A Substack post titled "AI Erodes a Legacy of Reading" appeared on Hacker News last week and collected 12 points with 4 comments.
The thread centers on how large language models alter long-form reading patterns. Early comments note measurable drops in sustained attention when users default to AI summaries instead of original texts.
Core Claim from the Essay
The piece argues that repeated exposure to AI-generated condensations reduces tolerance for dense, uninterrupted text. It cites internal platform data showing average reading session length falling from 22 minutes in 2021 to 11 minutes in 2024 among heavy AI tool users.
No central mechanism is proposed; the author frames the change as an unintended side effect of convenience features now embedded in browsers and note-taking apps.
Attention Metrics Cited
- Average time on long-form articles dropped 50% for users who enable AI summaries.
- Comprehension quiz scores on 2,000-word essays fell 18% when participants read AI excerpts versus full text.
- Print book sales among 18-34 age group declined 7% year-over-year in markets with highest AI adoption.
These figures come directly from the post's referenced analytics.
How the Shift Occurs
Users encounter an article, trigger an LLM summary, and rarely return to the source. The pattern repeats across news, research papers, and books. Over months, the habit rewires expectations toward shorter, pre-digested content.
The essay contrasts this with pre-2022 reading logs where full-text engagement remained the default.
Tradeoffs Observed
- Faster information intake for surface-level facts.
- Reduced retention of nuance and counter-arguments.
- Lower ability to reconstruct arguments without external scaffolding.
Commenters on Hacker News echoed the retention concern while questioning whether the data sample was self-selected.
Comparison with Prior Tools
| Approach | Avg. Session Length | Retention Score | Source Engagement |
|---|---|---|---|
| AI summary only | 4 min | 62% | Low |
| Traditional skimming | 9 min | 71% | Medium |
| Full-text reading | 22 min | 84% | High |
The table uses numbers referenced in the thread.
Who This Affects Most
Researchers and students who previously relied on close reading of primary sources see the largest reported change. Casual readers seeking quick updates report minimal downside.
Developers building reading interfaces should consider optional "full text only" modes to preserve deeper engagement options.
Practical Steps Forward
Track personal reading logs for one week with and without AI assistance. Compare recall accuracy on key claims from each session. Adjust tool settings accordingly.
Bottom line: The Substack post and its Hacker News discussion document a measurable contraction in deep reading time tied to routine AI summary use.
The pattern suggests future interfaces may need deliberate friction to protect long-form attention rather than optimize solely for speed.
Top comments (0)