Most reading systems are designed around saving. Mine is increasingly designed around reuse.
Saving is easy. I can mark a paragraph, clip an article, bookmark a thread, and feel like I captured something. But capture is just the start. A saved idea that never changes a decision or a draft is mostly storage.
The Readwise workflow exists to push reading past storage.
The source material starts as marked passages and notes across books, articles, essays, and papers. I don't treat those marks as the full book or the full argument. They are evidence of what caught my attention at the time. That distinction keeps the output honest. A book-summary post is not "everything this content says." It is what I captured, why it mattered, and which ideas seem reusable.
That framing changes the writing. It makes the note personal and focused. I do not need to pretend I have produced the definitive summary. I need to show the trail between the original passage and my interpretation.
The workflow starts with the source. Local Readwise exports are preferred when they exist. Live Readwise data fills gaps. Duplicates get deduped. Counts matter because they tell me how much evidence the note stands on. A short note with 15 saved passages should feel different from one built on 150.
Then I look for the themes. The useful question is not "what did I save?" but "what pattern keeps showing up?" Sometimes the pattern is operational: how a company built a sales motion or how a founder made decisions. Sometimes it is personal: how to think about energy, fear, ambition, or attention.
After that, the synthesis has to become a note someone else can read. This is where a lot of AI-assisted writing goes wrong. It becomes too smooth. It sands off the texture of the source material and replaces it with generic lessons. I want the opposite. The note should keep the specificity of the content while making the ideas reusable.
The cleanup pass is where the note becomes publishable. I check for first-person leakage that sounds like the agent speaking as me, weak phrasing, overconfident claims, and inflated prose. The humanizer score is a useful diagnostic, not a moral authority. A low score tells me where to look; it does not get to rewrite the piece blindly.
Only then does it become a Ghost draft. The draft gets a clear title, tags, source link when appropriate, and a final verification pass. If it is based only on saved passages, the text should say so. The reader should not be misled about the source.
What makes the workflow matter is not that it produces posts quickly. Speed is nice, but speed alone just creates a faster pile. The point is that reading becomes part of a larger operating system.
A book note can later feed an essay. An essay can feed an about page. A profile can reuse an operating pattern from a book. A deep dive can borrow a question from a reading note. A content recommendation can point back to an old passage that suddenly matters again.
That is when reading compounds. Not when the library grows, but when the ideas become available for future work.
The workflow also changes my relationship to unfinished reading. I do not need every content to become a post. Some content produce one useful idea. Some produce a dozen. Some produce nothing worth publishing. That is fine. The system is not a trophy case for completion. It is a filter for usefulness.
The best reading notes have three qualities.
They are grounded. You can see they came from actual saved passages, not a generic summary floating above the book.
They are interpretive. They don't restate the material. They say what seems useful, surprising, questionable, or reusable.
They are connected. They point beyond themselves into the rest of the work: company building, operating cadence, sales, product, AI, personal systems, and judgment.
The best notes also keep a little of the original encounter. I want to remember why the idea caught me, not the category it belongs in. A clean summary is useful for search. A note with a point of view is useful for thinking. The difference is small on the page and large in practice.
The bad version of this workflow is automated content. That is easy and mostly worthless. The better version is a personal reading memory that becomes public when it has enough shape.
This is part 3 of 8 in Operating a Public Notebook.