Dating Fraud Detection Signals and Systems

Types of Fake Profiles and Bots

Type 1: Bot Accounts

Automated accounts controlled by scripts rather than humans. Created to:

• Generate engagement (inflate user count, activity metrics)
• Spam links to external sites
• Harvest user data
• Create false matches for scam

Characteristics:

• Identical generic messages to many users
• Respond to keywords with scripted replies
• No variation in conversation
• Never initiate conversations naturally
• Exactly repeated patterns across accounts

Type 2: Romance Scammer Accounts

Real person(s) controlling multiple accounts to conduct romance scams. Covered in detail in separate article, but detection overlaps with fake profile detection.

Characteristics:

• Rapid emotional escalation
• Stolen photos from Instagram/modeling sites
• Reluctance to video chat
• Money requests within 2-4 weeks
• Inconsistent personal details

Type 3: Catfish Profiles

Real person using another person's photos. Motivations vary:

• Insecurity about appearance
• Testing platform (using celebrity photos)
• Attempting fraud/romance scam
• Revenge (posting ex's photos)

Characteristics:

• Photos don't match (old, obviously mismatched with age)
• Reverse image search finds same photos used elsewhere
• Inconsistent in video calls (claim camera broken)
• Personality inconsistent with photos

Type 4: Fake Verification Profiles

Accounts created to appear legitimate:

• Using real person's identity with fake details
• Professional photos with false credentials
• Fake credentials (doctor, CEO, athlete)

Type 5: Commercial Spam

Accounts promoting services, products, or websites:

• Cryptocurrency schemes
• Unlicensed escort services
• Money-making opportunities
• Dropshipping promotions

Characteristics:

• Generic attractiveness photos
• Bio includes commercial links
• Messages include promotional content
• Targets many users with same message

Type 6: AI-Generated Face Profiles

Recent phenomenon using AI (GANs) to generate fake faces:

• Photos of people who don't exist
• Extremely hard to detect with human eye
• Used for catfishing and romance scams
• Detection requires specialized tools

Why Bots and Fakes Are a Problem

User Experience Impact

Users join dating platforms expecting to connect with real people. Fake profiles:

• Waste user time (matches that lead nowhere)
• Damage trust (feeling deceived)
• Lower match quality
• Create frustration and churn

Users will leave if 10-20% of their matches are fakes.

Safety Concerns

Fake profiles create risks:

• Romance scams (money theft)
• Catfishing (emotional harm)
• Photo theft (non-consensual use of images)
• Data harvesting (stealing user information)
• Predation (bad actors using fake profiles)

Business Impact

Fake profiles affect:

• User acquisition (churn when users encounter fakes)
• User engagement (users stop messaging)
• Advertiser confidence (don't want ads near fake content)
• Platform reputation (reviews suffer)
• Regulatory issues (Ofcom notices if safety issues exist)

Detection Signals and Red Flags

Profile-Level Red Flags

Photo analysis:

• All photos appear AI-generated (too perfect, unusual artifacts)
• Reverse image search finds photos elsewhere (Instagram, modeling sites)
• Heavy, obvious filters or heavy photoshop
• Professional photos mixed with casual (using stock photos)
• Photos appear to be same person at different ages/appearances

Profile information:

• Contradictory information (age 25 but job "retired CEO")
• Too good to be true (model-looking person, extremely wealthy)
• Missing information (bio is single generic line, no interests)
• Mismatched details (claims London local but email is from Russia)
• Duplicated information (same bio as other accounts)

Profile behavior:

• Created very recently but has full profile
• No activity history (no profile views, matches, etc.)
• Claims to have been on platform long but account is new
• No profile photo (or only one low-quality photo)
• Unusual username (random numbers, obvious fake)

Behavioral Red Flags

Messaging patterns:

• Identical messages sent to many users
• Copy-paste responses (user asks specific question, gets generic reply)
• No variation or personalization in messages
• Response time oddly consistent (exactly 5 minutes every time)
• Messages don't match claimed time zone (respond at odd hours)

Conversation characteristics:

• Won't answer specific questions about themselves
• Answers are generic ("I love traveling and dining out")
• Avoids video chat with endless excuses
• Tries to move conversation off-platform very quickly
• Conversations always lead to same topic (usually money or external link)

Engagement patterns:

• Messages many users in short timeframe (suggests automated)
• Messages only at certain times (suggests part-time operator)
• Always initiates conversations (real users sometimes respond)
• Messages everyone regardless of match criteria
• Matches with everyone (real users are selective)

Account-Level Signals

Creation patterns:

• Account created very recently (within 24-48 hours)
• Multiple accounts from same email domain (suggests bulk creation)
• Multiple accounts from same IP address
• Multiple accounts registered with similar details
• Email never confirmed (many fakes use fake emails)

Engagement patterns:

• No real interactions (messages but no matches, likes but no responses)
• Suspicious timing (starts messaging at exact same time daily)
• Targeting specific user types (new users, young users)
• High message volume but low response rate
• Activity bursts followed by disappearance

AI-Generated and Deepfake Detection

AI-Generated Face Detection

Modern AI (GANs) can generate convincing fake faces. These are emerging threat.

Visual indicators of AI-generated faces:

• Asymmetrical facial features (AI often makes subtle errors)
• Unusual teeth arrangement (common AI failure point)
• Odd hair consistency (especially fine details)
• Background artifacts (blurred in suspicious ways)
• Eye consistency issues (slight differences between eyes)
• Skin texture too perfect (no real skin imperfections)

But these aren't reliable - AI is improving rapidly.

Tools for detection:

• Media Forensics (detects GAN artifacts)
• Sensity (specialized in deepfake detection)
• Adobe Forensics (Adobe's research)
• Steg Alert (steganography detection)
• Custom ML models trained on fake face databases

Accuracy: Most tools are 80-95% accurate for current generation GANs. But accuracy decreases as generation quality improves.

Deepfake Video Detection

If you allow video profiles or video verification:

• Deepfake videos can show fake person
• Detection requires specialized tools
• Processing is time-intensive

Tools:

• Microsoft Video Authenticator
• Sensity deepfake detection
• Custom models

Implementation Strategy

For profile photos:

1. Use standard photo moderation
2. For suspicious profiles, run through AI detection tool
3. Escalate to human review
4. Remove if confirmed AI-generated or deepfake

For video verification:

1. Use liveness checks (prove it's real person, live)
2. Facial recognition (match to ID)
3. Deepfake detection if specialized tool available
4. Human review for high-risk cases

Behavioral Pattern Analysis

User Behavior Scoring

Create a fakeness score based on behavior patterns:

!Comparison of fake profile types showing bot accounts, catfish, and scammer profiles *Comparison of fake profile types showing bot accounts, catfish, and scammer profiles*

Accounts scoring 20+ should be flagged for review. 30+ should be removed.

Markov Chain Detection

Some bots follow mathematical patterns. Analyze user behavior as sequence of states:

• State 1: New account
• State 2: Complete profile
• State 3: Send messages
• State 4: Move to external chat
• State 5: Request money

Real users have random state transitions. Bots follow predictable patterns.

Tools can detect when user behavior follows mathematical pattern rather than natural randomness.

Graph Analysis

Map user behavior networks:

• Who messages whom
• How quickly relationships develop
• Are multiple accounts behaving identically

Bots often create characteristic network patterns:

• Many accounts messaging same target users
• Rapid message volume spike then disappearance
• Coordination between accounts (same IP, similar details)

Automated Detection Systems

Tools and Platforms

Crisp Thinking Specializes in behavioral safety, can flag suspicious patterns.

Two Hat Security Focuses on identifying coordinated harmful behavior, good for bot rings.

Custom ML Models Build on your own data - best approach for dating-specific detection.

Pattern Matching Rules-based systems flagging known bot behaviors.

Implementation

1. Real-time screening

• Scan profiles on creation
• Flag obvious bots immediately (empty profile, stolen photo)
• Prevent bot profile from becoming fully visible

1. Message analysis

• Scan messages for generic/spam patterns
• Flag accounts sending identical messages to many users
• Prevent bot from reaching critical mass of targets

1. Network analysis

• Identify coordinated accounts
• Ban related accounts together
• Prevent account proliferation

1. Ongoing monitoring

• Periodically rescan accounts
• Update scoring based on new behaviors
• Identify emerging bot patterns

Human Review Process

Automated systems flag, humans decide.

Tier 1: Automated Flags

Clearly fake profiles are auto-removed:

• Empty profile (no bio, one photo)
• Stolen photos (reverse image search match)
• Email never confirmed after 7 days
• Messaging 50+ people with identical message in 24 hours
• Account replicates known bot pattern exactly

False positive rate: <1% (very confident removals)

Tier 2: Human Review

Profiles flagged as suspicious go to human reviewers:

• Photo analysis (is it real person or AI-generated?)
• Behavior assessment (is pattern consistent with bot?)
• Context understanding (could be real but weird behavior)

Reviewers decide: remove, restrict, or approve.

Tier 3: Appeal

Users can appeal removal. Provide clear reason for removal and allow re-submission with new photos or information.

Information to Preserve

When removing fake/bot account, preserve:

• Account creation details
• IP address and device info
• Photo data (for pattern matching)
• Message pattern details
• Any third-party confirmation (law enforcement, other platforms)

Use this to identify related accounts and improve detection.

Profile Verification Strategies

At Signup

Email verification

• Send verification link to email
• User must click to confirm email is real
• Prevents bots using fake emails

Phone verification

• Send code via SMS
• User enters code to confirm
• Higher friction than email but stronger signal of real person

Age verification

• Covered in separate article
• Prevents minors, also filters some bots

Initial photo requirement

• Require selfie on signup
• Run through AI detection for obvious fakes
• Prevent empty profiles

Ongoing Verification

Video verification (optional)

• Offer incentive for verified badge
• Run liveness check
• Facial recognition against ID
• Strongest proof of real person

Social verification

• Link social media account
• Verify account age and activity
• Signals real person with social footprint

Community verification

• Allow real users to mark profiles as "verified"
• Risky if abused, but adds signal

Prevention at Scale

Account Creation Rules

Friction at signup:

• Require working email
• Require phone verification
• Require at least 3 photos
• Require 100-character bio (bot-generated are often shorter/generic)
• Require age verification

Each friction point reduces bot proliferation.

Rate limiting:

• Limit account creation from single IP (1 per day max)
• Limit account creation from single email domain
• Limit profile completion within 1 hour of signup (prevents rapid profile farming)

Email validation:

• Flag free email domains at scale (bot use corporate-free email services)
• Require email confirmation before messaging
• Block emails used in other accounts

Messaging Controls

Anti-spam measures:

• Limit messages per user (max 20 per day)
• Prevent identical message to multiple users in short timeframe
• Require some customization in each message
• Flag accounts messaging more than matching ratio
• Prevent initial message containing external links

Bot Ring Detection

Sophisticated bots operate in coordinated rings. Detect:

• Multiple accounts created on same IP
• Multiple accounts using similar photos (rotated, filtered)
• Multiple accounts with same generic message
• Multiple accounts targeting same user
• Same payment method across accounts (for premium accounts)

Ban entire ring when detected.

Key Takeaways

1. Fake profiles and bots damage user experience and platform trust. Detection and removal is ongoing priority.

!Machine learning pipeline for fake profile detection showing training data, feature extraction, and classification *Machine learning pipeline for fake profile detection showing training data, feature extraction, and classification*

1. Three main types: bots (automated), catfish (wrong photos), and romance scammers (money requests).

1. Detection combines profile analysis (photo verification, reversed image search), behavioral patterns (messaging speed, conversation quality), and AI tools (AI-generated face detection).

1. Red flags include: stolen photos, generic messages, refusal to video chat, rapid escalation, money requests, and patterns of identical behavior across accounts.

1. AI-generated faces are emerging threat. Use specialized tools for detection, though accuracy decreases as generation quality improves.

1. Behavioral scoring flags suspicious accounts. Combine multiple signals for confidence.

1. Automated tools handle obvious fakes (empty profiles, stolen photos). Humans review borderline cases.

1. Verify at signup with email, phone, age check, and initial photo. Optional video verification provides strongest proof.

1. Rate limiting and anti-spam measures reduce bot proliferation at scale.

1. Preserve evidence when removing accounts to identify patterns and related accounts.

Cross-Links

• How to Prevent Romance Scams on Your Dating Platform
• Age Verification for Dating Sites: Requirements and Solutions
• Content Moderation for Dating Sites: Tools and Strategies