← Back to App

Transparency · Artificial Intelligence

AI & Datatransparency

P. Bourdieu

Important Notice

The generated habitus profiles are AI interpretations — not scientifically validated diagnoses. Bourdieu would insist we say so.

Which AI Model

• Single profiles: Claude Haiku 4.5 (Anthropic)
• Pair comparison: Claude Sonnet 4.6 (Anthropic)
• No model training with user data (Anthropic API ToS)

How the analysis works

Song lists are sent to Anthropic with a sociological prompt. The model is instructed to apply Bourdieu's concepts to musical taste and return a structured JSON profile.

System Prompt (Summary)

"You are a cultural sociology analyst specializing in Bourdieu's practice theory (1982). Include Peterson & Kern (1996), Stuart Hall (1980), and Sinus Milieus (2023). Respond exclusively with a valid JSON object."

Advertisement

Known Limitations and Biases

• Training data bias: Primarily Anglophone — Austrian music culture may be underweighted
• Spotify demographics: Streaming users are not representative
• No biographical context: The AI only sees titles and artists, not life stories
• Structural constraint: The model must fill all fields — even when uncertain

Your Rights (GDPR Article 22)

HabitusMatch does not make automated decisions with legal consequences. The profiles are for entertainment and scientific reflection only.

Status: May 2026 · Updated when model changes occur

Research Survey — Data Collection

The optional research questionnaire captures your assessment of the AI-generated habitus profile. Only the following are stored:

• Your ratings of AI statements (scales 1–7)
• Sociodemographic self-assessments (milieu, capital, age)
• Cultural practice data (anonymous)
• Optional: email for interview contact — strictly separated from response data, only with explicit consent

The song list is stored only with explicit consent in anonymized form (no names, no play counts). All data is stored in Cloudflare KV in the EU.

Milieu Playlist Generation

The milieu playlists are generated based on aggregated, anonymized listening profiles from the survey. The process:

1. Survey participants from the same Sinus milieu are grouped (minimum size: n≥10)
2. Frequently mentioned artists and genres are extracted
3. Claude generates a playlist suggestion with reasoning
4. No identification of individuals is possible

The playlists are scientifically interpretive and illustrate sociological patterns — they are not algorithmic recommendations.