Trust Badge

Fake reviews are a billion-dollar problem. Trust Badge is a review intelligence platform that combines AI-powered content moderation, real-time behavioral tracking, and multi-channel verification to surface authentic feedback and expose fraudulent activity before it damages anyone.

The Review Fraud Problem

Fake reviews distort purchase decisions, hurt legitimate businesses, and erode consumer trust. Existing moderation is either manual and slow, or keyword-based and easily gamed. A well-written fake review from a real-looking account sails through most filters.

Trust Badge approaches moderation differently: instead of only analysing review content, it analyses reviewer behavior — account age, review velocity, device fingerprint, IP history, and pattern correlation across multiple reviews from the same source.

Behavioral Fingerprinting

When a reviewer submits feedback, a behavioral snapshot is collected and stored alongside the review:

interface ReviewerFingerprint {
  ipAddress: string;
  userAgent: string;
  deviceId: string;          // canvas + WebGL hash
  sessionDuration: number;   // time spent before submitting
  keystrokeCadence: number;  // avg ms between keystrokes
  accountAgedays: number;
  previousReviewCount: number;
  velocityScore: number;     // reviews submitted in last 24h
}

Each factor is scored and weighted. A brand new account submitting a 5-star review with a 3-second session duration and no prior review history scores very differently from an established account with varied rating history and a 90-second session.

AI Content Moderation

The content layer uses GPT-4 with a structured output schema. Reviews are classified across multiple dimensions: sentiment authenticity, specificity (genuine reviews tend to mention specific product details), linguistic patterns associated with incentivised reviews, and whether the content matches the product category being reviewed.

// OpenAI structured output schema
interface ModerationResult {
  authenticityScore: number;   // 0-100
  flags: ReviewFlag[];
  reasoning: string;
  recommendation: 'approve' | 'review' | 'reject';
}

type ReviewFlag =
  | 'generic_praise'
  | 'suspiciously_perfect'
  | 'no_specifics'
  | 'promotional_language'
  | 'off_topic';

Real-Time Processing with Redis

Review submission triggers a processing pipeline. The behavioral score is computed synchronously (fast, in-memory). The AI moderation call is async — queued in Redis and processed by a worker. The review is visible immediately with a "pending verification" badge; the final verdict updates it within seconds.

Redis also powers the velocity check: a sliding window counter per device ID and IP tracks review submission frequency. Spike detection triggers automatic hold and escalation to human review.

Trust Signal Widgets

The output-facing side of Trust Badge is embeddable widgets — a review aggregate badge, a review feed, and a "verified review" indicator that signals to consumers that the feedback has passed moderation. Widgets are generated as minimal JavaScript snippets with no framework dependencies, designed to add under 5KB to any page.

Multi-Channel Verification

Optionally, businesses can enable purchase verification: a review is only accepted if the reviewer's email or order ID matches a record in the business's system. This connects via a webhook that fires when an order is fulfilled, storing a short-lived verification token the reviewer redeems during submission.

What I Learned

Behavioral signals alone produce too many false positives. A legitimate power user who reviews many products has a velocity score that looks suspicious without context. The breakthrough was weighting account history heavily — long-term, consistent reviewers get significantly more trust credit regardless of their velocity.

The AI moderation layer is expensive at scale. Moving the most common rejection patterns into a fast rule-based pre-filter (checking for known spam phrases, duplicate content, and generic templates) reduced OpenAI API calls by about 40% without sacrificing moderation quality.

Status

Trust Badge is live at trust-badge.app. The behavioral fingerprinting, AI moderation, and embeddable widgets are all in production.