DeepRead

What Hacker News started reading after ChatGPT

Hacker News comments up to 1 November 2026 · numbers frozen on 3 November 2026

From December 2022 to 1 Nov 2026, the Science fiction novels region gained more ground in Hacker News book recommendations than any other: from 122 to 183 of every 1,000. 10 of the 15 books that gained most sit in it, led by Children of Time (1.70 → 4.52). AI books gained too, from 28 to 36 per 1,000 as a group, but spread over many titles: none of the 15 biggest single risers is an AI book; the first, Gödel, Escher, Bach, is 18th.

At least … recommenders after
Slope chart: the books whose share of Hacker News recommendations grew most from Jan 2019 – Nov 2022 to Dec 2022 – 1 Nov 2026. Children of Time 1.70 to 4.52; Hyperion 0.72 to 2.85; A Philosophy of Software Design 1.49 to 3.35; Project Hail Mary 1.08 to 2.91; Neuromancer 1.85 to 3.51; Blindsight 1.46 to 2.98; A Fire upon the Deep 1.61 to 3.05; Crafting Interpreters 3.50 to 4.92; A Deepness in the Sky 1.58 to 2.98; Dungeon Crawler Carl 0.05 to 1.37.
Other booksAI books

Values are per 1,000 recommendations in each period. Click a line or a row to read the comments behind it.

#BookBeforeper 1,000Afterper 1,000Change
1Children of Time711.701354.52+2.81
2Hyperion300.72852.85+2.13
3A Philosophy of Software Design621.491003.35+1.86
4Project Hail Mary451.08872.91+1.83
5Neuromancer771.851053.51+1.67
6Blindsight611.46892.98+1.52
7A Fire upon the Deep671.61913.05+1.44
8Crafting Interpreters1463.501474.92+1.42
9A Deepness in the Sky661.58892.98+1.40
10Dungeon Crawler Carl20.05411.37+1.32
11Amusing Ourselves to Death471.13732.44+1.32
12Permutation City902.161023.41+1.25
13There Is No Antimemetics Division170.41451.51+1.10
14The Dawn of Everything501.20662.21+1.01
15The Cuckoo’s Egg671.61782.61+1.00
18Gödel, Escher, BachAI1182.831063.55+0.72
26AccelerandoAI330.79421.41+0.61
35ExhalationAI370.89421.41+0.52

How this was measured

What counts as a recommendation. A Hacker News comment that names a book and is labelled as recommending it. The labels (recommend, neutral, critical) come with the HackerBooks data and are made automatically, so some are wrong. Each person counts once per book per period, however many times they recommend it. "Reading" in the title means exactly this: being recommended in a comment, not being read.

The two periods. ChatGPT was released on 30 November 2022. Before is Jan 2019 – Nov 2022; after is Dec 2022 – 1 Nov 2026, the last day in the data. Both are 47 months long. The window menu narrows the after period to a single year; each window is counted on its own, so a person who recommends a book in two years counts once in each year and once in the whole period.

Why shares and not counts. Hacker News makes fewer book recommendations than it used to: 41,660 person–book recommendations before, 29,873 after. On raw counts almost every book would seem to fall. So each book is measured by its share: the people recommending it divided by all person–book recommendations in the same period, shown per 1,000. Crafting Interpreters shows the difference: 146 recommenders before and 147 after, nearly flat as a count, but a rise from 3.50 to 4.92 per 1,000 as a share.

How books are ranked. By the change in share, after minus before. A book needs at least 10 recommenders in the after window to be ranked (the buttons change this to 5, 10, 20, 40), because a share made of a handful of people swings on one comment.

New books. A book first published in 2022 or later had little or no time to be recommended before December 2022, so it rises from almost nothing just by existing. Such books are left out unless you tick the box. A book whose year is unknown stays in.

AI books. A book counts as an AI book when HackerBooks files it under AI and machine learning, or when Open Library lists artificial intelligence, machine learning, neural networks, deep learning or large language models among its subjects. Novels about machine minds are usually tagged neither way and are not guessed at. As a group, AI books went from 27.6 to 36.3 per 1,000 recommendations. When none of them reaches the top 15, the 3 that gained most are drawn as dashed blue lines, labelled with their rank.

Regions. The 13 regions of the library, the areas of the DeepRead book map; every book sits in exactly one. The Science fiction novels region went from 121.6 to 182.8 per 1,000.

What this does not show

  • Why. The split marks when ChatGPT came out. The data shows what changed after that date, not that ChatGPT caused it; anything else that changed on Hacker News or in publishing in the same years could produce the same pattern.
  • Reading. These are recommendations in comments by people who chose to comment, not a measure of what anyone read.
  • Books outside the catalogue. Only the 15,732 books in the DeepRead library are counted, and only the comments matched to them.
  • Certainty. No error bars: with 10 to 40 recommenders a book's rank can move several places on chance alone. The pattern across many books is firmer than the order of any two.

Data

Snapshot: Hacker News comments from 1 January 2019 to 1 November 2026; database 9290be31d8c7; numbers frozen on 3 November 2026. Re-running the analysis on the same snapshot reproduces every number on this page.