author rejected it. The new one is type-led, with a page marked up in pen using the book's own 4-mark system.
Final proofread · facts
A date that was a year off
A fresh-eyes proofread caught that Adler's essay “How to Mark a Book” was published in 1941, not 1940, plus a mismatch between the hours per book and the book's own reading pace.
The review mine

What real readers complain about

Blueprint Page 11 says to read the 1-star and 2-star reviews of the top competing books. We did it after drafting (the wrong order), logging 32 real negative-review quotes. It still changed the book.

Complaint 1
Too long and padded
The most common complaint. The book stays at 44 pages, and says so on the back cover: you can finish it in an evening.
Complaint 2
Preachy and condescending
Readers punish books about reading that lecture them. The voice was kept as a mentor who has had the same problem.
Complaint 3
Unsupported claims
Every number in the book now has a named source or is labelled as a rule of thumb, with a sources page at the back.

It also found a gap no competitor owns: readers saying their job trains them to skim. That became a new passage in Module 2.

The finished pages

Eight pages from the book

Front cover
Front cover
Module 1: the diagnosis
Module 1: the diagnosis
Module 4: the 15-minute maths
Module 4: the 15-minute maths
Module 6: the four pen marks
Module 6: the four pen marks
Module 12: one book, three assets
Module 12: one book, three assets
Module 13: the reset ladder
Module 13: the reset ladder
Printable 30-day tracker
Printable 30-day tracker
Back cover
Back cover
What the test taught us

Five lessons about the Blueprint itself

Lesson 1
The order matters
Validation done last still changed the book. Done first, it would have shaped the draft instead of patching it.
Lesson 2
Book Bible before the draft
Drafting without one produced an unsourced reading-speed figure that undersold the book's main promise.
Lesson 3
Measure the voice
Two editing rounds still left the prose at about a third of the author's contraction rate. Only measuring showed it.
Lesson 4
Order a printed proof
The CMYK soft proof showed the screen violet prints noticeably duller. A printed proof copy is part of the export, not an extra.
Lesson 5
Automation needs a human read
A contraction script produced two grammar errors that only a page-by-page read caught. That is what the human-in-the-loop pages are for.

See also the first case study: one picture book, start to finish. The first draft of this book was written by AI (Claude) and then edited, fact-checked and designed under the author's direction; the diagrams and cover were also drawn by AI, as code.

The 2026 Author Blueprint · Coming Soon
The same system, for your book.
Join the waitlist