A sister publication · EXHIBITA
19 rules for making something spread, across 5 stages — each marked twice. Once for how confident we are, and once for what a corpus of published claims did to it when it was checked. The marks are the point, not the advice.
A separate body of work with a separate method, kept under its own name. 9 of the 19 rules are settled; 4 are contested or folklore and are printed as such rather than quietly dropped.
How to read it
The first mark is ours: how confident we are in the rule. The second is the corpus's: what happened to that rule's tactic when published claims about it were collected and checked against things that actually spread. A rule can be confidently held and undercut by the evidence at the same time — those are the interesting ones, and nothing here hides them.
The second mark describes the tactic a rule belongs to rather than the sentence itself; a tactic verdict is computed from many claims, and attaching it to one line would overstate it. Hover any mark for what it rests on.
01 · Before you make anything
Most posts fail here, before a single word is written. This is the highest-leverage stage and the one almost no advice covers.
Make something you could make again next week.
A one-off that spreads brings borrowed attention that leaves as fast as it arrived. A recognisable format brings the same people back, so each post starts from a higher floor than the last.
settled unmeasured short · long · image · text · community
One idea per post. If you can't say it in a sentence, it isn't ready.
Sharing requires the sharer to explain why. A post carrying three ideas gives them nothing to say, so it gets liked and not passed on.
settled undercut short · long · image · text · community
Aim for a specific feeling, not 'engaging'.
Posts that spread tend to provoke high-arousal responses — awe, delight, indignation, surprise. Interest alone is low-arousal and rarely moves anyone to share. The effect is real but the size varies enormously by audience.
mixed undercut short · image · text · community
02 · The packaging
Title, thumbnail, first line. The most-discussed stage by a wide margin, and the one where advice is most often overfitted to someone else's audience.
Open on the most interesting moment, not the setup.
The decision to keep watching or reading is made almost immediately, and every platform measures it. Context can be filled in afterwards; attention cannot be recovered once lost.
settled unmeasured short · long · text
Concrete and specific beats clever and vague.
A reader decides in a glance whether a post is about something they care about. Wordplay hides the subject; specificity reveals it. This is the single most reliable improvement available to most posts.
settled no lift short · long · image · text · community
Open a loop — but close it.
Withholding creates a pull to keep watching. It also creates the strongest negative signal on any platform when the payoff doesn't arrive, and repeat viewers learn quickly. Practitioners disagree sharply on where the line sits.
contested no lift short · long · text
The thumbnail has to read at the size of a fingernail.
Almost everyone sees it small, on a phone, in a scrolling feed. Detail that reads on your monitor disappears entirely at delivery size.
settled no lift long · short · image
A number helps — when it's a real one.
Numbers signal scope and finish, which is why listicles persist. The effect is modest, decays as everyone does it, and reverses when the number is obviously padded.
mixed no lift long · text · image
03 · The post itself
What happens after someone stops. Packaging buys the first two seconds; this stage decides whether the platform shows it to anyone else.
Optimise for finishing, not for starting.
Every major platform weights completion and watch time far above the click. A high stop-rate on something nobody finishes teaches the algorithm the wrong lesson and suppresses the next post.
settled unmeasured short · long
Make it the shortest length that still lands.
Shorter usually helps completion, but a post cut below its own payoff performs worse than a longer one that delivers. Length is a consequence of the idea, not a target to hit.
mixed unmeasured short · long · text
Deliver the promise, then add one more thing.
Ending exactly on the payoff invites an immediate exit. A short additional beat after it gives the viewer a reason to stay to the end, which is the signal that matters.
mixed undercut short · long · text · community
Caption everything.
A large share of feed video is watched with sound off. Uncaptioned video is simply invisible to those viewers, and this is one of the few genuinely uncontested points in the whole field.
settled no lift short · long
04 · Publishing and the first hour
The stage most heavily mythologised. Some of what follows is real; a well-known part of it is not.
Be present for the first hour.
Early engagement is how most platforms decide whether to widen distribution. Replying quickly raises the comment count and keeps the post surfacing. The effect is real; claims about its exact size usually aren't.
mixed unmeasured short · long · image · text · community
Post when your audience is awake. Beyond that, ignore the charts.
Universal 'best time to post' tables are among the most repeated and least supported claims in the field. Once audience size is controlled for, the effect largely disappears. Your own analytics beat any published chart.
contested unmeasured short · long · image · text · community
A few relevant tags. Nothing is gained by thirty.
Hashtag-count advice is close to pure folklore, and the tested claims that exist mostly find no effect. Tags help categorisation, not distribution.
folklore unmeasured short · image · text
Repurpose, but re-cut for each platform.
The same idea can work everywhere; the same file usually can't. Reposts carrying another platform's watermark or aspect ratio are demoted on most services.
mixed unmeasured short · image · text
05 · Staying viral
What separates a spike from an audience. Almost nothing in circulation covers this, which is why most people who go viral once never do it again.
Have the next one ready before the first one peaks.
A spike sends people looking for more. If there is nothing there, the attention disperses and does not come back. This is the most common and most costly mistake made by people who break out.
settled unmeasured short · long · image · text · community
Send them somewhere you own.
Reach on a platform is a lease that can be revoked by a ranking change. An email list, a newsletter, a site — anything with a URL you control — is the only way a spike becomes an audience.
settled unmeasured short · long · image · text · community
A rate you can sustain beats a sprint.
Posting more does raise your number of chances. It also degrades quality and burns people out, and the evidence for any specific cadence is weak. The sustainable rate is the one that matters.
contested unmeasured short · long · image · text
Caught spreading
Collected off the open web while the claims above were being checked. These are specimens, not exemplars — the point is what travelled, which is not the same as what was good. The figures are not comparable to each other: each is whatever its own platform counts, named beside it, and the ones carrying no figure were never measured at all.
The Amazon tax684 points
Google has acquired the data of failed US airlin536 points
Cursor launches Origin, GitHub alternative346 points
Using the railway network as a flatbed scanner339 points
Memory prices climb 500% in 12 months331 points
Babies born under sugar rationing grew into adul223 points
Teaching my kid to code with a modern MUD197 points
Norway Should Buy OpenAI135 points
Python Polars Cheatsheet (based on our O'Reilly 133 points
California's new tire efficiency rules could sav113 points
We've flown a radiation-blocking vest to the Moo60 points
Pacing model development in an era of cyber-crit29 points
2,500-year-old sculpture discovered at UNESCO si27 points
How to organize 3 acquired companies into one co13 points
Mycorrhizal Infrastructure Map9 points
Visiting the History of Computing and Play5 points
llms.txt: a proposed standard no major AI platfo4 points
IndieWeb Homebrew Website Club Asia Pacific: Ref4 points
Emulating ALiBi with Rope4 points
Drug dealer jailed after taking gun to probation
How reusable heat packs work
Android ported to reMarkable's color tablet
Figures are whatever each surface publishes — searches, points, comments — and are not comparable across platforms. 378 specimens held in all.
The claims themselves
The articles the marks above were computed from — every one that shipped a picture. These are the claims, not the findings: read them as the field talking to itself.
How to Get More Followers on Instagram in 2026Buffer Resources
13 Trending Songs on TikTok in August 2026 (+ Buffer Resources
Leonardo da Vinci Would Be Invisible in the AgJeff Bullas
AI SEO tools that fit your growth stackHubSpot Marketing
Growth experimentation: A guide for growing maHubSpot Marketing
Movie that went viral for terrible animation bHacker News: viral
X algorithm insights highlight key posting strSocial Media Today
Brand tracking tools for scaling companiesHubSpot Marketing
How we made a viral commercial for developersHacker News: viral
The YouTube algorithm just had a huge change… Think Media
New all-time high thanks to old video? IBM stoHacker News: viral
Viral Halloween 'De-Flock' campaign gains momeHacker News: viral
I Had 200+ Unread Articles in My List — So I BBuffer Resources
Top 11 Social Media Engagement Tools to Build Buffer Resources
The Fastest Way to Get Monetized on YouTube ShTubeBuddy
This Might Be the Best Youtube Shorts Update ETubeBuddy
How an Independent Agency Manages 23 Social MeBuffer Resources
YouTube’s Algorithm Has Changed Again (Full BrThink Media
Running Doom on Our Custom CPU and Going ViralHacker News: viral
Show HN: Mock Chats – Make viral chat-story viHacker News: viral
Ultra-rare Star Wars Lego collection went missHacker News: viral
The skeptic's guide to humanoid robots going vHacker News: viral
YouTube Just DROPPED a Netflix-Style Feature fTubeBuddy
Build your SEO strategy as if search engines dSearch Engine Land
Brands shake up back-to-school marketing to reMarketing Dive
Honey loses over 7M users and 7k stores followHacker News: viral
YouTube is changing how it counts viewsSocial Media Today
Youtube Shorts NEW Revenue ChangesvidIQ
If You're Worried About The New YouTube MonetvidIQ
Stop treating every customer touchpoint the saMarTech
YouTube Is Harder Now… But There’s a Huge OppoThink Media
118 articles held in all; 36 shown, the ones carrying a picture. Their claims are what the second mark is computed against.
On the feeds now
Straight off the same kind of source the corpus was built from, refreshed with the rest of the site. Unmarked, because nothing has been checked yet — that is what the marks above are for.
The engagement illusion: What actually gets atMarTech
Snapchat puts the “social” in social media by Tubefilter
The answer to your martech problem depends on MarTech
How to create B2B content that buyers keep usiMarTech
AI is putting marketing’s in-house model to thMarTech
28 Digital Marketing Tools for Small BusinesseBuffer
Your marketing agent is working. Can you see wMarTech
How YouTube’s Algorithm Can Find Your CustomerSocial Media Examiner
MIPCOM 2026 expands creator economy programminTubefilter
In the sports world, influencers are facing baTubefilter
YouTube wants creators to pitch brands with guTubefilter
From the cozy games to the horror hits, TwitchTubefilter
Live from MarTech, Tubefilter, Buffer, Social Media Examiner. Treat these as claims in the wild, not as advice — none of them carries a mark.
Where it comes from
118 articles making claims about what spreads, 378 specimens of things that did, across 35 feeds. Collected 2026-08-18, 24 days ago.
The claim count is a property of the network on the day, not of the field: three runs one afternoon returned 159, 119 and 70. The scraper refuses to overwrite a corpus with one under three-quarters its size for exactly that reason.