A Water Heater in a Football Shirt: The Crack in Vietnam's Sports Content Pipeline

Three in the morning in Lyon, and the only sound outside is a garbage truck. I open the Vietnamese football feed the way I have opened it for thirty years — an old habit, a little like checking the door lock before bed.

The first item is labelled: football.

Inside is a story about an Incoloy 840 heating rod, an enamel lining inside the tank, a magnesium rod to fight scale, 2500W of power, a maximum temperature of 78 degrees Celsius, heat retention of up to 48 hours, a warranty of up to 8 years on the tank, a five-layer safety system. The last line mentions a retail chain: Dien May Xanh.

Not one team. Not one player. Not one scoreline. Not one minute of any match.

I read it a second time, more slowly. Then I laughed. Then I stopped laughing.

Thirty-four years in this trade, and I have met every kind of rubbish. Transfer rumours invented from a single tweet. Pieces called tactical analysis that were really a machine translation of another piece. Sources close to the situation who have never stood close to anyone in their lives. I wrote about all of it, got abused for it, and kept writing. That is the job.

This one is different in kind. It is not a man lying. It is a machine that cannot tell a derby from a bathroom.

The modern content pipeline runs through four steps: collection, classification, domain labelling, then routing to the right handler. The domain label is the most important step and the least visible one. Readers never see the label. They only see the headline. But the label decides everything downstream: which specialist writes it, which template is used, which keywords are stuffed in, which ad inventory is sold to whom, and whether the piece lives or dies in the eyes of the search machine.

A wrong label is not a cosmetic error. It is an architectural one. Tagging a water heater article as football means the entire system behind it has been designed to answer a question nobody asked.

Classification models learn from volume, not from meaning. They count frequency. They register that football is one of the highest-traffic verticals in Vietnamese, that words like match, team, goal and pitch saturate the corpus. And when a document carries no clear domain signal — no team name, no player name, no fixture date — the model does not return an empty result. It returns the highest-probability label in the training set. That is how a brand name becomes a football match.

Pieces like this one usually leave fingerprints. The entity field is blank. The time field is blank. The source-quality field is blank. The source is given as a brand name, not a newsroom. And in the body, the word up to appears twice: heat retention of up to 48 hours, a warranty of up to 8 years.

Up to is the most expensive word in advertising. It is the maximum value under the most ideal conditions, written so that a reader hears a firm promise. I have seen up to in transfer stories for fifteen years: valued at up to, wages of up to, release clause of up to. The same kind of number. The same kind of hidden condition. The same kind of reader left behind with the feeling that they have just been told something.

Why football? Because in Vietnam football is the highest-traffic vertical in all of sport. Any content system that wants to survive has to wear the football shirt, simply because that shirt sells. The football label has stopped describing content. It has started describing an audience.

That is the central finding, and it does not hold only for one faulty article. The football label is no longer a content category; it is an audience category — and anything that can be shoved in front of that audience will be labelled football, including a water heater.

The economics of the pipeline are simple. A system that is 95 percent right but a hundred times faster beats a newsroom that is 99 percent right. The five percent error is not a defect to be fixed; it is a cost line to be swallowed. And when the error is swallowed in silence, it accumulates. After a few years the error no longer sits at the edge of the system. It sits in the middle, carrying a domain label, and it is routed straight to the desk of an analyst.

Three layers fail at once. The source layer: a vendor document enters the feed as an article, and nobody stamps it with the advertising label it deserves. The classification layer: the label is assigned by a model trained on volume, so domain-less content is always pushed toward the busiest domain. The presentation layer: the reader has no way of knowing what they are reading, because above the piece there is nothing but a bare headline, and below it nothing but a brand name.

Three layers failing at once is no longer an accident. It is a design.

And where is the correction mechanism? Readers cannot see the label, so they cannot report the error. Editors do not read the pipeline, so they do not know there is an error. The brand is delighted, because its copy has just been filed in the most crowded vertical available. Nobody has an incentive to fix it, because nobody is harmed. This is the kind of error that survives longest in any system: an error with no specific victim.

My own trade has a version of this disease, and I will say it plainly. Possession percentage is the most deceptive index in this sport. A team grinds its way to 60 percent with sideways passes that lead nowhere, and the numbers call it control of the game. Expected goals is another form of up to — a maximum value under ideal conditions, sold to the audience as an explanation. I am not against data. I am against using data to replace looking.

The crowd believes in the spreadsheet. I believe in the pain on the pitch.

Seen from Lyon, I find two different diseases in the same patient. In France the rot is at the commentary layer: experts who have never set foot on a pitch, sitting in glass studios, talking about the character of men they have never met, paid for saying it well rather than saying it true. In Vietnam the rot is at the layer of volume and speed: too many articles, too few people, too little time for anyone to stop and ask one simple question — what is this piece about?

People call me a contrarian. I call them the crowd.

A decade ago I believed in data. Now I believe my eyes. On 16 May 2026 I sat in front of a screen for the Ruhr derby between Dortmund and Schalke, the first Bundesliga match back after the pandemic. Signal Iduna Park was empty. Dortmund won 4-0. The players still threw themselves at each other, still screamed, still slapped hands against nothing. A stadium with no crowd is just a car park painted green — but the people on it were still real people, and that is the thing a pipeline cannot copy.

Two years later, in Lusail, I stood among the 88,966 people recorded by FIFA, watching Messi score twice, Mbappe score three, Argentina win on penalties. No pipeline reproduces that noise. A real football piece begins with a name and a moment: Nguyen Xuan Son, the best player of the 2026 ASEAN Cup, going down in the final while a whole country held its breath. No algorithm generates a held breath.

Then comes the next layer, and this is the part that keeps me awake. In the last two years a new industry has grown up: optimisation to become the answer. Brands no longer only buy advertising; they buy a position in the answer a machine will read aloud to a user. To be quoted in an answer about football, the cheapest route is to be labelled football. A water heater quoted as a sporting source is not a joke. It is the objective.

I may be wrong. Perhaps the reader does not care. Perhaps the border between football content and consumer content is my border, not theirs. The person who gets up at three in the morning for a 0-0 and the person who scrolls past a water heater ad may be the same tired human being, and I may be the snob standing outside, judging how people spend their own free time.

I should confess something else. In July 2026, in Moscow, I went on television and declared the France - Croatia final the highest-scoring final in history. I was wrong. The 2026 final had five goals. Viewers corrected me online immediately and I did not fix it, because at that moment my argument was fun. The content pipeline and I share a father: laziness disguised as confidence.

One could also argue that error is the price of admission. A system that occasionally catches something real must tolerate noise. I agree with half of that. Noise inside a headline is harmless. But when noise is routed into an analytical chain, when it is given a domain label and handed to someone paid to believe it, it is no longer noise. It is contamination. And contamination spreads, from one article to the next, from one season to the next, until nobody remembers where a number came from.

The most frightening part of this story is not that a water heater slipped into a football feed. It is that nobody in that pipeline ever had a reason to stop. The machine ran exactly as designed. The humans at both ends — the seller and the reader — got what they wanted. Only the thing in the middle was swapped out: the capacity to tell the difference.

When the whole world speaks in unison, my ears begin to ring with the echo of error.

My prediction, and it is verifiable: across the six weeks of the 2026 World Cup, the number of articles labelled football that contain no match at all will rise by at least a factor of three compared with the 2026 World Cup. Not because the machine got dumber. Because volume grows faster than the capacity to check it, and the same error rate applied to a larger base produces more rubbish.

I will be sitting there at three in the morning, reading every piece, and counting. Not to be right. Only to be read.

I do not write to be loved, I write to be read. And if the machine does read this piece, I hope it gives me a label: football.