"Treat the highlights as an extra emphasis since I resonated with those parts. It is like going through a book and underlining something; you only underline or highlight if it resonates with you. Also, generally, if I highlight something, there's a near certainty that it fits somewhere into a mental model of mine, or connects to something I've already heard of in the past or would like to explore deeper in the future. It works as a very reliable signal."
What this means in practice
A highlight in the user's brain is not equivalent to a citation or a quote pulled from the text. It is:
A resonance flag — the user marked this as connecting to something they already care about or want to know more about.
A signal of mental-model membership — the highlighted passage fits a framework the user is building. It is more likely to be load-bearing in their thinking than a random passage.
A signal of future-exploration interest — the user picked it because they expect it to connect to things they will dig into later.
Reliable enough to weight as primary — the user explicitly said "near certainty." Treat this as a strong prior.
Operational rules
When a brain page (book, podcast, article, video) is being created or updated and the source has user highlights:
Highlights go ABOVE the LLM's own picks in the page hierarchy. They are weighted higher than synthesis-generated quotes.
The "Highlights" / "User resonance" section gets a header that signals its primary status — not a footnote, not an appendix. The page opens with synthesis, then highlights, then LLM-picked quotes.
Highlight pages in zotero-to-brain pipelines should NOT just be a "passes through" data point — they are the spine. A page with no highlights is weaker evidence of relevance than a page with highlights, even if the synthesis is excellent.
Cross-link from highlights to existing brain pages when a highlight matches a known concept (the user has explicitly asked for this kind of linking — see "connects to something I've already heard of").
Highlight density on a source is a relevance signal at the corpus level. A book with 12 highlights is more likely to be useful than a book with 0, all else equal — independent of how good the synthesis is.
Anti-patterns
Treating highlights as decoration (showing them in a small italic block at the bottom of the page).
Treating highlights as data (counting them, sorting by recency, treating as a popularity metric without weight).
Replacing user-picked highlights with LLM-picked quotes that look similar but weren't selected by the user.
Synthesizing over highlights without first reading them — the user has stated they want the highlight-resonance treated as primary. Reading the highlights IS the first step.
Class-level generalization
This pattern applies to all user-generated content, not just books:
Podcast episodes with Snipd saves — the Snipd saves are resonance markers.
YouTube videos with manual timestamps or "liked" — the timestamps/likes are resonance markers.
Articles saved to read-later or bookmarked — the save is a weaker resonance marker (often done by default), but explicit "re-read later" tags are strong.
Voice notes with explicit "I want to come back to this" framing — strongest resonance marker of all.
Whenever the synthesis pipeline encounters source content with user-generated resonance markers, the markers are primary signal, not decoration.
Source
User statement (verbatim, 2026-07-05): "treat the highlights as an extra emphasis since I resonated with those parts. It is like going through a book and underlining something; you only underline or highlight if it resonates with you. Also, generally, if I highlight something, there's a near certainty that it fits somewhere into a mental model of mine, or connects to something I've already heard of in the past or would like to explore deeper in the future. It works as a very reliable signal."
Context: response to the question of how to handle existing highlights-only books (e.g. Golden Son) when introducing the synthesis-first format for full-text-extracted books.