An Algorithm Tagged an Organ-Donation Story as Football: The Real Gap Sits in Verification

A foggy October morning in Shenzhen. The training ground of Shenzhen FC still held dew on the grass. I sat on the familiar concrete step, notebook open, writing down the steady bounce of the ball during a recovery session. A few paces away, the veteran supporters' group was still beating the old drum, slow and even, a rhythm I have heard for almost fifty years in this trade. My phone buzzed. A young editor sent me a spreadsheet of sports content that the system had auto-tagged during the week. On row eleven, sitting between transfer stories and match previews, was an article labelled "football". Its headline was a guide telling residents of Mexico City how to register as organ and tissue donors.

I read it three times. Then I read the whole spreadsheet. Among the twenty-nine information points the system extracted from that article, there was not a single club, player, coach, competition, contract or tactical metric. Nothing belonging to football. The label, however, stayed right there, neat and confident, as if it knew what it was talking about. When new media knocks loudly at the door, I still hear the old drum from the stands.

What the original story was about

The source article is a public-health report from Mexico City, published during a campaign to register organ and tissue donors around Mexico's National Day of Organ and Tissue Donation and Transplantation. The campaign is led by Clara Brugada, head of the city government. The communications centrepiece was staged at Museo Yancuic, in the Iztapalapa district. The piece walks readers through the steps of registering as a voluntary donor and explains the legal and medical process: donation is altruistic and free, and the family of the donor takes part in the final decision at the moment of death.

The figures in the article belong to public health: more than 3,000 people waiting for a transplant, more than 50,000 people already registered as voluntary donors, kidneys accounting for roughly 60 percent of transplant demand, and seven out of ten registered donors being women. According to data published by the city government within that same campaign, the core message is framed behaviourally: turn solidarity into a decision made before an emergency happens. It is a serious, objective service piece. It contains exactly one error, and the error does not belong to the reporter.

The error belongs to the label. The Stage-1 classification system assigned the domain "football" to the article. Moving into deep Stage-2 analysis, the analyst refused to invent conclusions. He stated plainly that football analysis cannot be built from non-football material, and marked every category — tactics, finance, transfers, standings, governance, dressing room — as insufficient information. The risk was rated high, but its nature was defined precisely: data-integrity risk, not sporting risk.

Why the genre makes the error worse

The source piece is a service article, the kind that tells readers how to do something specific. That genre is standard on general news desks, where health and civic reporters produce it constantly. Football barely has such a genre. Supporters do not need a step-by-step guide to becoming supporters, and clubs do not publish instruction manuals on registering your love for a team.

When a health service piece lands in a football content pool, the damage does not stop at one bad row. It distorts trend tracking, because the system records that this topic is heating up in football circles. The next day, another model reads that board, assumes football readers care about organ donation, and pushes more of the same. A wrong label does not stand still. It reproduces.

One geographical note matters here. Mexico City appears constantly in health coverage simply because it concentrates the country's hospitals and transplant procedures. That concentration has nothing to do with football, even though to a keyword filter, the sheer frequency of a major place name easily reads as a sporting signal.

What produces a wrong label

Looking at the recurring tokens in the source text, the path of the error becomes guessable. "CDMX", "campaña", "registrarse" are strings a keyword classifier grabs easily. In Spanish, "campaña" means both a health campaign and a season. A system running on string matching cannot tell the two apart, because it has never watched a match.

This is the kind of mistake that worries me more than sensationalism. A wrong article gets caught by readers. A wrong label sits quietly in a database, waiting to train another model, and then produces more wrong labels.

My trade began before labels existed

In 2026, the Independent had just been founded in London, and I began my writing career. I entered the profession not by reading spreadsheets, but by sitting in stands until I could hear how the rhythm of a match changes with the scoreline. Forty-nine years later, I still do one thing: I follow the team. Not the match, the team. I go to the ground not to score, but to keep the rhythm of the stories.

Based on my experience watching matches across nearly half a century, information about a team only has value when it has been verified by eye at the training ground, in the corridor, at the bus stop at eleven at night after an away trip. A camera does not record who lingered longer than necessary outside the medical room.

In March 2026, at 56, I followed Shenzhen FC in China League One. A new platform posted a training report twenty minutes after the session ended. It ran two hundred words. It skipped the club's financial crisis and the sale of two key players. That same week, my five-thousand-word series on Shenzhen's supporter culture, built on a survey of three hundred people, was dismissed by young editors as too long and not catchy enough for digital. I stood between two lines: speed on one side, depth on the other. The wrong label I received this October morning is a child of the first line.

A name is an entire person

In June 2026, thanks to the previous year's analytical series, a television station invited me to commentate on Portugal against Spain in Sochi. In the first half I mispronounced the name of midfielder Isco three times. Chinese social media mocked me on Weibo for days. The following week I sat through the full qualifying footage of all thirty-two teams and noted the local pronunciation of seven hundred and thirty-six players' names. Isco taught me that a name is an entire person, and that no syllable deserves to be read wrongly.

I bring that up because it belongs to the same family as the label. A mispronounced syllable disrespects a journey. A misassigned label does the same. When I write about a player, I always reserve at least one paragraph for his hometown, his voice, his family, his habits, the scars invisible on television. Strip those away and what remains is a meaningless name sitting in a data cell.

The letter in the empty dressing room

In September 2026, the stadium had been empty for seven months because of the pandemic. I met Dai Weijun, a twenty-one-year-old midfielder wearing number twenty-one, sitting alone in the dressing room. He told me quietly: "Uncle, with no spectators, I don't know who I'm playing for." I did not write that line down bare. I encouraged him to write an open letter to the supporters. The piece built around that letter was shared fifty thousand times. The empty dressing room that day still carried the smell of grass and someone's tears.

That moment taught me that the job of a beat keeper is not to record facts, but to help a human being speak in his own voice.

Three weeks of silence in Qatar

In December 2026, while the World Cup was running in Qatar, Dai Weijun's agent told me the player would go on loan to a Dutch club in the winter window. I held the information for three weeks, waited for the contract to be signed, then published the exclusive. It reached two million reads. A group of supporters immediately accused me of lacking transparency. I held an online question-and-answer session, listened to them out, and understood that the collective interest has to come before the writer's own.

Every transfer window is a parting, but the heart of a club never leaves.

What the machine counts and what it cannot

Back to the wrong label. That classifier and the models now flooding professional football share one weakness: they count brilliantly and understand poorly. A tracking system can tell me that a midfielder ran eleven point four kilometres in a match. Data on progressive passes, pressures per minute, ball recoveries in the opposition half — all of it is correct and useful in the analysis room. But none of it tells me that his father was sitting in stand A, or that he played that match with an unhealed wrist.

Data analysts are walking into the dressing room, and their conclusions often detach from the team's real rhythm. They arrive with the spreadsheet first and the person second. That is exactly how a machine labelled an organ-donation story as football: it processed the surface of language and then felt certain it had grasped the substance.

There is another wrong label I have watched for years, and it costs far more than one row in a spreadsheet. It is when a Gulf state attaches the label "football development project" to a tourism marketing campaign and recruits stars past their peak to serve as image ambassadors. The label reads as sporting. The internal rhythm is not football's rhythm. A football nation does not grow out of a two-year contract for a thirty-five-year-old.

The counter-view

The familiar story I hear at every conference is that artificial intelligence will replace sports reporters. The label incident reveals a different danger, and it runs the other way. The industry's problem is not a shortage of content generation but a shortage of content verification. We have far too much text and far too few people who read to the end.

Dig deeper, and football itself has been mislabelling people long before algorithms arrived. We call a player by his transfer fee. We call a goalkeeper by his goals-conceded tally. We call a young midfielder by a potential index. Those labels are convenient, shareable and often statistically accurate. But every time we use them to describe a whole human being, we re-enact the machine's exact error.

One thing the Stage-2 report did right deserves credit. Its author refused to invent tactical conclusions. He stamped "insufficient information" on every question he could not answer, rather than filling the space with plausible-sounding speculation. In today's transfer-rumour economy, that virtue is worth gold. One piece that dares to say "I don't know" is more trustworthy than a hundred that say "I am certain".

What I am watching for next season

I am not waiting for a smarter system. I am waiting for a checkpoint that knows how to stop. A few big clubs have begun building content-provenance procedures before publication, and in the coming years I believe sports newsrooms will have to do the same: a real person, who has actually watched a match, sitting at the end of the data pipeline to sign off before a label enters the system. That person's job will not be to write faster, but to know what is wrong.

I keep carrying my notebook of player pronunciations and my handwritten notes from training sessions. It does not scale, it cannot be automated, and nobody prices it on the market. But when a machine can misname a human being in a split second, the one who stays behind to get the name right is still the most useful person in the room.