the algorithm doesn't want your best work.

we ran an experiment on ourselves. for six weeks, every time we shipped a post, we rated it on a 1–5 scale before we hit publish. simple rubric: did we like it? would we save it if a competitor posted it? then we left it alone for thirty days and pulled the data.

the result was not subtle. the work we'd rated 5—our favourites averaged 23,000 views. the work we'd rated 2—the stuff we shipped because the calendar said today averaged 71,000. the gap was three to one in favour of the work we didn't believe in.

you can read this two ways. one: we have terrible taste. two: the algorithm has a different job than the one we keep accusing it of. both are true. but the second one is where the leverage is.

— section onethe algorithm is not a critic.

we keep talking about platforms as if they have editorial taste. as if there's a tiktok that prefers craft, a meta that respects nuance, a youtube that rewards depth. they don't, they don't, and they sort of do but not in any way you can plan around.

what platforms actually optimise for is a single physics question: given two people in this exact second, will the next swipe come faster on this post or the next one? that's it. that's the whole job. quality, in the human sense of the word, is incidental to the question. sometimes it correlates. usually it doesn't.

the algorithm is not asking "is this good." it is asking "will this delay the swipe by 1.4 seconds." those are different questions and you have to choose which one to write for.

the work we rated 5 was almost always slow-build. it had a pre-amble. it asked you to remember something from earlier in the carousel. it had a punchline that required you to have read the setup. all of which is great editorial work and terrible swipe-physics.

the work we rated 2 was, embarrassingly, built for the swipe. hook in 800 milliseconds. payoff at the four-second mark. nothing required from the viewer except their attention, briefly, while the slot-machine handle was being pulled.

— section twowhat we changed about the brief.

after week three, we re-wrote our internal content brief. the old brief had a section called "creative ambition" which is exactly as embarrassing as it sounds. we deleted it. we replaced it with two questions:

the brief is now two questions. the answer is yes/yes, or we don't ship. not "is it creative," not "is it on-brand," not "is it strategic." those are downstream of these two. if the post can't survive the swipe, none of the other questions matter. if it survives the swipe but blurs the brand, it's a borrowed-attention loan we'll pay back with interest.

the post-mortem changed too

we used to look at views, saves, comments, shares. four metrics, weighted intuitively. that's a dashboard, not a decision. now we ask one question after every post: did this post make the next post easier or harder to write?

it's a strange metric and it took us a while to operationalise. but it captures something the dashboard misses. a viral post that doesn't compound — one that exists for itself, not for the next one — is a tax. it sets a hook the next post has to pay off. if the next post can't pay it off, the viral post becomes a noose.

— section threewhat we stopped doing.

three things we cut from the workflow, in rough order of how much pain they were causing:

• the "best work" pile. we used to keep an internal slack channel of pieces we were proud of, separate from what we'd shipped. it had become a graveyard. we deleted it. if it's worth keeping, it's worth shipping — and if we won't ship it, it's not worth keeping.
• the awards-style retrospective. end-of-quarter recaps that highlighted "creative wins." they were morale theater — they made us feel like artists, which made the work worse, because feeling like artists is downstream of being read, not upstream.
• the brief sentence "something different this week." ambient pressure to vary, with no theory of why. it produced posts that were neither in-format nor out of it. neither here nor there. nothing kills compounding faster than reflexive variation.

and three things we added

1. the swipe-test before any creative review. play the first 1.5 seconds with the sound off and the screen at half-brightness. if you, the person who made it, want to swipe, ship it back to the editor.
2. the brand-survivor test after the swipe-test. if you removed the logo, the handle, and the watermark, would a regular viewer guess which kind of account this came from. if no, it's a borrowed-attention post.
3. a weekly “why didn't we ship this” review. for every five posts that shipped, we look at one we didn't — and we name the actual reason out loud. not "wasn't ready." not "didn't fit." the real reason, which is usually that we got embarrassed about something thin and called it "not strong enough."

— section fourthe part that's hard to admit.

here's the thing none of this solves. some of the posts that win are bad. by every meaningful measure of craft, they're worse than the posts they outperform. they're shorter, dumber, more obvious, more derivative.

and we ship them anyway. because the algorithm is a distribution channel, not an editor, and the job of a distribution channel is to distribute. using the channel without acknowledging its physics is just a slow way of going broke. so the practice now is: build the body of work in two places at once. the channel-shaped pieces ship every day. the craft-shaped pieces ship occasionally, and they're priced — in time, in budget, in expectation — differently.

the body of work is not the channel feed. the body of work lives somewhere else — the case studies, the talks, the long-form, the dispatch. the channel feeds it traffic. it does not contain it.

this is the bit that makes creative people nervous and it should. we are saying, in plain language, that the medium we publish in every day is not the medium we are building in. if that sounds like cheating, fine. it's also true.

— closingthe question you should ask yourself.

here's the prompt we'd offer if you're running a content team and any of this is uncomfortable: what's your ratio? what percentage of your output is for the channel, what percentage is for the body of work, and have you ever named those buckets out loud?

most teams haven't. most teams have one bucket called "content" and a vague hope that quality and reach are the same thing. they aren't. they were maybe, briefly, in 2015. they are not in 2025. the sooner you separate the two, the sooner the calendar stops feeling like a punishment.

our ratio, currently, is about 85/15. eighty-five percent for the channel, fifteen for the body of work. we're trying to push the body-of-work number up. we'll let you know how it goes.

this essay is part of the dispatch — one piece of writing on social, content, and the operations of running both, sent on the last friday of every month. no roundups, no listicles, no spam. subscribe here →