How Goodreads broke book reviews

Q: Are there any reliable book-review sources?

Yes — NYT Book Review, Kirkus, Publishers Weekly, and Booklist are institutional professional-critic sources with editorial rigor and substantially less inflation than crowd platforms. They're the backbone of Rankquant's book source weighting.

Q: Does Goodreads' rating have any information in it?

Yes, but it's most useful as a popularity signal (which books a large engaged audience read and rated, regardless of specific score) combined with comment-level sentiment analysis rather than the raw average. A Goodreads 4.7 can tell you a book has passionate fans; it cannot tell you whether the book is better than another book rated 4.6.

Q: What about DNF (did not finish) rates?

Goodreads does publish DNF statistics, and they're more informative than star ratings in some ways — a book with a 50% DNF rate has failed a large fraction of its readers, regardless of the 4.4 average. Rankquant's book methodology will incorporate DNF rates as a secondary signal where the platform exposes them.

Q: Does Amazon book reviews matter?

Moderately. Amazon reviews on books are known to be manipulable (paid ARC reviews, AI-generated reviews, author-friend voting), but the sheer volume provides a noisy-but-useful signal. Rankquant weights them at 2 (versus 9 for NYT Book Review) for that reason.

By Ryan Siegal · Founder and Principal

Published 2026-04-23

The three failures

1. The 1-5 scale is too coarse for books

Books have enormous quality variance. A great novel might resonate with a reader for decades; a mediocre one is forgotten by the next weekend. Compressing that variance into five discrete buckets erases most of the information a reader would need to compare two books.

Amazon uses the same 1-5 scale and has comparable inflation problems, but books are worse because the top three buckets ("it was amazing" = 5, "really liked it" = 4, "liked it" = 3) all feel socially acceptable, while the bottom two ("it was OK" = 2, "didn't like it" = 1) carry a reputational cost most reviewers avoid. Academic studies on the platform have found that > 85% of reviews are 3+ stars and > 60% are 4+ stars across the full catalog.

2. Friend-network voting amplifies positive bias

Unlike Amazon (reviews are anonymous-ish and transactional) or Letterboxd (reviews are read by a cinephile peer group), Goodreads shows reviews to the reviewer's literal friend network first. A 1-star review on Goodreads is visible to your book-club. A Kirkus critic panning a novel is a professional opinion; a reader panning it is an awkward conversation at next month's meeting.

The incentive structure pushes readers toward DNF ("did not finish") and no-review for books they dislike, rather than a 1-star review. The negative tail of the rating distribution is artificially suppressed.

3. Readers pre-select into books they expect to enjoy

Unlike headphones or hotels, which people buy out of necessity, readers choose books they already expect to like — often on the recommendation of a friend, a trusted list, a BookTok/BookTube influencer, or the book's own cover and premise. The reader base for any given book is a positively-selected population before the first page is read.

This is different from Amazon's electronics sections, where people buy products because they need a laptop, regardless of whether they'll love it. On Goodreads, the readers most likely to log a book are the readers most predisposed to enjoy it. Self-selection bias is baked into the platform.

4.1/5 avg

Literary fiction average rating on Goodreads. The nominal midpoint of the scale is 3.

Goodreads Year-in-Books aggregates

4.3+/5 avg

Genre fiction (romance, fantasy, YA, thriller) average ratings. All well above the scale midpoint.

Goodreads category audits, 2024

<3%

Fraction of books on Goodreads rating below 3.5 stars. The bottom half of the scale is nearly empty.

Aggregate audit of Goodreads ratings

Why StoryGraph, Hardcover, Fable didn't fix it

A wave of post-2020 Goodreads alternatives have tried to address user-experience failures (bad search, ugly UI, no privacy) but none have changed the underlying scoring mechanics. They're mostly using 1-5 or 1-10 scales with friend-visibility and self-selection biases intact. Same inflation problem, different UI skin.

What actually works: professional + crowd, weighted + normalized

The right approach for book reviews combines professional editorial sources (NYT Book Review, Kirkus, Publishers Weekly, Booklist, LitHub, LA Times) with crowd platforms (Goodreads, Amazon) — but weights them by credibility, applies Bayesian prior adjustment for low-review books, and z-scores against a per-genre peer set.

Rankquant's source weights for books:

NYT Book Review — 9
Kirkus Reviews — 8
Publishers Weekly — 8
Booklist — 7
LitHub / NPR / LA Times — 6
Goodreads — 3 (high volume, low rigor per review)
Amazon book reviews — 2 (known manipulation)

Normalizing this way inside genre peer sets (contemporary literary, historical fiction, fantasy, memoir, etc.) recovers the signal that a raw Goodreads average throws away. A book that's 4.3 on Goodreads might land at 2.8 normalized (below-average for its genre despite high raw score) or at 4.7 normalized (genuinely exceptional). The number actually tells you something.

A book's normalized score only means something if we show you what we compared it against. Every review page exposes the peer set — "Normalized within: contemporary literary fiction, 2020-2024, hardcover-first (2,147 books)" — so you can judge the comparison.
— Rankquant editorial policy

Frequently asked questions

Are there any reliable book-review sources?+

Yes — NYT Book Review, Kirkus, Publishers Weekly, and Booklist are institutional professional-critic sources with editorial rigor and substantially less inflation than crowd platforms. They're the backbone of Rankquant's book source weighting.

Does Goodreads' rating have any information in it?+

Yes, but it's most useful as a popularity signal (which books a large engaged audience read and rated, regardless of specific score) combined with comment-level sentiment analysis rather than the raw average. A Goodreads 4.7 can tell you a book has passionate fans; it cannot tell you whether the book is better than another book rated 4.6.

What about DNF (did not finish) rates?+

Goodreads does publish DNF statistics, and they're more informative than star ratings in some ways — a book with a 50% DNF rate has failed a large fraction of its readers, regardless of the 4.4 average. Rankquant's book methodology will incorporate DNF rates as a secondary signal where the platform exposes them.

Does Amazon book reviews matter?+

Moderately. Amazon reviews on books are known to be manipulable (paid ARC reviews, AI-generated reviews, author-friend voting), but the sheer volume provides a noisy-but-useful signal. Rankquant weights them at 2 (versus 9 for NYT Book Review) for that reason.