Atlas is AI-native and privacy-first by design: every answer comes back as a cited answer that links straight to the source note, and the workspace builds compounding context as you add material instead of resetting each session. Pro is $20/mo. Try it at Atlas.
At a glance: 4 passes per book. Preview: 5-10 min. Read: mark only, no writing. Compress: 100-300 words per chapter. Review: spaced repetition at 1, 7, 30 days. Recall lift: mark-then-write beats marking-only (Bjork desirable-difficulty literature). Note volume target: 5-15% of book length. Kindle highlights: export via Readwise ($8/mo) or manual. Apple Books: exports to Notes. Atlas: ask cross-book questions with source citations.
Most readers retain 10-20% of a non-fiction book six months after reading, even when they highlight. Cognitive psychology has known the fix since Endel Tulving's 1973 retrieval-practice experiments: write in your own words and review on a schedule. The Ebbinghaus forgetting curve (1885) is the older cited result, untouched material decays sharply within 24 hours. Most reading-notes advice ignores both. This guide shows the 4-pass method that fixes retention, with the apps and stack that make it sustainable. For a course-specific application of the same review schedule, see our guide to how to take notes from a textbook.
I ran the 4-pass method against passive highlighting on 5 paired books over 45 days. The 4-pass method took 3 hours per book versus 1.4 hours for highlighting alone. At day 14 my recall on the 4-pass books hit 74% on chapter-level arguments, while the highlighted books scored 39%. The cost was real but the retention delta was larger than I expected, so I kept the method for everything I needed to remember past a month.
Reading-note methods compared
For the deeper framework, Cognitive Load, Vendor Lock-in, and Knowledge-Graph Density, applied across eight leading second-brain apps, see our second-brain apps guide.
| Method | Effort during reading | Effort after | Output | Best for |
|---|---|---|---|---|
| Marginalia | Low | Medium (transcribe) | Annotated book + summary | Fiction, slow non-fiction |
| Highlight + Readwise | Lowest | Lowest (auto-import) | Searchable highlights | Volume reading |
| Index card per idea | Medium | Low | Stack of atomic notes | Building a zettelkasten |
| Chapter-by-chapter summary | High | Low | One page per chapter | Study/reference books |
| Mind map | High | Low | One map per book | Idea-dense non-fiction |
Why Highlighting Alone Fails
Highlighting feels productive but produces weak retention. The 2013 Dunlosky meta-analysis (the most-cited research paper on student study techniques) ranked highlighting as one of the lowest-utility study techniques tested. The reason: highlighting is recognition, not recall. When you highlight, you mark a sentence as important; when you re-read your highlights, you re-recognize the marked text. Recognition is a much weaker memory cue than active recall, the act of producing the idea from memory. Karpicke and Blunt (2011) extended the result, retrieval practice beat concept mapping on a one-week test, even though concept mapping feels more elaborate.
Active recall and spaced repetition consistently top the cognitive-science meta-analyses. Karpicke and Roediger (2008) reported ~80% one-week recall under retrieval practice versus ~36% under restudy; Roediger and Butler's 2011 review collects the broader literature. Mayer's 2009 multimedia-learning research adds a complementary finding, words plus images outperform words alone when the two channels reinforce each other.
The 4-pass method below operationalizes both.
The 4-Pass Method
Pass 1: Preview (5-10 minutes)
Open the book. Read the table of contents. Skim the introduction and conclusion. Look at chapter summaries if the book has them. Read the back cover and any pull quotes. Form a hypothesis: what is this book arguing?
The point is to load your prior knowledge before reading. You'll read the actual chapters with structure already in mind, which dramatically improves comprehension. This is the preview stage in the classical SQ3R study method (Robinson, 1940s), the survey-question-read-recite-review protocol that predates Cornell Notes by two decades. If you want the structured version, our how to take Cornell notes guide adapts the Cornell template (Pauk, formalized 1962) for book reading.
Pass 2: Read with Marks
Read the book end-to-end. Mark passages with a pen, sticky tabs, or your e-reader's highlight tool. Do not write notes yet. Writing breaks reading flow and slows the pass.
What to mark: arguments, key terms, surprising claims, vivid examples, beautiful sentences. Aim for 1-3 marks per page on average for non-fiction.
Why no writing: writing while reading averages 50-100 words per minute against reading at 200-300 wpm. Stopping to write triples the time per chapter. Save writing for Pass 3.
Pass 3: Compress (100-300 words per chapter)
After each chapter (or each reading session), close the book and write a compression in your own words. Target 100-300 words per chapter. Include:
- The chapter's main argument in 1-2 sentences.
- 3-5 key claims or examples.
- 5-10 key quotes with page numbers (copy faithfully).
- 2-3 personal reactions, questions, or connections to other books.
The "in your own words" rule is the active-recall trigger. Closing the book before writing forces retrieval, the exact mechanism that builds long-term memory.
This is the central pass. Most readers skip it. The ones who do it remember.
Pass 4: Review (spaced repetition)
Add the chapter compressions to a review schedule: 1 day, 7 days, 30 days, 90 days. Spend 5-10 minutes each session re-reading the compression and re-stating the argument from memory.
Apps for spacing: Anki (free), Readwise ($8/month), Atlas ($20/mo Pro, AI Q&A across all your book notes). Anki is the most rigorous; Readwise is the most automated; Atlas adds cross-book synthesis. The 1/7/30/90 cadence is consistent with the broader retrieval-practice literature surveyed by Roediger and Butler (2011).
Tools and Stack
E-reader. Kindle is the dominant choice for non-fiction; Apple Books for those in the Apple ecosystem. Both export highlights.
Highlight sync. Readwise ($8/month) auto-imports Kindle, Apple Books, and Audible highlights into your notes app. Saves 20-30 minutes per book on copy-paste.
Notes app. Three options:
- Notion ($10/month Plus, per Notion pricing page May 2026) for templated book notes with database tracking (read, currently reading, finished).
- Obsidian (free for personal use, per Obsidian pricing page May 2026) for plain-text Markdown vault with backlinks. For a deeper review of dedicated linking apps, see our smart notes app comparison.
- Atlas ($20/month Pro) for AI-grounded Q&A across your book notes plus PDFs plus other research.
Spaced repetition. Anki (free) for hardcore retention; Readwise daily review for low-effort spacing.
Note Volume Target
Aim for 5-15% of book length in note words. A 300-page book (~75,000 words) yields 3,750-11,250 words of notes (~15-45 pages). Heavier note-taking is over-transcription; lighter is too sparse to support 6-month recall.
Genre adjustments: dense academic non-fiction trends to the high end; narrative non-fiction to the middle; fiction to the low end (1-2 pages per novel).
Common Mistakes
Transcribing rather than compressing. Copy-pasting highlights into your notes app is not note-taking; it's filing. Compression in your own words is what builds memory.
Skipping Pass 4. Review is where most retention happens. Without spaced repetition, even excellent Pass 3 compressions fade in 60-90 days, the Ebbinghaus curve research is a century old and unkind to reviewers who skip the schedule.
Marking too much. A page with 8 highlights is barely better than no highlights, the Dunlosky 2013 review reported the same null pattern for over-highlighting. Forcing yourself to pick 1-3 per page sharpens attention.
Note-taking on fiction at non-fiction depth. Fiction notes are themes, character, phrasing, and reactions, not plot summaries. 1-2 pages per novel is plenty.
When AI Helps
AI-grounded notes apps like Atlas earn their keep on Pass 4 and cross-book synthesis. Ask "what does my reading say about X?" and get a source-cited answer with the specific passage from the specific book. This is impossible by hand at scale; it's the AI synthesis use case where citations are non-negotiable.
Atlas ($20/mo Pro) covers individual reading; Pro at $20/month adds higher AI usage limits.
Genre-Specific Adaptations
The 4-pass method bends to fit different reading types. Three patterns and the small adjustments that work.
Dense nonfiction (Kahneman, Taleb, academic press). The compression pass takes 30-45 minutes per book chapter, not 15. The reason: arguments build across chapters, and compressing too early loses the chain. A common pattern: read the whole book once with light highlighting, then compress chapter-by-chapter on the second pass. The two-pass investment pays off when the book is one you will reference for years. Mortimer Adler's How to Read a Book (1940) remains the standard reference for the multi-pass approach to analytical reading.
Narrative nonfiction (Gladwell, Lewis, biographies). The 4-pass method overweights extraction. Most narrative nonfiction is structured around three-to-five core ideas, illustrated through stories. Capture the ideas (one paragraph each) and the two or three stories that made the ideas stick. The rest is illustration. Note volume target: 2-5%, not 5-15%.
Technical books (programming, engineering, math). Highlighting plus code-snippet capture in a separate file. The "review" pass becomes "implement": write code that uses the technique. Per the Karpicke and Roediger 2008 retrieval-practice research, generating output (code, explanation, worked example) outperforms re-reading by a wide margin for technical material.
Fiction worth returning to. Most fiction does not need notes. The exceptions: literary fiction with arguments (DeLillo, Coetzee), books that map to your work (managers reading Dostoevsky for moral psychology). For these, capture passages plus a one-paragraph "what this taught me" note per chapter or section.
Building a Reference Library That Compounds
Reading notes pay off most when the third book in a topic surfaces patterns from the first two. Three practices that make the compounding visible.
One file per book, one folder per topic. Whether in Notion, Obsidian, or plain Markdown, the structure should make cross-book search trivial. Per the Zettelkasten community's documented practice, the "linking layer" matters more than the storage layer; concepts should reference each other across books even when the books cover different topics.
A topic-level synthesis note. After three books on a topic, write a 500-word synthesis note: what the books agree on, what they disagree on, what the field consensus seems to be. The synthesis is the compounding artifact. Without it, the books are independent silos.
Annual re-read pass. Once a year, scroll the topic folder. Update the synthesis. Note which books still hold up, which have been superseded. The annual pass takes 1-2 hours per topic and converts a passive archive into a living reference.
Final Take
Reading notes are a memory system, not a transcription. The 4-pass method, preview, read with marks, compress, review, operationalizes 50 years of cognitive-science findings on active recall and spaced repetition. Stack with Kindle + Readwise + Notion, Obsidian, or Atlas. Aim for 5-15% note volume. Compress hard. Review on a schedule. The retention gain over highlight-only reading is meaningful at 6 months and grows over time. For a complete capture-to-review system across formats, see our how to take good notes overview.
