Automated match reporting has been in use for some time and works reliably for a narrow class of writing. Its limitations reveal what a report is actually for.
Structured facts translate directly into sentences
A match produces a clean record of goals, scorers, times, cards and substitutions. Converting that record into readable prose is a well-understood transformation.
Systems built for this purpose have covered lower divisions and youth competitions for years, generating reports for fixtures no publication could afford to staff.
The output is accurate, unremarkable and genuinely useful where the alternative is no coverage at all.
Significance is not present in the data
Whether a goal mattered depends on the league position of both clubs, the manager's situation, what happened in the previous fixture and what supporters expected. None of that is in an event feed.
A model can be supplied with league context and recent results, which helps with the mechanical part of significance and not with the human part.
The judgement that a substitution was an admission of a failed plan requires knowing what the plan was, which nobody recorded.
Causal claims are where errors appear
Reports naturally explain why something happened, and explanation requires linking events that the data merely lists in sequence.
A model generating fluent causal language will produce plausible explanations whether or not they are supported, because fluency and accuracy are separate properties.
This is the failure mode that makes automated reporting risky for prominent fixtures, where a wrong explanation is worse than a bare summary.
Tone carries editorial judgement
Describing a defeat requires decisions about how critical to be, and those decisions reflect a publication's relationship with its readers and the club.
Models trained on existing reports absorb an average of those decisions, which suits nobody in particular and reads as generic.
Outlets using automated writing therefore constrain tone tightly through templates rather than letting the model choose.
The useful division of labour is emerging
Automated systems now handle the factual scaffolding, statistical context and routine coverage that would otherwise go unwritten.
Human writers concentrate on the parts that require judgement, which is the analysis of why a match unfolded as it did.
That split works because the two tasks were always different, and the arrival of capable text generation mostly clarified which parts of a match report were ever writing.