Tony Wang9 min readGoodreads' Own Data Says 2018 Outpublished 2019 Through 2024. A Reprint Mill Explains Why.
A publisher labeled 'Generic' relabels its reprint year as publication year — explaining 96% of every Leather Bound book in a 20.9M-record Goodreads crawl.
If you asked Goodreads' own crawled data which year saw the most books "published," the answer is 2025 — narrowly ahead of 2018. Not 2023, not 2024, not any year in between. 2018 alone outranks every single year from 2019 through 2024, a seven-year stretch that includes the peak of the ebook and self-publishing boom.
That's not how publishing works. So we pulled the raw records behind the number, and one very specific print shop explains almost all of it.
A curve that shouldn't exist
Publication-year counts should roughly track real publishing volume — rising over time, dipping only for the most recent, still-incomplete years. Goodreads' own data doesn't do that.
2025 leading is at least plausible on its face — new releases, momentum, a currently-hot year. 2018 beating six later years, including 2023 and 2024, isn't plausible on any reading of real publishing activity.
Ruling out the boring explanation first
Before trusting a weird shape, we checked whether it was a Crawlora-side measurement problem — the same check that caught real bugs in two other datasets we tested this month. We cross-referenced the facet's year counts against an exact date-range search:
Exact matches, both years. The anomaly is real data, not a broken aggregation — which means the explanation has to live in what the records themselves actually say.
Tracing the spike: one format, one publisher, one boilerplate sentence
Cross-tabbing publication_year against format finds the mechanism immediately. Filtered to format=Leather Bound:
96.3% of the entire format
The individual records explain exactly why. Nine spot-checked titles all came from a publisher listed as "Generic" — itself a placeholder-looking value — sharing near-identical templated description text:
Show all 9 spot-checked titles
| Title / author | publication_year | True original year |
|---|---|---|
| Sophocles, Philoktetes | 2018 | 1851 |
| Marco Polo, The Travels | 2018 | 1855 |
| Émile Zola, Le Rêve | 2018 | 1910 |
| Jean de la Fontaine, Fables | 2018 | 1913 |
| George Orwell, Nineteen Eighty-Four | 2018 | 1949 |
| Victor Hugo, Les Misérables | 2018 | pre-1862 |
| Marguerite Henry, Stormy, Misty's Foal | 2018 | 1963 |
| Benito Pérez Galdós, Doña Perfecta | 2022 | 1925 |
| Thérèse of Lisieux, Historia de un alma | 2022 | 1926 |
The description text says it outright, on the record:
"Leather Binding on Spine and Corners with Golden Leaf Printing on round Spine ... Reprinted in 2018 with the help of original edition published long back [1851]." — Sophocles' Philoktetes,
publication_year: 2018
"Reprinted in 2022 from the original edition published many years ago [1926]." — Thérèse of Lisieux, Historia de un alma,
publication_year: 2022
This isn't fabricated data. Every one of these is a real, crawled book listing with a real description. But for tens of thousands of records, the publication_year field means "when this specific print-on-demand operation ran off a new leather-bound copy of a public-domain classic," not "when this book was first published" — and nothing in the field name, schema, or facet response signals that distinction. publisher: "Generic" alone totals 154,607 records across the whole 20.9M-book dataset — 0.74% of the catalog — but it's concentrated almost entirely into this one seven-year window, which is exactly how a publisher that small ends up visibly distorting a year-level chart.
So does this explain 2025 too? No.
The obvious next question is whether the same reprint-mill pattern explains 2025's even bigger peak. It doesn't, and we're not going to pretend it does.
Leather Bound isn't even in 2025's top-10 format list — Kindle Edition leads at 258,784, followed by Paperback (234,463), Hardcover (70,357), and Audible Audio (54,201). Whatever's driving 2025, it isn't a reprint mill stamping public-domain classics.
We checked a second, simpler hypothesis — that 2025 skews toward non-English or translated editions, which might proliferate faster for currently-popular titles — and it doesn't hold up cleanly either: English makes up 59.9% of 2025's total (382,929 of 639,617) versus 57.8% of 2010's (163,778 of 283,377). That's a real but modest 2-point difference, nowhere near enough to explain a whole-year outlier on its own.
Two more things worth knowing before you cite this data
Genres are multi-label, not a partition. Fiction leads the genre facet at 767,506, Nonfiction at 443,907 — and Audiobook ranks 3rd at 408,810. A book carries 8-10 genre tags on average in our samples, mixing genuine genres with format labels like "Audiobook" or "Ebooks." Genre counts don't sum to the catalog size and shouldn't be read as a clean distribution.
This is a crawl, not Goodreads' full catalog. ~20.9M records is a real, large, live number — but it describes what curated lists, search sweeps, and author-bibliography expansion have reached, not a claim about how many books exist on Goodreads overall.
What it's for
Public-domain reprints are a real, useful category of books — this isn't an argument against print-on-demand classics, and every individual record we checked is a genuine, accurate listing on its own terms. The problem is purely aggregation: a publication_year field built to answer "when was this edition printed" gets read as "when was this work first published" the moment you sum it into a year-over-year chart, and the two questions have very different answers for a public-domain classic.
Book catalog data — titles, authors, series, genres, format, and the full ratings/reviews breakdown — is queryable at /datasets/goodreads over one REST API. If a "books published per year" number ever looks too flat, too spiky, or too convenient, it's worth checking what format and publisher are actually driving it before you chart it.
Frequently asked questions
Was 2018 really a bigger year for book publishing than 2019 through 2024?
No. Goodreads' own crawled publication_year data shows 2018 (627,704 books) outranking every year from 2019 through 2024, including 2023 (526,586) and 2024 (523,921) — but this isn't real publishing volume. 96.3% of every Leather Bound-format book in the dataset (241,403 of 250,618) is dated 2018-2024, traced to a print-on-demand reprint operation that labels its reprint year as publication_year instead of the work's true original publication date.
How do you know the publication_year field itself isn't just broken or capped?
We cross-checked it first. Both spike years' facet counts matched an exact date-range search total exactly — 639,617 for 2025, 627,704 for 2018 — ruling out a measurement bug. The anomaly is real underlying data, not a Crawlora aggregation artifact, which is what let us trace it to its actual source instead of dismissing it as a crawl error.
What's the actual mechanism behind the 2018/2022 publication-year spike?
A specific publisher, listed as 'Generic' in the data, reprints public-domain classics in a leather-bound format and stamps the reprint run's year as publication_year. Sample records quote it directly: 'Reprinted in 2018 with the help of original edition published long back [1851]' for Sophocles' Philoktetes. Other titles caught the same way include works by George Orwell (1949), Victor Hugo, Marco Polo (1855), and Marguerite Henry (1963) — all relabeled with a 2018 or 2022 publication_year.
Is Goodreads' 2025 publication-year peak explained by the same reprint-mill mechanism?
No — and we're reporting that honestly rather than force-fitting an explanation. Leather Bound format isn't even in 2025's top-10 format list (Kindle Edition and Paperback dominate instead). A translation-share hypothesis was also checked and doesn't hold up cleanly (59.9% English in 2025 vs. 57.8% in 2010, too small a gap to explain a whole-year outlier). The leading unverified hypothesis is that currently-popular books simply get more editions discovered by this crawl's search-sweep-based methodology.
Is Goodreads' genres field a clean breakdown of book categories?
No, it's multi-label. A book typically carries 8-10 genre tags mixing genuine genres with format labels — Audiobook ranks as the 3rd-largest 'genre' at 408,810 books, right behind Fiction and Nonfiction. Genre counts don't sum to the catalog size and shouldn't be read as a distribution.