AI Transparency

1. AI Models We Use

Horizon Reports uses a multi-provider AI architecture for reliability and cost efficiency. No single AI provider has a monopoly on our content generation:

Models are selected dynamically based on availability and task type. If one provider experiences downtime, the system automatically falls back to the next provider in the chain.

2. How Articles Are Generated

Our content pipeline transforms raw news data into original articles through a multi-step process. At no point does the AI see or copy source article text directly.

Step 1: Aggregation

Our RSS engine monitors 20+ international news sources around the clock, including BBC, Reuters, Associated Press, Al Jazeera, The Guardian, and other verified outlets. Raw articles are ingested and stored for processing.

Step 2: Clustering & Analysis

AI groups related articles about the same event into clusters using semantic similarity. Each cluster is scored for newsworthiness, and key entities (people, organizations, locations) are extracted and cross-referenced.

Step 3: Fact Sheet Extraction

From each cluster, the AI extracts a structured fact sheet containing only confirmed facts, attributed quotes, and verified data points. This fact sheet — not the original articles — is what the generator receives. This architectural isolation prevents plagiarism at the source level.

Step 4: Original Article Generation

The AI generates a completely new article from the fact sheet. It never sees the original source text. Each article is written in one of several editorial styles (analytical, investigative, breaking news, feature) matched to the story type.

Step 5: Quality Gate

Every generated article must pass a multi-layer quality gate before publication. Articles that fail are either regenerated with a different approach or rejected entirely. See section 3 below for details.

Step 6: Publication

Approved articles are assigned to one of our 100 specialized AI reporters based on topic match, categorized, and published with full source attribution. Social media distribution follows automatically.

3. Fact-Checking & Quality Assurance

Every article passes through a comprehensive quality gate with the following checks:

Articles receive a composite quality score (0.00 to 1.00) based on weighted scores across all checks. Only articles meeting the pass threshold are published. Failed articles are regenerated up to 3 times with different editorial styles before being permanently rejected.

4. Source Attribution Policy

Every article published on Horizon Reports is built from facts sourced from verified news outlets. Our source attribution standards:

• Source outlets are identified by name in our fact extraction pipeline
• Direct quotes are only used when attributed to named individuals in source material
• No source article text is ever reproduced verbatim — our AI writes entirely new prose
• Source credibility is scored based on outlet tier (Tier 1: major wire services, Tier 2: major nationals, Tier 3: regional/trade)
• Articles from low-credibility sources receive a quality score penalty

5. Editorial Standards

Our AI is instructed to follow strict editorial guidelines:

• Present facts neutrally without editorial opinion or bias
• Attribute all claims to their sources — never state opinions as facts
• Use professional journalistic tone appropriate to the story type
• Avoid sensationalism, clickbait, and emotional manipulation
• Include multiple perspectives when covering controversial topics
• Never fabricate quotes, sources, or statistics
• Clearly distinguish between confirmed facts and developing reports

6. How to Report an AI Error

AI-generated content can contain errors despite our quality checks. If you find an inaccuracy, fabricated information, or any other issue in an article, please let us know:

We take every report seriously and will investigate promptly. If an error is confirmed, the article will be corrected or removed.

7. What Data We Collect vs. What the AI Uses

User data and AI content generation are completely separate systems. No personal user data is ever used in article generation. For full details, see our Privacy Policy.