The Most Useful Page in My Second Brain is a Catalog of What I Refuse to Read

The most useful page in my second brain isn’t a record of what I’ve kept. It’s a record of what I’ve decided to refuse.

I save a lot of creator-economy content — X threads, Instagram carousels, Threads posts, YouTube clips about AI side hustles and Claude prompt packs and “how I made $10K with this one weird workflow.” Most of it is junk. I keep saving it anyway, and that’s not the mistake it looks like. The interesting question isn’t why am I saving this — it’s why a catalog of what I refuse to read is the highest-leverage page in the whole system.

This is the first post in a series on what a second brain has to refuse. It’s about a specific inverse problem: what happens when capture volume grows past the point where you can re-evaluate every save from scratch.

The problem with “save less”

If you’ve been running a second-brain setup for any length of time, you’ve hit this: the more useful the system becomes, the more aggressively you capture. The more you capture, the noisier the inbox gets. The naive answer is save less. Be more selective at the point of capture.

The naive answer is wrong, for one specific reason: real signal sometimes hides inside the noisy genre. I save creator-economy content because once in a while it contains a usable artifact — a specific prompt that turns out to be reusable, a one-line observation that names something I’d been circling around, a mechanism worth lifting. Save-less filters everything by genre; the signal inside the genre disappears with the noise.

So the actual problem isn’t volume. The actual problem is re-evaluation cost: every templated post that hits my inbox costs me processing budget to decide whether it’s the rare one with substance — every single time, from scratch. By the seventh near-identical “BREAKING: These 10 Claude prompts that built my $50K side hustle” post, I’m doing the same evaluation work I did on the first one, with no compounding.

The move is to make the refusal compound, the same way the rest of the second brain compounds.

Naming the noise

The page I’m building is called Viral Copy-Paste Templates. It does one thing: it documents recurring spam patterns I’ve seen across multiple accounts, and the mechanism each one uses to scale. Once a pattern is named, the next fifty instances of that pattern cost roughly zero to refuse — they get a one-line bundle entry, a cross-link to the template’s main entry, and they move out of my evaluation queue.

It’s an extremely boring page. It’s also among the highest-leverage pages in the system, and the explanation is the same explanation that runs through every other piece of a working knowledge architecture: you only do the work once. Memory-is-markdown works because you stop re-explaining your context to the model every session. The noise catalog works because you stop re-evaluating noise patterns every save.

Same structural move, different layer. The piece of the system that refuses has to compound or it eats you alive.

Three patterns, briefly

I’ll walk through three of the templates I’ve cataloged — not because they’re the most interesting (they aren’t) but because the mechanisms they use are worth seeing once. The point isn’t the templates themselves. The point is what a useful catalog entry looks like.

Template 1 — “BREAKING: These [N] Claude prompts…” The most common one in the AI side-hustle subgenre. Carrier post hooks with an all-caps “BREAKING:” or “CLAUDE CAN BUILD…” opener, then a numbered prompt list (usually 10+), then the last post in the thread pivots to “DOWNLOAD MY FULL COLLECTION” via a Gumroad lead-magnet URL. I’ve documented six different accounts running it within a 5-week window, with the openers verbatim-identical. The funnel converts engagement metrics into a mailing list at zero cost. The Gumroad endpoint is the real product; the prompts are the bait. Tell: the carrier post is identical across accounts. If three accounts open with the same hook in the same fortnight, it’s a distributed template.

Template 3 — Listicle-as-funnel with handle-topic mismatch. This is the most elegant one, and the one most worth seeing once because the mechanism repeats across genres. Author handle is @dailyprompter (“Artificial Intelligence | Prompts | Technology”). Carrier topic: Android camera settings. Seven numbered tips on phone photography, all genuinely useful enough to retain audience attention. Then step 8 reveals the actual product: “I killed prompt packs forever. Built a free prompt improver instead. Trained on thousands of prompts I’ve written over 2 years. Try it: dailyprompting.com.” The listicle isn’t the content. The listicle is the vehicle the audience rides into a product they didn’t know they were being marketed. Tell: when the author’s handle and the carrier topic don’t match, always read the last comment of a numbered listicle, not just the first one. That’s where the funnel lives.

Template 5 — “Comment X to receive Y” DM funnel. Generalises across platforms and languages. @thederekgray on Instagram in English: “Comment AGENT and I’ll show how to access my full strategy & AI Agent.” @edrian.exe on Instagram in Spanish: “Comenta LIBRO y te envío el enlace de preventa directo.” @ZayvenKnox on X in English with a Polymarket-bot story wrapped around the hook. Different niches, different languages, same template. The hook claims something valuable exists; a “friction wall” requires commenting a specific keyword; an auto-DM (likely ManyChat or similar) delivers the link. Comments and follows are both algorithmic boost signals, so the template pays double — engagement amplifies the post, which gets it in front of more people, who in turn comment the keyword, which amplifies it more. Tell: when the access route to content is a comment keyword rather than a link, you’re looking at a funnel, not a piece of content.

I haven’t written this catalog to expose the people running these templates. I’ve written it because a pattern named once costs nothing to refuse the next fifty times. That’s the entire point.

What survives the filter

Here’s the part that makes the catalog worth keeping rather than just a smug list.

Every once in a while, a templated post wraps a real artifact. Not often. But often enough that blanket-dismissing the genre would cost me real signal.

One account, @the.alexmethod, ran the same Claude prompt five different times within two weeks — textbook self-spam pattern, exactly the behavior the catalog is built to flag. But the prompt itself turned out to be reusable: a “Kill or Push” framework for triaging ideas that I’ve used multiple times since. The template was noise; the artifact survived.

Another account, @lennox_saint, ran a generic Income-Streams-genre “Jimmy cheatsheet” post — all the standard packaging — but buried inside was one transferable line worth keeping: “Comments are market research.” Five words; permanent rule in how I think about audience signals.

This is why the catalog matters more than blanket dismissal. The discipline is: extract the testable artifact (prompt, line, mechanism) and reject the surrounding packaging. The artifact gets a permanent home in the wiki; the genre gets filed away as documented noise. Both compound — the signal compounds upward, the pattern-catalog compounds outward.

A pure “I refuse all of this” posture would have lost both the Kill-or-Push prompt and the comments are market research line. A pure “save everything, deal with it later” posture would have buried them under a thousand near-identical Gumroad funnels. The catalog is the third move: distinguish the wrapper from the artifact, name the wrapper once, keep the artifact for good.

The harder case — when the production budget goes up

The cases above are the easy ones. Templated wrapper, small artifact, easy diagnostic: name the wrapper, keep the artifact, move on.

The harder case shows up when the same template gets performed at higher production budget. The signature is identical: credentials front-loaded, list-of-N tiers, distribution-as-moat, Like/share closer. But the underlying content is real. I’ve watched three instances of this hit my inbox in a single week.

One: a $20M-revenue AI media operator publishes a two-part “Get Rich With AI in 2026” piece. “Three Architects” in Part One (Career / Income / Venture). “Three Plays” in Part Two (Distribution / Consulting / Product). Credentials front-loaded ($5M+ in digital product revenue, CEO/founder of a 30-person AI media company). Like/Repost close. Textbook viral-template structure. The substance is also real: “be the AI guy at your company” really is the highest-percentage play risk-adjusted, the SMB-consulting arbitrage exists, distribution-before-product holds. Same template, higher production value, more accurate claims.

Two: a vibe-coder publishes a 4-agent Claude Code pipeline architecture. “How to build a 4-agent team that ships a feature while you sleep (Exact Setup Inside).” Closes by funnelling to a $29/mo managed-agent service. Both viral-template markers — the “Exact Setup Inside” parenthetical, the closer — present. But the architecture works. Planner → Coder → Tester → Reviewer, handoff via .pipeline/spec.md → .pipeline/changes.md → .pipeline/test-results.md → .pipeline/review.md, one slash-command orchestrator. The four subagent definition files are real and lift cleanly without the funnel.

Three: a personal-brand operator publishes a piece on “building an AI agent that posts on X exactly like you.” Personal-brand framing, three-tier blueprints (Session Agent → Approval Pipeline → Autonomous Stack), heavy formatting with arrows and bold callouts. Also: an extracted summary of the actual xAI recommendation algorithm with the 14 specific scoring signals the model weights for. The GitHub repo he cites is real. The signals are extractable from the code.

The same diagnostic from the easy case (the wrapper is templated; extract the artifact) also works here, but the artifact is bigger than a single prompt. The discipline is the same move scaled up: the primary-source check. Does the GitHub repo actually exist? Does the underlying architecture work without the funnel? Are the claimed numbers checkable? When the answer is yes, the substance gets a permanent home in the system and the template gets logged once in the catalog. When the answer is no, the production value was the actual content.

What makes these harder is the temptation to grade on production value. A $20M-revenue credential reads as authority; a working code example reads as substance; both can be true while the surface is still templated. The catalog has to flag both signals, real content and templated surface, without collapsing one into the other. “This works and also follows the template” is a complete catalog entry.

The compounding point

The same structural move runs through every layer of a working knowledge architecture: chat is a working surface, but the durable layer has to live in files. Your inbox of saves is a working surface too. The decisions you’ve made about what’s worth your attention have to live somewhere durable — or you’ll keep making them, expensively, every week, forever.

People build elaborate AI workflows to summarise their reading queue, classify it, surface it back to them. Some of that helps. None of it solves the underlying problem, which is that the same noise patterns keep arriving and the system keeps re-evaluating them. If you can name the pattern once and refuse it cheaply forever, that’s a compounding asset hiding inside the system you’ve already built.

If you can’t, no amount of summarisation will fix it. The model will keep dutifully summarising the same junk into a slightly tidier version of itself.

The boring page is the load-bearing one. The catalog of what you refuse is doing structural work no amount of cleverness elsewhere in the stack will replicate.

What to actually do this week

If you’ve got a knowledge system you’re already running — Notion, Obsidian, Apple Notes, a folder of markdown files, whatever — make a new page in it called something like Noise Patterns or What I Won’t Read. Leave it blank.

Then the next time you almost-save something and stop yourself with “oh, this is the same thing as that other post,” don’t just close the tab. Write down what the pattern is. What’s the carrier opener? Where’s the funnel? Why does it scale? Who else have you seen running it? One paragraph, no more. Cross-link to one example.

Repeat for the next five times you catch yourself in that “oh, this again” moment. After two weeks you’ll have a small catalog of recurring patterns, and you’ll start noticing instances faster — because the act of writing the pattern down trains you to see it. After a month you’ll be running the catalog without thinking about it, and your evaluation budget will quietly free up for the rare posts inside those genres that actually contain something.

The thing nobody tells you about building a second brain is that the part that compounds isn’t the part that captures. It’s the part that refuses, when the refusal has been thought through once and written down.

Part 1 of an open series on what a second brain has to refuse. Future entries will cover the larger epistemic-hygiene problem (when published content is shaped to bypass your filters), the boundary between healthy skepticism and dismissiveness, and the small set of sources I treat as automatic signal — and why naming them helps. (Links here when published.)

Sources

The work this post is built on is wiki-internal, but the underlying patterns are public. The receipts:

• The catalog itself — Viral Copy-Paste Templates is a living concept page in my wiki, currently documenting six templates with detection heuristics. The patterns called out in this post (Templates 1, 3, and 5) are observable on any active X, Threads, or Instagram feed if you scroll through the AI-side-hustle subgenre for an afternoon.
• @digitalshippers et al. — Template 1 canonical — the “BREAKING: These [N] Claude prompts…” opener as it ran across six unrelated accounts in May 2026. Variants by @digitalpro.motive, @digitalprofitmx, @digitalprofitspdf, @digital.profits, @promptprism on Threads — all running the same Gumroad-funnel structure.
• @dailyprompter — Template 3 canonical — the Android-camera-tips listicle with the prompt-improver reveal in step 8. The cleanest single example of the listicle-as-funnel mechanism I’ve seen.
• @thederekgray + @edrian.exe + @ZayvenKnox — Template 5 across languages and platforms — the “Comment X to receive Y” DM-funnel template generalises across English / Spanish, Instagram / Threads / X, and across niches (investing, history, polymarket bots). The template doesn’t care about subject matter.
• @the.alexmethod — the “Kill or Push” prompt — the templated wrapper was self-spam; the artifact survived the filter. Worth the extraction.
• @lennox_saint — “comments are market research” — five words; permanent rule. Templated post, real signal.
• @aiedge_ — operator-tier viral-template at $20M+ revenue scale — the “How to WIN the 2026 AI Gold Rush” + “If I Wanted to Get Rich with AI in 2026, I’d Do This” two-part bundle by Miles Deutscher. Three Architects + Three Plays. Credentials front-loaded, Like/Repost close. Substance is real-but-conventional. The cleanest current instance of same template, higher production budget.
• @zodchiii — substantive 4-agent pipeline + Teamly funnel — the “How to build a 4-agent team that ships a feature while you sleep (Exact Setup Inside)” article. Planner → Coder → Tester → Reviewer, .pipeline/*.md handoff pattern, slash-command orchestrator. The architecture lifts cleanly without the $29/mo Teamly close.
• @doublenickk — xAI algorithm extraction inside personal-brand frame — the “Build a personal AI Agent that posts on X exactly like you” piece. Personal-brand X-growth wrapping. The underlying primary source — github.com/xai-org/x-algorithm — is real. The 14 scoring signals he extracts are verifiable in the code.
• Companion piece (forthcoming) — Memory is Markdown — the structural move parallel to this one, at a different layer of the stack. Both arguments are about compounding-via-naming: memory compounds when context lives in files instead of being re-explained; refusal compounds when noise patterns are catalogued instead of re-evaluated. (Link added here when published.)
• The CLAUDE.md operating manual in my wiki — the file that tells the system how to behave, including the “refuse the obvious” rule that this post’s catalog implements at scale. Patterned on Karpathy’s LLM Wiki gist: https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f