News Outlet Media Bias and Reliability Rating Prompt

This prompt is designed for verification pipelines that process news-derived claims and need to dynamically assess the outlet behind a source before using it as evidence. Use it when your system must decide whether to trust a news source, weight its evidence, or flag it for human review. The prompt produces a structured rating covering political bias, factual reliability, reporting track record, and opinion-versus-news classification. It is built for automated fact-checking workflows where source credibility must be evaluated at runtime, not hardcoded from static lists.

The ideal user is a verification engineer or pipeline operator who already has the outlet name and optionally a sample of its recent reporting. The prompt assumes you can provide at least a domain or outlet identifier; richer context such as a snippet of a current article or a list of recent headlines improves accuracy. The output is a structured JSON object with fields for bias rating, factual reporting level, and a confidence score, making it directly consumable by downstream evidence-weighting or routing logic. You should wire this prompt into a broader verification harness that includes validation of the output schema, retry logic for malformed responses, and a human-review queue for low-confidence or high-risk assessments.

Do not use this prompt as a replacement for human editorial judgment in high-stakes decisions such as legal filings, medical claims, or safety-critical content moderation. It is not a static media bias chart; it is a dynamic assessment tool that can drift with model updates and should be regularly evaluated against a golden dataset of known outlets. Avoid using it for non-news sources like personal blogs, corporate marketing pages, or government propaganda outlets without first adapting the evaluation criteria. For sources where the outlet name is ambiguous or the reporting sample is too short, expect lower confidence scores and route those cases to human review rather than forcing a classification.

This prompt template produces a structured assessment of a news outlet's bias, reliability, and factual reporting track record. It is designed for verification pipelines that process news-derived claims and need to weight sources dynamically. The prompt forces the model to separate factual reporting history from opinion content, detect false balance patterns, and flag outlets that present opinion as news. Before using this prompt, ensure you have the outlet's name, a representative sample of recent content, and any known metadata about ownership or funding. Do not use this prompt for real-time breaking news assessment where the outlet's track record on the specific event is still forming.

textCopy
You are a media bias and reliability analyst. Your task is to evaluate a news outlet and produce a structured rating with supporting evidence.

## INPUT
Outlet Name: [OUTLET_NAME]
Content Sample: [CONTENT_SAMPLE]
Additional Context: [OUTLET_CONTEXT]

## OUTPUT SCHEMA
Return valid JSON matching this structure:
{
  "outlet_name": "string",
  "assessment_date": "YYYY-MM-DD",
  "overall_reliability_score": 0-100,
  "reliability_confidence": "high|medium|low",
  "bias_rating": {
    "political_spectrum": "far_left|left|center_left|center|center_right|right|far_right",
    "spectrum_confidence": "high|medium|low",
    "bias_indicators": ["string"],
    "false_balance_detected": true|false,
    "false_balance_examples": ["string"]
  },
  "factual_reporting_track_record": {
    "factual_accuracy_score": 0-100,
    "correction_policy": "robust|adequate|minimal|none_detected",
    "major_retractions_count": 0-N,
    "fact_check_failure_rate": "low|moderate|high|unknown",
    "track_record_notes": "string"
  },
  "opinion_vs_news_classification": {
    "opinion_labeling_practice": "clearly_labeled|inconsistently_labeled|not_labeled",
    "opinion_presented_as_news": true|false,
    "misclassification_examples": ["string"]
  },
  "source_transparency": {
    "ownership_disclosure": "full|partial|none",
    "funding_disclosure": "full|partial|none",
    "methodology_transparency": "high|moderate|low",
    "anonymous_source_policy": "string"
  },
  "evidence_summary": "string",
  "limitations": ["string"]
}

## CONSTRAINTS
1. Base all ratings on the provided content sample and context. Do not rely on training data cutoff knowledge unless explicitly noted in limitations.
2. When evidence is insufficient for a rating, use "unknown" or set confidence to "low" and explain in limitations.
3. False balance occurs when an outlet presents opposing views as equally valid despite one side lacking factual support. Flag specific examples.
4. Distinguish between news reporting and opinion/analysis content. Flag instances where opinion is not clearly labeled.
5. Do not equate "center" political positioning with higher reliability. A centrist outlet can have poor factual accuracy.
6. If the content sample is too small or unrepresentative, set confidence to "low" and note this in limitations.

## RISK_LEVEL: [RISK_LEVEL]
If RISK_LEVEL is "high", require human review before finalizing any rating used in verification pipelines.

To adapt this prompt, replace the square-bracket placeholders with real values. The [CONTENT_SAMPLE] should include at least 5-10 recent articles spanning different topics and formats (news, analysis, opinion). The [OUTLET_CONTEXT] field should contain known metadata: ownership structure, funding model, geographic focus, and any prior reliability assessments. Set [RISK_LEVEL] to "high" when the outlet's rating will be used to accept or reject evidence in automated fact-checking pipelines. For lower-stakes use cases like research briefings, "medium" or "low" is acceptable. After receiving the model output, validate the JSON structure before ingestion. Run the output through your source credibility eval checks to detect calibration drift, over-reliance on political spectrum as a proxy for reliability, and false balance false positives. If the model returns opinion misclassification examples, manually verify at least a sample before accepting the rating.

This prompt is designed to be a single step in a larger verification pipeline, not a standalone chatbot interaction. The primary integration point is after claim extraction but before evidence matching: when a claim is attributed to a news outlet, this prompt assesses the outlet's bias and reliability so downstream evidence weighting can adjust confidence scores accordingly. The implementation harness must handle input normalization, output validation, retry logic, and audit logging because bias ratings are inherently subjective and will be scrutinized by users and reviewers.

The harness should enforce a strict JSON output contract. After receiving the model response, validate that the bias_rating field matches one of the allowed enum values (e.g., left, lean_left, center, lean_right, right), that reliability_score is an integer between 0 and 100, and that factual_reporting_track_record contains at least one specific incident or citation. If validation fails, implement a single retry with the validation error message injected into the prompt as additional context. Do not retry more than once for bias classification tasks—escalate to human review instead. Log every assessment with the model version, timestamp, outlet name, and raw output for audit trails. For high-stakes verification pipelines, route any assessment where reliability_score falls below 40 or where false_balance_flag is true to a human review queue before using the rating to weight evidence.

Model choice matters here. Use a model with strong reasoning capabilities (GPT-4o, Claude 3.5 Sonnet, or equivalent) because the task requires nuanced distinction between opinion journalism, analysis, and straight news reporting. Avoid smaller or older models that default to centrist ratings as a safe response. If your pipeline processes high volumes, consider caching ratings for frequently assessed outlets with a 30-day expiration to reduce API costs. Never use this prompt's output as the sole basis for suppressing or promoting content to end users without human oversight—bias assessment is an input to verification workflows, not a publishing decision engine.