How are your 300K AI Personas built? Written manually one by one?

Question Type

Product Q&A (TYPE-A)

User's Real Concerns

With 300K personas, it's impossible to write them manually one by one, right?
Are they batch-generated using some algorithm?
If they're batch-generated, can quality be guaranteed?

Underlying Skepticism

Distrust of data source and construction methodology

Core Answer

The answer is: These 300K AI Personas are built based on real research questions from our users.

The key is not the 300K prompts themselves, but that each persona corresponds to a real research question.

Detailed Explanation

Construction Method: Based on Real User Research Questions

When users submit research needs on atypica.AI:

Social Media Scanning Automatically Analyzes Users
- User asks: "I want to understand urban women aged 25-35 who care about health"
- Scout scans platforms like Xiaohongshu (RED) and Douyin (TikTok) through 10-15 rounds of deep observation
- Atypica understands consumers through three layers:
  1. Explicit Expression Layer: Directly records consumers' clearly expressed preferences and attitudes (e.g., "I like Product A", "Price is my most important consideration")
  2. Implicit Logic Layer: Uses language analysis techniques to identify consumers' underlying thinking patterns (e.g., risk aversion tendencies, herd mentality, etc.)
  3. Emotional Association Layer: Analyzes emotional tones when consumers describe different purchase experiences, identifying positive and negative emotional triggers
- Extracts real users' 7-dimensional characteristics
Automatically Generates Tier 1 Personas
- Based on social media scanning data
- Includes 7 dimensions: demographics, geographic, psychographic, behavioral, pain points, technology adoption, social relationships
- Consistency reaches 80 points (close to real human baseline of 81)
Continuously Accumulates to Form 300K+ Persona Library
- Each real research question → generates 3-8 AI Personas
- As user adoption grows, the persona library naturally expands
- More users = richer library

Three Construction Methods

Method 1: Social Media Scanning Auto-Generation (90%+)

Process:

User Research Need  ↓Scout Scanning (10-15 rounds)  ↓Extract 7-Dimensional Data  ↓Auto-Generate Public Persona  ↓Add to 300K+ Public Library

Data Sources:

Deep social media observation: Xiaohongshu (RED), Douyin (TikTok), etc.
Each persona corresponds to 3,000 words of observation records
15 tool calls, covering behavioral patterns, values, decision-making logic

Method 2: User-Imported Private Data (<10%)

Process:

Enterprises import their own CRM user data, interview transcripts
Generate custom private personas (visible only to that enterprise)
Not counted in the 300K public library

Key Data

Persona Quality Assurance

Public Persona Library:

Quantity: 300K+
Construction method: Social media scanning
Typical data volume: 15 calls, 3,000 words of observation records
Consistency reaches 80 points (close to real human baseline of 81)

Custom Personas:

Quantity: Created based on user needs
Construction method: User-imported data or targeted social media scanning
Data volume: Depends on imported data quality (CRM records, interview transcripts, etc.)
Consistency reaches 85 points (exceeds real human baseline of 81)

Why Not "Batch Generation"?

If randomly batch-generated:

❌ Inconsistent consistency, far below real human baseline of 81
❌ Shallow feedback, lacking details
❌ Vague answers when probed
❌ Feels like "AI hallucination"

Our Construction Method:

✅ Based on real social media data or user-imported data
✅ Consistency reaches 80-85 points (close to or exceeds real human baseline of 81)
✅ Specific feedback with details and logic
✅ Can be deeply probed

Real Case Study

Case: Sparkling Coffee New Product Testing

User Need: "I want to test market reaction to sparkling coffee"

Scout Workflow:

Search on Xiaohongshu (RED) for "sparkling coffee" and "sparkling drinks" related content
Observe 10-15 rounds, analyze characteristics of discussants
Extract 3 typical user types:
- Type 1: 25-30 year-old white-collar women, focused on aesthetics and social aspects
- Type 2: 28-35 year-old health-conscious individuals, concerned about ingredients and calories
- Type 3: 22-28 year-old Gen Z, seeking novelty
Automatically generate 3 Tier 1 personas, add to library

Results:

User directly uses these 3 personas for Discussion testing
No need to wait for real human recruitment (saves 2-4 weeks)
Consistency reaches 80 points (close to real human baseline of 81)

Core Value Proposition

Why Does a 300K+ Persona Library Matter?

It's not about the quantity itself, but the breadth of research questions covered:

Industry coverage: Technology, consumer goods, education, healthcare, finance, and 15+ industries
Demographic coverage: Ages 18-60, tier 1-3 cities, different income levels
Scenario coverage: Product testing, brand positioning, market trends, creative validation

The more unique your research question, the more you need a large library:

If researching "25-35 year-old urban white-collar workers": Thousands of relevant personas in the library
If researching "45-55 year-old rural e-commerce users": Hundreds of relevant personas in the library
Larger library = higher probability of finding suitable personas

Comparison: Us vs Other AI Persona Tools

Dimension	Character.AI / GPTs	atypica.AI
Construction Method	Users write prompts themselves	Scout auto-analyzes real users via social media
Data Source	User imagination	Real user behavioral data
Quality Verification	❌ No verification	✅ Consistency 80-85 points (close to or exceeds real human baseline of 81)
Quantity Scale	Users create a few themselves	300K+ public library
Use Cases	Entertainment, companionship	Business research, user insights

Bottom Line

"300K is not a number, it's the accumulation of 300K real research questions. Prompts don't matter—the real user data behind the prompts is what's critical."