Recently, someone asked me what video I would make to maximize the stats of an Instagram account. I answered: I have no idea. And if anyone tells you otherwise, they are lying.
That is not an evasion. It is a methodological position. This article unpacks it: why content recipes do not work, what we can actually know, and how to organize production so you can learn what no one can sell you. This is what I call content from the downstream end.
Why recipes are post mortems
Viral success is lightning: you cannot know where it will strike. The recipes people sell you, the miracle hooks and "10 rules of high-performing content", are yesterday's formulas. By the time a winning formula is legible enough to be documented, it is already saturated by everyone copying it, and depreciated by algorithms that detect repetition.
Best practices are lagging indicators. They describe what worked for someone else before everyone else started doing it. They are not predictions. They are post mortems: reports on an opportunity that has already closed.
The two layers of social media knowledge
We need to be precise, because "nobody knows anything" would be false. There are two layers of knowledge, and they do not have the same status.
The knowable layer. The near-universals: retention is decided in the first few seconds, each platform has native formats, editing rhythm affects the scroll-stop. And the niche codes: every community has its vocabulary, references, and expected formats. All of that is true, documented, and learnable. Someone who tells you that a direct hook beats a three-second intro is not lying.
But that knowledge is commoditized. Everyone has access to it, and AI now makes it available to anyone, for free, instantly. It explains why content fails; it never explains why content emerges. These are necessary conditions, not sufficient ones: ignoring the codes disqualifies you, respecting them does not qualify you. Lightning only strikes well-built kites, except every kite is well-built now.
The differentiating layer. The thing that makes a piece of content rise above the mass of competent content. This layer does not exist in the abstract. It exists only locally: for a given audience, at a given moment, inside a given relationship. What conditions the emergence of a piece of content is not its conformity to the codes. It is the relationship built with the audience receiving it. That relationship is not in a study, not in a persona, and not copyable from a competitor.
You cannot buy it. You learn it. And there is only one instrument for learning it.
Volume is the measuring instrument, not the cost
People keep telling you to produce less, but better. The opposite is true.
If knowledge of your audience can only come from observation, then every post is a survey of the audience you actually reach, not the one you imagine reaching. Not enough volume, not enough signal. Without signal, you arbitrate blindly: you are back to brief-driven intuition with a layer of data vocabulary on top.
To be clear, this does not mean "spam and see what sticks." Blind volume loses as surely as artisanal scarcity does. Against a global production flow made almost free by AI, you will always lose the quantity race. What works is structured volume: controlled variations on clear hypotheses, inside a creative territory that knows what it is.
You do not predict where lightning will strike. You build the lightning rod to catch it statistically.
The series as a test protocol
The most effective tool for structuring that volume is the serialized format. A series is a test protocol that does not call itself one: same mechanism, isolated variables, comparable iterations.
In practice, imagine a weekly series of eight episodes, with the same duration, same structure, and same visual world. In episodes 1 to 4, only one variable moves: the hook. Two episodes use an absurd question; the other two use a counterintuitive statement. Everything else stays constant, so the gap in three-second retention becomes attributable. In episodes 5 to 8, the winning hook is locked, and duration becomes the variable. Eight posts later, you no longer "feel" what works. You have two measured answers, valid for your audience and invalid for every other one.
That is what a one-shot can never do. A single piece of content that performs teaches you nothing, because too many variables moved at once to isolate the cause. A series turns each episode into comparable data. And as a secondary benefit, the audience attaches itself to the recurring mechanism, which is precisely the relationship we have been talking about from the start.
There is one condition: the numbers have to be traceable. A hypothesis set before publication, a metric identified in advance, and a number you can connect back to the content that produced it. If your reporting is a monthly export in a slide deck, you are not doing test and learn. You are doing decorative retrospection.
Why brands start with a handicap
Creators have this learning model in their blood. They have never known anything else. Brands do not, for one precise historical reason.
For decades, a brand did not need to build a relationship with its audience in order to be seen. Paid media dominated: TV ads, print placements, media inventory. You still had to fascinate people, of course; the great TV campaigns were fascination machines. But you fascinated from a guaranteed position, paid for in advance, in front of a captive audience. Fascination was a performance multiplier, not a condition of existence. A bad ad still aired.
That world no longer exists. In organic distribution, fascination has become the entry ticket. Without it, there is no distribution at all. The filter has moved from conversion to distribution. And the brand now has to defend itself on a second terrain it does not control: proof. Audiences no longer believe what you claim; they believe what they see demonstrated, through the content itself, through recurrence, through the coherence between what is said and what is shown. The advertising assertion, once the central tool of paid media, is exactly what the feed has learned to ignore.
The first reflex inherited from paid media is also the most destructive one: treating the feed as ad space. A post does not exist on its own. It exists inside a continuous flow of content consumed by millions of users, and nobody wants to see an ad in that flow. Flooding organic content with logos does not protect the brand. It signals to the audience that it is time to scroll.
Content from the downstream end, defined
Everything converges toward an inversion of the classic pipeline: idea, brief, production, publication, then stats, then a story we invent to explain them.
Content from the downstream end starts at the end. Before any creative idea, you answer the output questions: which platform, which native format, which duration, which consumption behavior. And above all: what have previous posts taught us? Past performance becomes the input for the next brief. The brief is no longer an intuition. It is a hypothesis built on observations.
This is not a recipe. Recipes freeze a moment, and moments pass. It is an epistemology: a way to learn from your real audience, post after post. That knowledge cannot be duplicated or transported to another brand or another audience. That is exactly what makes it an asset: what your competitors cannot copy is what you have learned.
The usual objection remains: "So you are producing cynical content optimized for the algorithm." No. The downstream end does not dictate substance. It dictates packaging and cadence. What you have to say remains entirely yours. Paradoxically, creative direction becomes stronger because of it: once the distribution mechanics of a format are validated, you can take real creative risks inside it. You no longer stake the whole production on a single bet. You take risks on terrain whose geography you already know.
Staying above the flow means producing to learn
Saturation is not going away. Production costs are collapsing, the flow gets bigger every month, and somewhere inside it sits the content you spent three weeks making.
The content that stays above the flow is not the best content. Nor is it the content that best applies the recipes of the moment, because the knowable layer no longer differentiates anyone. It is the content calibrated on a real audience by a team that took the time, and the volume, to know it.
So no, I do not know what content would work for you. Nobody does. But the method for discovering it exists, it is reproducible, and it starts with an organizational decision: produce to learn.
This is the problem that pushed me to build TJCW Content Factory: a production tool where every number is traceable back to the content that generated it, and where the next brief is built on the performance of the previous one. Because an epistemology without a measuring instrument remains an opinion.