AI Disclosure
VeriFix uses large language models and classical machine-learning components to gather, group, summarise, translate, and verify news. This page explains what AI does — and does not — do on the platform, so you can judge every story for yourself.
Where AI appears
- Verified summaries (30-second, 2-minute, 5-minute reads) are written by an AI model from the underlying source documents.
- Translations between languages are AI-generated.
- Story resolution — deciding which articles belong to the same Living Story — combines deterministic rules with AI similarity scoring.
- Verification signals — confidence score, source spread, contradictions, and freshness — are computed by our Verification Engine from source metadata and model analysis.
- Fact-check tool: claims you paste are analysed by AI against retrieved sources.
How verification works
Every Living Story surfaces the evidence behind it: the publishers reporting the story, their independence and trust tier, the timeline of events, and any contradictions detected across sources. The confidence score reflects the strength and diversity of that evidence — not an editorial verdict.
Human oversight
VeriFix is a small team. We monitor pipeline health, source trust, and model quality, and we can manually merge, split, or take down stories. However, most story updates are produced without a human in the loop on each individual item. Treat AI output as a research aid, not as ground truth.
Known limits
- Models can hallucinate, misattribute quotes, or over-simplify nuanced context.
- Translations may lose subtle meaning, especially in low-resource languages.
- Confidence scores reflect source signals, not journalistic certainty.
- Breaking stories change quickly — always check the timestamp and linked sources.
Providers
We work with several leading AI providers and choose the best model for each task based on quality, availability, and cost. The specific model used for a given output can change over time as quality evolves.
Data sent to models
We send only the content required to complete the task — for example, article text and metadata for summarisation, or the claim and retrieved sources for a fact-check. We do not send your account identifiers to model providers, and we do not use your personal reading history to train third-party models.
Reporting issues
If you spot an incorrect summary, misclassified source, or hallucinated claim, please email corrections@verifix.news with a link to the story. We take corrections seriously.
Last updated: July 2026