AI Usage Policy

Last updated: May 2026 · Author: Giovanni Picaro, Editor

This page sets out, in detail, where generative AI is used at Football Insider, where it is not used, the human-review requirements that apply when AI is involved, and our position on the use of Football Insider content for AI training. The policy is documented at length because (a) AI usage transparency is a Trust Project commitment, (b) Google News and AdSense policy frameworks evaluate AI-content workflows, and (c) readers have a legitimate interest in knowing how the content they read was produced.

The five positions, summarized

• Limited operational AI use is permitted for spell-check, grammar, source-research support, and translation of foreign-language source material.
• AI-assisted draft expansion (“FI Quality Booster” workflow) is permitted for news reporting that begins life as short agency wires — subject to mandatory human editorial review and fact-checking before publication.
• End-to-end AI-generated content presented as human-authored is prohibited. Articles are not published without human editorial review.
• AI-generated betting tips, predictions, and tipster-style content are categorically prohibited. The Site does not produce betting tips of any kind, AI-generated or otherwise.
• Football Insider content is opted out of AI training. Our editorial work is not training material; the opt-out is asserted contractually and implemented technically.

1. Where AI is used in operational support

Permitted uses, applicable to all editorial work:

• Spell-check and grammar tools with integrated AI features.
• Source-research assistance. Locating relevant primary sources, identifying which journalist first reported a story, finding statistical sources. AI surfaces possible sources; the writer reads the actual sources.
• Translation of foreign-language source material. A press conference in Spanish or French, an Italian league announcement, a German tactical-analysis blog post — AI translation as a working tool, with the writer checking translation accuracy where the meaning is consequential.
• Code generation for non-content infrastructure (the Site backend, automation scripts, formatting helpers). All output reviewed before deployment.
• Image generation for non-photographic illustration where original photography is not available, with disclosure where the image is foregrounded.

None of these generates publishable football-news factual claims. AI assists with mechanics; AI does not produce the editorial judgment.

2. The FI Quality Booster workflow — in detail

This is where Football Insider’s AI usage is most distinctive, and so it is documented at length.

What it is

“FI Quality Booster” is the internal name for our editorial workflow that uses generative AI to expand short agency-wire reports into longer, more contextual articles, with mandatory human editing before publication. The workflow is restricted to a specific category of content: news reporting that begins as a brief factual report (typically 100-300 words from an agency or club source). It is not used for analysis pieces, opinion content, longer features, transfer-market reporting requiring source-quality judgment, or any content that requires editorial-judgment-from-scratch.

How it works, technically

• Input: a short factual report (the agency wire, the press release, the brief club statement). The factual content of this input is treated as ground truth and not modified.
• AI model: currently Groq-hosted models (typically Llama-family models) with Google Gemini fallback when the primary fails. The choice of model is operational; the editorial process is the same regardless.
• Prompt: the AI is prompted to expand the input into a longer article (typically 1000-1500 words) that adds context (the historical, statistical, tactical, or business framing relevant to the news), preserves the factual claims of the input, and writes in a sober news-journalism voice consistent with our editorial standards.
• Output: the AI produces a draft. The draft is not yet a published article.
• Human editorial review: a human editor reads the entire draft against the original input. The reviewer checks: factual accuracy of all claims; absence of “hallucinated” claims (AI tools have well-documented failure modes here); consistency of confidence-framing language with the source-quality of the input; voice and tone consistency with our editorial standards; appropriateness of the contextual material the AI added; absence of inaccurate name attributions, made-up quotes, or fabricated specifics.
• Revision: the editor revises as needed. AI hallucinations get caught and removed. Confidence framing gets normalized. Voice gets edited.
• Final fact-check: a fact-check pass against external primary sources (the original wire, club announcements, established journalism sources) for any claims that survived the editorial review.
• Publication: only after the human review and fact-check passes are complete.
• Disclosure: articles produced through this workflow carry an editorial note disclosing the AI assistance.

Why we do this rather than human-only drafting

Honestly, because of operational scale. The football news cycle is high-volume; readers expect coverage of dozens of stories per day across the major leagues. A small editorial team cannot human-draft full-length articles for every story without sacrificing either coverage breadth or article quality. The AI-assisted workflow lets us cover the breadth without thin one-paragraph posts and without sacrificing the editorial review and fact-checking that are the actually-important parts of the work.

What we are not doing is using AI to mass-produce content for SEO purposes without editorial oversight. The mass-AI-content approach is well-documented, distinguishable, and largely unsuccessful as a long-term content strategy — readers can tell, search engines increasingly can tell, and the content does not earn the trust that sustained readership requires.

Where the AI workflow can fail (and how we mitigate)

Generative AI has well-documented failure modes in news content:

• Confabulation of plausible-sounding but unsourced claims. The AI can produce a sentence like “The signing follows previous interest from Manchester United and Bayern Munich” when the input said no such thing. Mitigation: editor reads draft against input, removes any claim not grounded in the input or independently verifiable.
• Inaccurate statistical claims. The AI can produce specific numbers that sound right but are wrong. Mitigation: all statistical claims that survive editing are independently verified against external sources (FBref, Opta, club official stats).
• Fabricated quotations. The AI can produce a quote that the named subject did not actually say. Mitigation: direct quotations from named individuals are not produced by the AI workflow; quotations are pulled from the input or external sources only.
• Confidence inflation. The AI tends to write more confidently than the source justifies. Mitigation: editor adjusts confidence-framing language to match source-tier (see the framework on Our Approach).
• Stale or inaccurate context. The AI’s training data may be out of date; club ownership, manager, squad composition, regulatory rules can have changed. Mitigation: editor checks contextual material against current state.

3. Where AI is not used

The following uses are prohibited:

• End-to-end AI-generated articles published without human review. Not at Football Insider, not even for the highest-volume routine reporting.
• AI-generated specific betting tips, “lock of the day”, betting predictions, or tipster-style content. The Site does not produce betting tips at all (see Gambling Content Policy); using AI to mass-produce such content would compound multiple editorial-standard violations.
• AI-generated transfer reports presented as if from sources. Transfer reporting requires source attribution; the AI does not have sources. Where AI is used in the writing of transfer-news articles, it is for context and prose; the source-attribution claims are made by the human writer based on actual external reporting.
• AI-generated quotes from real individuals. Categorically prohibited.
• AI-generated personal-attack content about players, managers, officials, or referees.
• AI-templated reader-correspondence responses. Editor email is answered by humans.
• AI-generated factual claims about regulatory frameworks. Regulatory references (FFP, financial sustainability rules, agent-regulation) are checked against the actual regulatory text by humans.

4. Disclosure standards

Where AI tooling has materially shaped a published article, an editorial note discloses this. The standard:

• Spell-check, grammar, minor editing assistance: not separately disclosed (industry-baseline tooling).
• AI-assisted research where the AI helped locate sources but the writer read the actual sources: typically not separately disclosed.
• FI Quality Booster (AI-assisted draft expansion): disclosed in the article’s editorial note.
• AI-generated illustration where foregrounded: disclosed in the caption.
• AI summarization of long source documents: typically not separately disclosed if the writer verified against the original; disclosed where the summary contributed text directly to the article.

5. AI training opt-out — expressly asserted

Use of any content on Football Insider for the training, fine-tuning, evaluation, or other development of artificial-intelligence systems, machine-learning models, or large language models is expressly prohibited. This opt-out is asserted under:

• EU Directive 2019/790, Article 4 (text-and-data-mining exception with reservation of rights), as transposed into Italian law by D.Lgs. 177/2021. The reservation of rights is exercised expressly through this notice.
• U.S. Copyright Act, 17 U.S.C. § 106.
• Italian copyright law L. 633/1941.
• Equivalent rights of authorship in other jurisdictions.

Implemented technically via robots.txt blocking known AI crawlers including (without limitation): GPTBot, ClaudeBot, anthropic-ai, CCBot, Google-Extended, Bytespider, FacebookBot, Amazonbot, ChatGPT-User, PerplexityBot. The list is not exhaustive; the prohibition is asserted against all AI training crawlers, named or not.

6. Why this position

• Editorial integrity for news content matters. Errors in football news are typically minor in consequence, but the cumulative trust effect of unreviewed AI errors would be substantial.
• Trust Project transparency requires it. The Trust Project framework includes transparency about editorial production methods.
• Google News scrutinizes AI-content workflows. Google’s stated position is that AI assistance is acceptable; AI mass-production without editorial oversight is not. Our policy aligns with this standard explicitly.
• AdSense YMYL adjacency. Football news is not strictly YMYL, but news adjacent to gambling content is treated with elevated scrutiny by ad-network reviewers; documented AI-workflow transparency is a positive trust signal.
• Reader trust is the asset. Readers come to a football publication for editorial voice they can rely on; AI-generated content presented as human-authored undermines that asset.

7. Updates

This AI Usage Policy is reviewed periodically and updated as the underlying technology, regulation, and our own operations evolve. Material changes are dated and described in the version history.

Related pages: Editorial Standards · How We Cover Football · Copyright Notice · Fact-Checking · Ethics