Press Release vs Independent Reporting Distinction Prompt

This prompt is for verification engineers and pipeline builders who need to prevent corporate press releases, government PR, and paid content from being treated as independent evidence. Automated fact-checking systems often ingest content from news aggregators, RSS feeds, and web scrapes where press releases are syndicated alongside original reporting. Without explicit classification, a verification pipeline can cite a company's own announcement as corroborating evidence for its claims. This prompt produces a structured independence classification with re-reporting chain analysis, original source identification, and adversarial disguise detection. Use it as a gate before evidence matching: if a source is classified as a press release, route it for human review or downgrade its evidence weight to zero for independent corroboration.

The ideal user is a pipeline operator or verification system architect who controls the ingestion and evidence-weighting stages of a fact-checking workflow. You should have access to the full text of the source article, its URL, and any available syndication metadata. The prompt works best when it can analyze the content's structure, attribution patterns, and disclosure language. It is designed to catch not only obvious press releases with company boilerplate and contact information but also more subtle cases where PR content has been lightly rewritten by a wire service or where a government agency's announcement is being passed off as independent journalism. The adversarial disguise detection capability specifically looks for cases where paid or promotional content mimics the style and format of editorial reporting.

Do not use this prompt for general credibility scoring, bias assessment, or domain authority evaluation. Those are separate tasks covered by sibling playbooks in the Source Authority and Provenance Assessment group. This prompt is specifically scoped to the binary question of independence: is this content an original, independently reported piece of journalism, or is it a form of controlled messaging from an interested party? If you need to assess whether a source is authoritative for a specific claim domain, use the Domain Authority Assessment Prompt. If you need to detect funding bias or conflicts of interest, use the Conflict of Interest and Funding Bias Detection Prompt. Using this prompt for broader credibility tasks will produce noisy results and miss the specific failure modes it is designed to catch.

This prompt template is designed to be copied directly into your system prompt or user message template. It instructs the model to act as a media provenance analyst, distinguishing between original independent journalism, corporate or government public relations material, and adversarial cases where PR is disguised as reporting. The output is a structured classification with a confidence score and a trace of the re-reporting chain, enabling verification pipelines to weight evidence appropriately.

textCopy
You are a media provenance analyst. Your task is to classify a source document as a press release, independent reporting, or a hybrid, and to trace its re-reporting chain.

## INPUT
[SOURCE_TEXT]
[SOURCE_METADATA]

## OUTPUT_SCHEMA
Return a single JSON object with the following fields:
- "classification": "press_release" | "independent_reporting" | "hybrid" | "unable_to_determine"
- "confidence": 0.0-1.0
- "indicators": ["list", "of", "specific", "textual", "or", "metadata", "indicators"]
- "re_reporting_chain": [
    {
      "outlet": "string",
      "type": "original_reporting" | "press_release" | "syndication" | "aggregation" | "unknown",
      "evidence": "string explaining the basis for this step"
    }
  ]
- "adversarial_indicators": ["list of signs PR is disguised as journalism, if any"]
- "rationale": "A concise summary of the reasoning behind the classification."

## CONSTRAINTS
- Base your analysis strictly on the provided [SOURCE_TEXT] and [SOURCE_METADATA]. Do not use external knowledge about the outlet unless it is present in the metadata.
- A press release is defined as content originating from an organization about itself or its activities, typically with promotional intent and lacking independent editorial scrutiny.
- Independent reporting is defined as content produced by a journalistic entity with editorial independence, featuring original fact-gathering, multiple sources, and a clear separation of fact from opinion.
- A hybrid source contains elements of both, such as a press release that has been lightly edited and republished by a news outlet without original reporting.
- Flag any adversarial indicators, such as the use of journalistic stylistic conventions (byline, dateline, quotes) to mask promotional content, or the omission of the originating organization's name.
- If you cannot determine the classification from the provided information, set the classification to "unable_to_determine" and explain what information is missing.

## RISK_LEVEL
[HIGH] - Incorrectly classifying a press release as independent reporting can lead to circular evidence and false corroboration in automated fact-checking pipelines. Always require human review for "hybrid" and "unable_to_determine" classifications before using the source as evidence.

To adapt this template, replace the [SOURCE_TEXT] and [SOURCE_METADATA] placeholders with the full body of the document and any available metadata (such as the URL, publication date, author, and outlet name). For high-throughput pipelines, consider adding a pre-filtering step to extract the byline and any disclosure statements (e.g., 'Sponsored Content', 'Press Release') before invoking this prompt, as these are strong signals. The [OUTPUT_SCHEMA] is designed for direct parsing; validate the JSON structure and enum values in your application layer before routing the result. For adversarial cases, log the adversarial_indicators field for periodic review to improve your detection patterns.

This prompt is designed as a classification and analysis step within a larger source authority assessment pipeline. It should be invoked after source ingestion and metadata extraction, but before evidence matching or claim verification. The model receives the full text of a source document plus available provenance metadata (URL, author, publisher, funding disclosures) and returns a structured independence classification with a re-reporting chain analysis. The output is not a final verdict on source credibility—it is one signal that feeds into downstream source weighting, evidence sufficiency scoring, and human review routing decisions.

Integration pattern: Call this prompt as a single-turn classification task with a strict JSON output contract. Use a model with strong instruction-following and structured output support (GPT-4o, Claude 3.5 Sonnet, or equivalent). Set temperature=0 to minimize classification drift. Wrap the call in a validation layer that checks: (1) the independence_classification field matches one of the allowed enum values (independent_reporting, press_release, sponsored_content, government_release, disguised_pr, syndicated_release, unable_to_determine), (2) the re_reporting_chain array contains at least one entry when classification is not independent_reporting, and (3) the confidence_score is a float between 0.0 and 1.0. If validation fails, retry once with the validation errors appended to the prompt as [CONSTRAINTS] feedback. If the second attempt also fails, route to a human review queue with the raw model output and validation failure reasons attached.

Adversarial case handling: The disguised_pr classification is the highest-risk output. When the model returns this label, do not auto-accept it. Instead, flag the source for mandatory human review and attach the model's indicators_of_disguised_pr array as the review context. This is the failure mode where PR content is deliberately formatted to mimic independent journalism—missing it can poison downstream verification. In production, consider running a second-pass confirmation prompt with different phrasing on all disguised_pr classifications before surfacing to reviewers. Logging: Capture the full prompt, model response, validation result, retry count, and final disposition for every call. This audit trail is essential for debugging classification drift, calibrating confidence thresholds, and defending source-weighting decisions to compliance stakeholders.

Model choice and latency budget: This prompt is not latency-sensitive in most pipeline architectures—it runs during source ingestion, not during user-facing query time. Prefer accuracy over speed. If processing high-volume news feeds, batch sources and run this prompt asynchronously. For real-time verification use cases, consider caching classifications by source domain and publisher to avoid redundant calls. Tool use and RAG: This prompt does not require external tool calls or retrieval in its base form—the model classifies based on the provided document text and metadata. However, for the re-reporting chain analysis, you may optionally provide retrieved context about known PR wire services, syndication networks, and publisher ownership structures to improve accuracy on edge cases. If you add retrieval, ground the model's chain analysis in the retrieved evidence and require it to cite specific retrieval results in the re_reporting_chain entries.

Next steps after classification: Route independent_reporting sources to standard evidence matching. Route press_release, government_release, and syndicated_release sources with a source-type flag that downstream verifiers can use to adjust evidence weight (these are not automatically discarded—government releases may be authoritative for policy claims, for example). Route sponsored_content and disguised_pr to a restricted evidence pool or exclude them from verification entirely, depending on your risk tolerance. Always log the classification decision alongside the source record so that verification reports can explain why a source was weighted or excluded.