Open dataset · Amazon KDP · 2026
540 books. €30,702. The catalog grew, and the money per book collapsed.
The full de-identified ledger behind WriteAIBook's 350-book case study: 540 titles from one KDP catalog, including 524 AI-era releases from March 2025 onward. Sixteen months of royalties through April 2026, every book row included.
The finding: more titles, less per title
Publishing volume worked early. It stopped working as the catalog grew. Revenue per catalog book peaked at €16.51 in July 2025, when the catalog held 142 titles. By March 2026 it was €4.14 across 528 titles, a 75% drop. In April 2026 it fell to €2.26.
The fairest cohort comparison is first-month revenue, because every cohort gets the same 30-day window. Median first-month revenue fell from €13.80 for July 2025 books to €4.10–€5.70 for the November 2025–March 2026 cohorts.
Lifetime hit rates point the same way, but they are not a fair cohort comparison: the later books are younger. As of the cut, 66.7% of July 2025 books had passed €50 lifetime, versus 6.7% of the November 2025–April 2026 group. Use the first-month column for the clean signal.
What 16 months of royalties look like
Catalog health, in five labels
Each title is labelled by its last-90-day royalties at the data cut. The thresholds are descriptive, not a recommendation to delete books.
| Status | Books | Share | Definition |
|---|---|---|---|
winner | 34 | 6.3% | last 90 days > €30 |
active | 151 | 28.0% | last 90 days €10–30 |
watch | 137 | 25.4% | last 90 days €2–10 |
dead | 121 | 22.4% | last 90 days < €2 |
too_new | 97 | 18.0% | published after January 2026 |
“Dead” means under €2 in the last 90 days, not zero lifetime revenue. One title had zero lifetime revenue at the cut.
By publication cohort
The July 2025 cohort is the outlier on both first-month and lifetime revenue. Later cohorts started lower in their first month, before the observation-window difference can explain the gap; lifetime columns are as of the cut.
| Cohort | Books | Lifetime | Median lifetime | Median first month | ≥ €50 lifetime |
|---|---|---|---|---|---|
| 2025-03 | 34 | €1,971 | €31.56 | €3.90 | 26.5% |
| 2025-04 | 10 | €775 | €76.28 | €13.00 | 80.0% |
| 2025-06 | 28 | €2,116 | €37.17 | €8.40 | 42.9% |
| 2025-07 | 54 | €4,738 | €82.42 | €13.80 | 66.7% |
| 2025-08 | 76 | €4,332 | €51.37 | €10.90 | 52.6% |
| 2025-09 | 82 | €3,806 | €43.70 | €7.50 | 43.9% |
| 2025-10 | 46 | €2,512 | €42.87 | €8.80 | 45.6% |
| 2025-11 | 33 | €696 | €21.94 | €4.30 | 3.0% |
| 2025-12 | 32 | €788 | €24.41 | €5.70 | 6.2% |
| 2026-01 | 32 | €916 | €24.99 | €4.10 | 9.4% |
| 2026-02 | 43 | €886 | €16.15 | €5.50 | 9.3% |
| 2026-03 | 42 | €780 | €16.30 | €5.00 | 7.1% |
| 2026-04 | 12 | €90 | €5.70 | €0.00 | 0.0% |
AI-era cohorts only. Legacy titles published before 2025 are included in the row-level CSV but excluded from this table. Median first month is the fair comparison; lifetime figures for recent cohorts are younger. Full 23-cohort data is in cohorts.csv.
What is in the dataset
Every public file is de-identified. Titles, ASINs and pen names are not included, and the private mapping is not published. The row-level file has 540 rows, not a sample.
| File | Size | What it contains |
|---|---|---|
catalog_books.csv | 540 rows | One row per de-identified title: anonymous IDs, cohort, status, lifetime and last-90-day revenue. |
monthly.csv | 16 rows | Monthly royalties, active books, catalog size and revenue per catalog book. |
cohorts.csv | 23 rows | Books, lifetime revenue, median revenue and hit rates by publication month. |
pennames.csv | 8 rows | Anonymous pen-name clusters, with no names. |
status.csv | 5 rows | Catalog status definitions and counts. |
revenue_bands.csv | 6 rows | Distribution of titles by lifetime revenue band. |
summary.json | 1 object | Recomputed headline metrics, including the corrections. |
datapackage.json | 1 object | Machine-readable schema, license and resource list. |
checksums.sha256 | 12 lines | SHA-256 hashes for every data and documentation file. |
METHODOLOGY.md | document | Source, definitions, reconciliation and limitations. |
DATA_DICTIONARY.md | document | Field-by-field documentation for every file. |
README.md | document | Overview, key numbers, file list and citation. |
LICENSE.md | document | CC BY 4.0 terms and attribution. |
The ZIP contains the CSVs, JSON summary, data package, methodology, data dictionary, README and license.
Methodology, briefly
Source
Amazon KDP royalty reports from one publisher account, consolidated through 2026-04-19. The catalog is primarily Amazon.de and priced in EUR. It contains fiction and some nonfiction, published under several pen names.
Revenue
Lifetime and last-90-day royalties include eBook sales and an estimated Kindle Unlimited page-read component of €0.0027 per KENP page, calibrated to the reported April total. Revenue is gross royalties, not profit. It does not subtract generation costs, editing, covers, ads, refunds or chargebacks beyond what KDP reports.
Inclusion
All 540 titles in the analysed catalog are included. 524 were published from March 2025 onward (the AI-era cohorts); 16 are legacy titles from 2019–2022. The original 350-book milestone is a mid-experiment snapshot, not a separate dataset.
De-identification
Titles and ASINs are removed. Pen names are replaced with anonymous IDs; Unicode variants of the same name are merged. Book IDs are assigned after sorting by cohort, pen and ASIN; the mapping is not published. The raw workbook stays private.
Reconciliation
The 540 book rows sum to €30,702.00 lifetime and €6,251.85 last-90-day revenue. The 16 monthly rows sum to €25,395.00; the remaining €5,307 is pre-2025 legacy revenue not shown in the monthly table. Cohort rows sum to 540; status rows sum to 540.
Corrections
Earlier posts cited 350 books and roughly €18,000 for the first six months. This release supersedes those figures with the full catalog through April 2026. A summary-sheet peak of €3,825 for September 2025 was inconsistent with the monthly table; this release uses €3,671.
What this dataset cannot tell you
- It is one publisher's catalog, not a random sample of Amazon KDP. Do not use it to estimate the average author's earnings.
- It cannot prove that AI-generated books caused the decline, or that volume causes lower quality. Genre mix, covers, ads, seasonality and KDP algorithm changes are not controlled.
- Cohort lifetime comparisons are age-biased: the November 2025–April 2026 books have had less time to earn. The first-month column controls for that; the lifetime hit rates do not.
- It does not include genre, ad spend, production cost, page reads by title, or profit. A €5 generation cost is not subtracted from the revenue figures.
- KENP revenue is partly estimated, and Amazon can revise royalties after the data cut. The last 90 days are a snapshot, not a final settlement.
- “Winner” and “dead” are descriptive thresholds based on last-90-day revenue, not judgments about a book's quality or long-term value.
- De-identification removes titles and ASINs, so you cannot join this file to Amazon product pages or verify an individual book.
- It is a historical snapshot. The catalog may have changed after 2026-04-19; this dataset does not describe its later state.
Download
All files are free, require no email, and are released under CC BY 4.0. Credit WriteAIBook and link to this page.
Cite this
WriteAIBook. (2026). WriteAIBook KDP Catalog Study: 540 Titles From an AI-Assisted Catalog [Data set]. https://writeaibook.com/research/kdp-book-study. CC BY 4.0.
BibTeX
@misc{writeaibook_kdp_study_2026,
author = {{WriteAIBook}},
title = {{WriteAIBook KDP Catalog Study: 540 Titles From an AI-Assisted Catalog}},
year = {2026},
howpublished = {\url{https://writeaibook.com/research/kdp-book-study}},
note = {De-identified dataset, data cut 2026-04-19, CC BY 4.0}
}Dataset version 2026-04-19. If you use the data in an article, link to this page and say that the revenue figures are gross royalties from one publisher account, not a KDP-wide sample.
Questions
Why does the study say 350 books when the dataset has 540?
350 was a mid-experiment milestone. The dataset covers the full catalog through 19 April 2026: 540 titles, of which 524 were published in the AI-era cohorts from March 2025 onward.
Does the dataset include titles, ASINs or pen names?
No. Each title has an anonymous book ID and each pen name is replaced by an anonymous pen ID. The private mapping is not published.
How is revenue calculated?
Lifetime and last-90-day royalties in EUR, including an estimated Kindle Unlimited page-read component (€0.0027 per KENP page) calibrated to reported totals. Revenue is gross royalties, not profit.
Is this a random sample of KDP?
No. It is one publisher's catalog, primarily on Amazon.de. It is useful for catalog economics, not for estimating the average KDP author.
Can I reuse the data?
Yes, under CC BY 4.0. Credit WriteAIBook and link to this page.
Want the workflow behind the catalog?
WriteAIBook generates an editable manuscript chapter by chapter. The first chapter is free with 25 welcome credits, no card. The data above is what one catalog did; your results will depend on genre, quality and marketing.