Language and framing
How an article presents claims, shapes emphasis, and guides the reader toward a particular interpretation.
Viesly / Trust Framework
Methodology
Last updated: April 2026
Viesly evaluates articles through a structured, AI-assisted analysis pipeline focused on trust, framing, and bias-related signals.
The goal is not to reduce journalism to a single verdict. The goal is to give readers a clearer, more transparent view of how an article is written, what signals it contains, and where additional scrutiny may be useful.
What this page is for
This page explains what Viesly analyzes, how the evaluation is produced, what the Trust Score represents, and where the system's limits begin. It is intended to make the product easier to understand and easier to use responsibly.
What Viesly analyzes
Viesly examines article presentation through multiple lenses associated with trustworthiness, editorial framing, and interpretive bias.
Source credibility signals
Whether sourcing appears attributable, varied, and evidence-oriented rather than thin, circular, or opaque.
Emotional or manipulative wording
Whether wording appears measured and informative or leans on loaded phrasing, outrage cues, or persuasive pressure.
Balance and perspective
Whether material context, counterpoints, and competing perspectives are meaningfully represented.
Factual confidence indicators
Whether claims are presented with appropriate certainty, attribution, and internal consistency.
Overall trust patterns
How these signals combine to suggest stronger or weaker trustworthiness in the article's presentation.
How the analysis works
The evaluation is designed as a structured pipeline so the output can be both systematic and explainable.
01
Article extraction
The article content is parsed from the page you are reading (Chrome extension) or from a link you submit. Raw article text is not stored.
02
Signal extraction
A single Gemini 2.5 Flash pass evaluates linguistic patterns, sourcing quality, and produces a reader brief.
03
Synthesis
Compressed signals are synthesized into a Trust Score, flags, and an explanation in your chosen output language (Gemini 2.5 Flash).
04
Cross-validation
Claude Haiku 4.5 independently re-scores the same signals. If the models disagree by more than 15 points, you see a disagreement notice.
05
User-facing explanation
Results are presented as AI-detected patterns to support reading judgment, not as statements of fact.
Trust Score
The Trust Score is an interpretive indicator. It is not a statement of objective truth, and it should not be read in isolation from the explanation around it.
What it means
The score reflects patterns associated with trustworthiness and bias-related signals in an article's presentation. It summarizes how the system interprets sourcing, wording, balance, clarity, and related indicators.
Context still matters. A nuanced article may contain strong language for legitimate reasons, and a polished article may still omit important context. The score should support reading judgment, not replace it.
What influences the score
- Factual consistency signals
- Sensationalism or loaded wording
- Evidence and sourcing quality
- Framing balance
- Clarity versus manipulation
- Confidence of the model output
Why multiple models are used
Viesly uses multiple advanced AI systems in its evaluation pipeline to improve robustness, cross-check reasoning patterns, and produce more useful synthesized explanations.
Anthropic Claude (Haiku 4.5)
Provides an independent cross-check score from the same compressed signals. A different provider than the primary Gemini steps, so disagreement is meaningful.
Google Gemini (2.5 Flash)
Runs the combined signal extraction, reader brief, and final synthesis passes over the article text supplied for analysis.
Why this matters
Multiple models do not eliminate error or guarantee truth. They can, however, help surface disagreement, reduce overreliance on a single reasoning path, and produce outputs that are easier for users to interpret critically.
Limitations and transparency
Viesly is intended to be transparent about where AI analysis helps and where human judgment remains essential.
AI analysis can be useful, but it is not infallible.
Article quality, source access, and available context can affect the result.
Subtle satire, specialist subject matter, or missing background context may reduce accuracy.
Viesly should support critical reading, not replace it.
The product does not determine absolute truth.
Privacy note
Viesly is designed to analyze content with care and to avoid unnecessary data retention in line with the platform's privacy practices. For details on how privacy is handled across the product, refer to the Privacy Policy.
Read with judgment
Viesly is best used as a decision-support tool for deeper media literacy. Review the score, read the explanation, consider the article in context, and use the output as one input into informed judgment rather than a final answer.