Write to the thesis, not the algorithm

Almost everything published online is written to an algorithm. Not because the people writing it admire ranking systems, and not because they lack taste. It is written that way because of a simple fact about where the work lives: the channel decides whether the work travels, and the channel rewards what is cheap to count. A click is cheap to count. Time-on-page is cheap to count. A keyword match against a query is cheap to count. So the work gets shaped toward those quantities — and the shaping is invisible, because it is done by competent people who believe, correctly, that they are doing their jobs.

This is the essay where we say what the other end looks like. We do not write to be ranked. We write to be cited. The difference is not a matter of degree or polish. It is a difference in what the work is for, and it has to be settled before the first sentence, because the two targets shape the same sentence in opposite directions.

Two targets, one sentence

Start with the distinction, because the distinction is the whole argument.

Work written to rank is shaped by what the channel cheaply counts: the keyword the query will match, the hook that wins the click, the structure that holds attention long enough to register as a session. Because those quantities are generic, the work that chases them becomes generic too. One affiliate roundup of "the best X of 2026" is interchangeable with the next thousand. No specific person stands behind a specific, falsifiable claim, because a falsifiable claim is a liability in a piece whose only job is to be the page you land on. The page is the product. The claim inside it is filler.

Work written to be cited is shaped by a different question: will a specific claim, with its evidence and its named author attached, survive being lifted out of this page and carried somewhere else? That is a harder thing to make. It requires that there be a claim — one you are willing to attach your name to and be wrong in public about. It requires the evidence to travel with the claim, so the quote does not collapse into an assertion the moment it leaves home. The unit is not the page. The unit is the claim, and the page is just where the claim currently happens to live.

We made this concrete for ourselves first. Every Pagecut essay is built so its claims, its evidence, and its named author come apart from the page cleanly — readable, and machine-readable, without the visual layer. That is a capability, not a result. It says the work is built to be cited; it does not say the citations will arrive. We will come back to why that distinction matters.

Why the two targets pull apart

It would be convenient if you could write to rank and to be cited at once, and sometimes a single piece does both. So be precise about the claim: this is not a law of physics. Work that ranks well and gets cited widely exists. The claim is about selection pressure, which is weaker than a law and more reliable than a preference.

When you optimize a body of work primarily for what the channel cheaply counts, you are applying a steady force in one direction across thousands of small decisions — which topic, which headline, which sentence stays, which claim gets softened because it might cost a click. None of those decisions is decisive on its own. In aggregate they bend the work away from the things that make a claim citable: specificity, falsifiability, a named position. The pressure does not forbid citability. It just never selects for it, and over enough decisions, what is never selected for disappears. This is the same mechanism essay #5 named from the other side: “Noise is an org chart” argued that noise is manufactured when an editorial function is subordinated to a distribution function that answers to the channel. The output of that subordination is exactly the work described here — ranked, generic, authorless in the way that counts. The structure is the cause; write-to-the-algorithm is what it produces.

So the practical claim is directional. Optimize a publication primarily for ranking and it will, over time, drift away from being citable — not because any one piece must, but because the pressure never pushes the other way. You do not have to choose perfection. You have to choose which direction you are willing to be pulled.

The sorting test

A distinction you cannot apply is decoration. Here is the line, applied to real cases.

On the write-to-rank side: the affiliate "best X of 2026" roundup, structured to capture a high-intent search query, where the claims are swappable across a thousand near-identical pages and no author stands behind any of them. The content-farm explainer in the Demand Media and eHow lineage, where the topic was selected by query volume and handed down as a keyword brief — written to own the results page, never to be quoted as the source of anything. The engagement-bait feed post, tuned to the dwell-and-reply signals of a feed, where the "insight" is shaped to provoke a reaction and dies the moment the feed moves on.

On the write-to-be-cited side: Stripe Press and Stripe's long-form writing, built as durable, attributable arguments that people quote and reference years later, where the claim — not the pageview — is the asset. Gwern.net, which is radically extractable and citation-first, pairing claims with evidence and explicit authorship and remaining indifferent to any feed. A canonical RFC, or a well-argued primary source: the thing people quote the claim from and link back to, whose entire distribution is being the citation rather than winning a ranking.

The cases sort cleanly on one question: is the page the product, or is the claim the product? That is the line. Everything else is implementation.

The bet we are actually making

Now the honest part, because this is where it is easy to overclaim.

We are betting that the durable unit of distribution is the extractable, attributable claim rather than the ranked page. That is a bet, not a finding. It is broader than anything we can prove today, and we want to state it as what it is: a wager on thesis-logic. The reasoning is that a ranked page is rented from whatever counts it, and the rent can change without notice, while a claim that has been quoted and traced back to you is held by the people who carried it — a different kind of asset with a different kind of durability. We find that reasoning convincing. We do not present it as measured.

There is a market read sitting underneath the bet, and it deserves the same honesty. The intuition is that summarizing and citing surfaces — the ones that answer a question and name a source rather than handing back ten blue links — are taking on more of the work that the click used to do. We think that is happening. We are not going to hand you a number for it, because we do not have one we would stand behind, and a publication whose whole argument is "write claims you will defend" cannot smuggle in an undefended one about its own market. Treat it as our editorial read, not as a measured trend. If it is wrong, the bet above still has to stand on its own logic, and we think it does.

The objection worth answering

The sharpest version of the counter is not "rankings matter." It is: refusing to write to the algorithm is a luxury, and ignoring distribution is how good work dies unread. We take that seriously, because half of it is correct.

Distribution does matter, and writing to be cited is not the same as ignoring distribution. Essay #5 put the two functions in an order: editorial-first, distribution-aware, in that sequence. Distribution-aware means we care a great deal whether the work reaches a reader; it does not mean the channel gets a vote that overrides the claim. The line we are drawing is not between caring about reach and not caring. It is between letting reach shape the claim and letting it carry the claim once the claim is made. Make the claim citable first; then move it however you can. The order is the whole point, and reversing it is the move that manufactures noise.

The luxury charge is the part that has teeth, and we will not wave it off. Writing to be cited is more expensive per piece, slower to pay off, and it asks you to forgo a click you could have had today. For a team measured on this quarter's traffic, that is not a luxury, it is a risk. We are taking that risk deliberately, with our own publication, on the bet stated above — not recommending it to anyone whose structure cannot absorb the wait. The honest claim is not "everyone should do this." It is "this is the bet we are making, and here is why we think it holds."

Why this is the third move, not the first

Essay #4 argued that there is the one number you can’t buy: completion, the reader-side signal that capital cannot manufacture from the outside. Essay #5 argued that the standard org chart cannot reach that number, because it splits editorial from distribution and lets the channel win. This essay is the third move: it names what the unsplit structure actually produces. When the people who decide what is good also decide how it travels, they stop writing to the algorithm, because nothing in their structure is pulling them toward the channel's count. What they make instead is work built to be cited — claims that can leave the page intact.

That is the through-line of the second arc. The right number resists money (#4). The standard structure cannot reach it (#5). The structure that can reach it produces work written to the thesis, not the algorithm (#6). Each move is a claim you can lift out on its own. That is not an accident of style; it is the argument, performed.

What it means in practice

For us, writing to the thesis is a set of capabilities, stated plainly so they can be checked.

A Pagecut essay carries a claim its author will defend in public. The claim comes with its evidence attached, so the quote does not degrade into an assertion when it travels. The author is named, because an attributable claim needs someone to attribute it to. And the whole thing is extractable — its claims, evidence, and authorship readable without the page's visual layer, which is why this essay, like the others, ships in a form a downstream system can lift cleanly.

None of that promises an outcome. We are not claiming this wins rankings, earns more citations, or beats anyone's traffic; those are the channel's numbers, and chasing them is the exact move we are declining. We are claiming something narrower and checkable: the work is built to be cited, on purpose, against the easier instinct to build it to be found.

That instinct is strong, and most teams shipping write-to-rank work are not foolish — many of them know the trade and make it deliberately, because the channel pays now and the citation might never come. That is a real argument and we are not going to pretend it away. We are making the other bet, with our own publication, before we ask anyone else to. The discipline is the same one we have applied since the first essay: pick the harder, reader-side target, commit to it before the work starts, and let the structure — not willpower — keep us pointed at it.

Write to the thesis, not the algorithm. It is the only target that is still yours after the channel changes its mind.