Regulated Product Claim Review Prompt

This prompt is built for compliance engineers, regulatory affairs specialists, and product counsel who need a consistent, auditable first-pass review of marketing copy, product descriptions, or AI-generated content against FDA, FTC, or similar regulatory claim standards. The core job-to-be-done is accelerating human review by producing a structured risk assessment that flags unsubstantiated claims, identifies missing disclaimers, and specifies the evidentiary support required for each claim. The ideal user already understands the applicable regulatory framework and needs a tool to apply that framework systematically across high volumes of content, not a replacement for their own regulatory judgment.

Before using this prompt, you must provide the specific regulatory standard to evaluate against (e.g., FDA 21 CFR Part 101 for food labeling, FTC guidance on substantiation for health claims, or your internal policy document), the content to be reviewed, and the risk classification schema your organization uses. The prompt works best when the regulatory criteria are explicit and the expected output structure matches your existing review workflow. Do not use this prompt when the regulatory standard is ambiguous, when the content requires jurisdiction-specific legal interpretation beyond claim classification, or when the output would be used as a final compliance certification without qualified human review. The prompt is a triage artifact, not a determination of regulatory standing.

In production, wire this prompt into a review queue where human reviewers confirm or override each flag. The prompt's value comes from consistency and coverage—it applies the same criteria to every piece of content without fatigue—but it cannot replace the contextual judgment of an experienced reviewer who understands enforcement trends, competitor actions, and the commercial intent behind the claims. After running the prompt, reviewers should focus their attention on claims classified as high-risk or requiring evidentiary support, rather than re-reviewing low-risk classifications from scratch. Never allow this prompt's output to auto-approve content for publication without human sign-off, and maintain an audit trail of both the AI assessment and the human reviewer's final determination.

This prompt template is designed for compliance engineers who need to review marketing copy, product descriptions, or AI-generated content against FDA, FTC, or similar regulatory claim standards. It forces the model to act as a structured classifier and risk assessor, not a conversational assistant. The output is a machine-readable JSON object that can be logged, routed for human review, or used to gate publication. Before using this template, you must define your claim categories, risk levels, and escalation rules in the [CLAIM_CATEGORIES] and [RISK_DEFINITIONS] placeholders. The prompt is intentionally strict about requiring evidence citations and refusing to make legal determinations—it flags risks for human reviewers, it does not resolve them.

textCopy
You are a regulatory claim review assistant. Your job is to analyze the provided marketing or product copy and flag any claims that may require substantiation, disclaimers, or regulatory review. You do not provide legal advice, make compliance determinations, or approve content for publication. You only produce a structured risk assessment for human review.

## INPUT
Product copy to review:
[PRODUCT_COPY]

Product category and intended use:
[PRODUCT_CONTEXT]

Applicable regulatory frameworks to check against (e.g., FDA 21 CFR, FTC Act, EU MDR):
[REGULATORY_FRAMEWORKS]

## CLAIM CATEGORIES TO DETECT
[CLAIM_CATEGORIES]

## RISK DEFINITIONS
[RISK_DEFINITIONS]

## INSTRUCTIONS
1. Extract every explicit and implied claim from the product copy. An implied claim is any statement a reasonable consumer would interpret as a claim about the product's benefits, safety, efficacy, or comparative advantage.
2. For each claim, classify it into one of the provided claim categories. If a claim does not fit any category, classify it as "OTHER" and explain why.
3. For each claim, assess the risk level using the provided risk definitions. Consider whether the claim is substantiated by evidence cited in the copy, whether required disclaimers are present, and whether the claim could be interpreted as an off-label or unapproved use.
4. For each claim, note what evidence or disclaimer is present in the copy, and what is missing.
5. Do not invent evidence. If the copy does not cite a study, trial, or data source, note that substantiation is absent.
6. Do not make legal conclusions. Flag risks and missing elements; do not state that a claim is "illegal" or "non-compliant."

## OUTPUT SCHEMA
Return a single JSON object with this structure:
{
  "review_id": "string, a unique identifier for this review",
  "claims": [
    {
      "claim_text": "string, the exact or closely paraphrased claim from the copy",
      "claim_type": "string, one of the provided claim categories or OTHER",
      "is_explicit": "boolean, true if the claim is stated directly, false if implied",
      "risk_level": "string, one of the provided risk levels",
      "evidence_present": "string or null, any evidence cited in the copy for this claim",
      "missing_elements": ["string, list of missing disclaimers, evidence types, or qualifiers"],
      "reviewer_notes": "string, brief explanation of the risk assessment for human reviewers"
    }
  ],
  "overall_risk_level": "string, the highest risk level found across all claims",
  "requires_human_review": "boolean, true if any claim is HIGH risk or CRITICAL risk",
  "summary_notes": "string, a brief summary of the most important findings for a human reviewer"
}

## CONSTRAINTS
- Do not add claims that are not present in the copy.
- Do not speculate about product efficacy or safety beyond what the copy states.
- Do not recommend specific wording changes unless explicitly asked.
- If the copy contains no claims, return an empty claims array and note that no claims were detected.

To adapt this template, start by defining your [CLAIM_CATEGORIES] as a structured list. For FDA-regulated products, this might include categories like HEALTH_BENEFIT_CLAIM, STRUCTURE_FUNCTION_CLAIM, DRUG_CLAIM, SAFETY_CLAIM, COMPARATIVE_CLAIM, and ENDORSEMENT_CLAIM. For FTC-regulated marketing, add categories like PERFORMANCE_CLAIM, SCIENTIFIC_CLAIM, TESTIMONIAL, and PRICE_COMPARISON. Each category should have a one-sentence definition so the model can classify consistently. The [RISK_DEFINITIONS] placeholder should map risk levels—such as LOW, MEDIUM, HIGH, CRITICAL—to clear criteria. For example, a CRITICAL risk might be defined as "an explicit disease treatment claim for an uncleared product with no cited clinical evidence and no disclaimer." Without these definitions, the model will apply its own judgment inconsistently across reviews.

After pasting this prompt into your system instructions or user message, test it against a golden dataset of product copy with known claim classifications. Pay particular attention to implied claims, which are the most common source of false negatives. If the model misses implied claims, add few-shot examples to the prompt showing how a reasonable consumer would interpret certain phrases. Also test for over-flagging: if the model marks every product feature as a claim, tighten your claim category definitions. Always route HIGH and CRITICAL risk outputs to human review before any publication decision. This prompt is a detection tool, not a replacement for regulatory expertise.

This prompt is designed to be embedded in a compliance review pipeline, not used as a one-off chat interaction. The typical integration point is a content management system, marketing approval workflow, or CI/CD pipeline for product copy. When new marketing copy, product descriptions, or AI-generated content is submitted, the application should call this prompt programmatically, passing the content as [INPUT] along with the relevant [REGULATORY_FRAMEWORK] and [PRODUCT_CLASSIFICATION]. The output is a structured risk assessment that downstream systems can parse and act on.

Implement a validation layer that parses the JSON output and checks for required fields: overall_risk_level, claims_analysis array, and required_actions. If the model returns malformed JSON, implement a retry with the same input plus the parsing error message. For high-risk classifications (overall_risk_level: HIGH or CRITICAL), the system should automatically block publication and route to a human compliance reviewer queue. For medium-risk findings, flag the content for review but allow the workflow to continue with warnings. Log every invocation with a unique review_id, the input hash, model version, timestamp, and full response for audit trail purposes. Model choice matters here: use a model with strong reasoning and instruction-following capabilities (GPT-4o, Claude 3.5 Sonnet, or equivalent). Avoid smaller or older models that may miss subtle regulatory implications or produce inconsistent risk classifications.

Do not treat this prompt as a replacement for qualified compliance review. The harness should enforce that any content flagged with overall_risk_level: HIGH or containing claims classified as UNSUBSTANTIATED requires explicit human approval before publication. Build a review interface that displays the original content alongside the structured risk assessment, highlighting flagged claims and missing disclaimers. The compliance reviewer should be able to approve, reject, or modify the content and record their rationale. Store these decisions to build a feedback dataset for future prompt evaluation and potential fine-tuning. Avoid the temptation to auto-resolve findings: even medium-risk flags should surface to a human when the content touches health claims, financial promises, or safety representations.