AI Content Policy

Last reviewed: 18 April 2026

Women Live Score uses generative AI tools in parts of our editorial pipeline. We think being transparent about where and how is the only responsible default, so this page explains it in plain terms. If you have a question or concern about a specific article, write to womenlivescore@gmail.com.

Where we use AI

AI assistance is used in the following places on the site. In every case the underlying facts — scores, goals, cards, squad lists, standings — come from our licensed data provider, not from the model.

• Match recaps. After full-time, a structured summary of the match facts is passed to a large language model under a fixed template. The model writes the narrative paragraphs. The output is checked back against the event stream before publication.
• Match previews. The model generates short contextual sections (head-to-head framing, form summaries, key-player notes). Factual fields such as recent form, squad absences and head-to-head scores are pulled directly from data, not generated.
• Team and league descriptions. Reusable descriptions displayed on team and league landing pages are AI-drafted from a structured prompt and reviewed before deployment. These are static — they do not change from visit to visit.
• Probability narratives. Short natural-language explanations of the model probabilities (“Chelsea enter favourites at 62 %, Arsenal’s away form has…”) are AI-generated from the numerical output of the probability engine.

Where we do not use AI

Some things the model never touches. This is a hard line.

• Match facts. Scores, goal scorers, assists, cards, substitutions, timings, final results. These come from the data feed and are displayed exactly as received.
• Standings, player stats and league tables. Built directly from match data.
• Quotes and interviews. We do not publish AI-generated quotes. If we cite a player, coach or official, the quote is sourced from a named outlet, a club press release, or a verified social-media post.
• Corrections. Any change to a published article after the fact is written and approved by a human editor. The log on our corrections page is human-maintained.

How we fact-check AI output

Every AI-assisted article passes through a fact-check layer before publication. The checks, in order, are:

1. Schema validation — the article must contain the facts required by its template (final score, scorers, competition, date). Missing fields block publication.
2. Event-stream reconciliation — every factual claim that maps to a match event (a goal timing, a scorer name, a card) is matched against the event stream. Mismatches block publication.
3. Name-list check — player and club names in the article are matched against the official squad and league lists at the time of the match.
4. No-new-claims rule — the model is instructed not to introduce facts that are not in the prompt. Any claim that cannot be anchored to an input field is stripped before publication.

This pipeline catches the great majority of hallucinations. It is not infallible. If you see an error, report it — our turnaround target on confirmed errors is 24 hours.

Which models we use

We use large language models from two providers — Anthropic and Google — selected per task based on cost, latency and output quality. We do not fine-tune models on private user data. We do not send readers’ personal information to the models; the prompts contain structured match and league data only.

Model selection is reviewed every quarter. When we change models we test the pipeline against a held-out set of completed matches to confirm the replacement performs at least as well on our fact-check layer.

Disclosure

Pages that are AI-assisted display the label “AI-generated — reviewed by WLS Editorial” in the article footer. We believe readers are capable of weighing AI-assisted copy on its merits when they know what they are reading.

We do not use AI to simulate a byline, a photograph, an interview or a first-person perspective. WLS articles are bylined as “WLS Editorial Team” because they are produced collectively; there is no individual journalist whose identity a model could impersonate.

Feedback and appeals

If you believe an AI-assisted article contains an error, a misattribution, or language that mischaracterises a player, coach or club, write to womenlivescore@gmail.com. We review every request and respond within 48 hours on business days. Corrections we accept are logged on the corrections page with the original claim, the correction and the date.

This policy is reviewed at least quarterly.