Client Offer / AI Review System

AI-Assisted Publisher Risk Scoring System

A step-by-step system to help review affiliate and publisher applications faster, detect risky submissions, and surface a clear score with an AI-generated review summary.

Goal

Turn thousands of applications into a clear, prioritized review queue.

The goal is to create an automated application monitoring layer that enriches each publisher submission, analyzes the submitted website, detects risk signals, and pushes a clean review record into Notion with a score, summary, and recommended next step.

1

Reduce manual work

Automatically collect signals that reviewers normally check by hand.

2

Spot risky applicants

Flag suspicious IPs, proxies, thin websites, missing compliance pages, and inconsistent data.

3

Standardize decisions

Give every application the same structured review process and scoring logic.

Step-by-step flow

How the system will work

01

Publisher submits application

The form captures business details, website URL, email, phone, IP address, browser data, and submission timestamp.

Input
02

Fraud and identity signals are checked

The system checks for VPN/proxy/data center usage, suspicious IP reputation, invalid email, VoIP phone, location mismatch, and repeat-submission patterns.

Risk signals
03

AI crawler reviews the website

The crawler visits the website, reads key pages, extracts content, identifies the niche, checks basic legitimacy, and captures important website signals.

Crawler
04

Compliance and quality checks are applied

The system checks for missing privacy policy, weak contact information, thin content, risky verticals, misleading claims, and absent affiliate disclosures.

Review
05

AI generates score and explanation

Each applicant receives a normalized risk score, clear red flags, positive signals, and a recommended action for the review team.

AI score
06

Notion record is created or updated

The final result is sent into Notion as a clean review card with status, score, summary, screenshots, and key decision fields.

Dashboard
What we will check

Application intelligence layers

Each application will be reviewed across multiple signal groups, so the team can see not only whether the applicant is risky, but why the applicant was flagged.

Applicant & submission risk

Submission-level checks focused on whether the applicant looks real, consistent, and safe to review.

  • VPN, proxy, Tor, hosting, or data center usage
  • IP reputation and suspicious geolocation mismatch
  • Browser/device consistency and repeated applicant patterns
  • Email quality, disposable email risk, and deliverability
  • Phone validity, VoIP risk, country, and carrier consistency

Website legitimacy

Website-level checks focused on whether the publisher has a credible, functioning online property.

  • Website availability and crawlability
  • Homepage, About, Contact, Privacy, Terms, and disclosure pages
  • Visible company/person identity and contact details
  • Affiliate disclosures and monetization transparency
  • Website category, niche, and business model classification

Domain & infrastructure trust

Technical trust checks that help separate established publishers from recently created or suspicious properties.

  • Domain age, registration date, expiry, and recent changes
  • WHOIS/DNS consistency and ownership signals when available
  • Hosting provider, ASN, server location, and infrastructure risk
  • SSL/certificate signals and suspicious subdomain patterns
  • Malware, phishing, scam, and reputation indicators

Content quality

Editorial and content checks focused on whether the website has real value or looks like a thin affiliate shell.

  • Content depth, number of pages/articles, and update recency
  • Author identity, editorial structure, and original point of view
  • Thin pages, generic AI-style copy, copied templates, or placeholder content
  • Outbound links, affiliate redirects, and monetization patterns
  • Mismatch between claimed traffic/source and visible website quality

Compliance risk

Vertical-specific review for sensitive categories where publishers can create legal, brand, or network risk.

  • Finance, credit, insurance, health, supplement, crypto, and lead-gen claims
  • Missing disclaimers, misleading promises, or unrealistic outcomes
  • Privacy/TCPA language for lead generation forms
  • Restricted or sensitive category detection
  • Affiliate disclosure and advertising compliance checks

Internal history & duplicates

Pattern checks across the client’s own applicant database to catch repeat submissions and organized fraud rings.

  • Same IP, device, phone, email, domain, or company reused before
  • Applicants connected to previously rejected or banned records
  • Similar website templates across unrelated submissions
  • Repeated tracking IDs, contact details, or redirect patterns
  • Reviewer notes and historical approval/rejection outcomes
Suggested scoring model

Clear score, not just raw data

The system will transform raw checks into a structured decision layer: a normalized 0–100 score, risk level, recommended action, red flags, positive signals, and a short AI-written explanation for the reviewer.

0–100

Normalized publisher score. Higher score means stronger approval confidence; lower score means higher risk or missing proof.

Reviewer output: score, risk level, confidence, recommended action, and concise explanation.

Approve76–100
Conditional Approve56–75
Manual Review31–55
Reject / High Risk0–30

Example score explanation

“The applicant appears to operate a new finance content website. Positive signals include visible editorial content and working contact information. Risk signals include recent domain registration, missing affiliate disclosure, limited author identity, and proxy usage at submission.”

IP & device riskVPN/proxy/Tor, hosting/data center usage, suspicious IP reputation, device anomalies, and browser consistency.
20
Domain & infrastructure trustDomain age, WHOIS/DNS stability, hosting quality, SSL/certificate signals, and reputation indicators.
20
Website content qualityReal content depth, niche clarity, originality, author identity, page quality, and absence of thin/placeholder content.
20
Contact & company legitimacyWorking contact details, company identity, social presence, business consistency, and applicant-data alignment.
15
Traffic, SEO & social proofTraffic signals, organic footprint, backlink/authority indicators, audience proof, and claimed-volume consistency.
15
Compliance riskAffiliate disclosures, privacy/TCPA language, vertical-specific claims, restricted categories, and misleading advertising risk.
10
Internal duplicate/ring riskRepeat IPs, devices, emails, phones, domains, tracking IDs, templates, or links to previously rejected applicants.
20
Final normalized scoreThe relative weights are combined and normalized into a 0–100 score for easy review and reporting.
100
Important: the score should not be a black box. Each applicant should include component-level scores so the team can see whether the risk came from submission fraud, weak website quality, missing compliance, domain history, or internal duplicate patterns.

Decision rules and hard-stop flags

Some signals should override the score and immediately push the application to manual review or rejection.

Malware, phishing, or unsafe website reputation detected
Website is parked, unavailable, empty, or impossible to verify
VPN/proxy + brand-new domain + disposable email combination
Restricted vertical with missing disclosure or misleading claims
Applicant connected to previously rejected or banned records
Traffic/source claims strongly conflict with visible website evidence

AI review output

For every application, the AI layer should produce a reviewer-friendly summary, not just technical data.

Publisher type
Website niche
Final score
Risk level
Recommended action
Top red flags
Positive signals
Questions for applicant
Compliance notes
Deliverables

What Scale Your Web LLC will provide

✓
Application enrichment workflowAutomated checks triggered after every new publisher application.
✓
AI website crawlerCrawls submitted websites and extracts structured review data.
✓
Risk scoring logicCustom scoring model based on fraud, website, content, and compliance signals.
✓
Notion review dashboardClean applicant records with scores, summaries, statuses, and review fields.
✓
AI-generated summariesShort explanations that help the team understand why an applicant was flagged.
✓
Manual review queuePrioritized queue so reviewers focus on uncertain or high-risk applications first.
Implementation plan

Proposed project phases

Phase Focus Outcome
Phase 1 Workflow mapping, data fields, scoring categories, Notion structure. Approved blueprint for the review system.
Phase 2 Form webhook, applicant record creation, enrichment checks, and basic risk flags. New submissions automatically appear in Notion with initial signals.
Phase 3 AI crawler, website analysis, content classification, compliance checks. Each website receives an AI-generated review summary.
Phase 4 Scoring logic, reviewer statuses, QA, edge cases, and final automation polish. Production-ready review workflow for ongoing publisher monitoring.
Pricing

Pricing: TBD

Final pricing will depend on selected checks, application volume, data providers, crawler depth, and dashboard requirements.

TBD