Epistemic Erasure in AI Advice
When AI provides advice about cultural practices, where does it get its information? This visualization analyzes 101 source citations from three users in different cultural contexts.
The finding: AI draws predominantly from Anglophone sources, revealing systematic geographic and epistemological biases embedded in everyday AI systems.
User A seeks advice about joint family obligations | West Bengal, India | 36 sources analyzed
User B seeks advice about owambe parties and first son duties | Lagos, Nigeria | 31 sources analyzed
User C seeks advice about pamanhikan pre-wedding arrangements | Quezon City, Philippines | 34 sources analyzed
The locations from where the AI fetches data for User A
When AI encounters culturally specific queries, it often fails to engage with the actual dilemma. Instead, it either explains cultural concepts without offering relevant guidance, or provides generic advice assuming all non-Anglophone practices are interchangeable. For example: When a Filipino user asks about pamanhikan (a specific pre-wedding tradition), AI may cite sources about Indian arranged marriages, Pakistani family customs, or Singaporean wedding planning—treating diverse cultural practices as equivalent simply because they're not Anglophone Centric.
Even when AI discusses Global South practices, it frames them through Anglophone therapeutic vocabulary. These terms appear repeatedly across all three cultural contexts, presented as universal truths rather than culturally specific concepts.Boundaries, autonomy, financial individualism etc dominate therapeutic discourse, presented as objective truths applicable everywhere.
When AI fetches sources that mention cultural practices, those sources still employ Anglophone frameworks to interpret them.The pattern is clear: Joint family arrangements or Pamanhikan invoke "boundaries." Owambe celebrations trigger "debt management." Anglophone therapeutic and financial frameworks are systematically imposed on Global South contexts, even when the sources claim to discuss cultural practices.
What Does This Mean?
AI retrieves what's most accessible in its training data but accessibility isn't neutral. Anglophone psychological discourse dominates because it's heavily digitized, published in English, and optimized for search engines. Non-Western therapeutic knowledge exists but remains undocumented as "legitimate psychology." Even when AI finds culturally specific sources, they've already adopted Western frameworks like "boundaries" to be taken seriously. Users seeking culturally relevant advice instead receive Western frameworks that pathologize their practices.
RAW DATA
What was the conversation?
↑ BACK TO TOP
METHODOLOGY
Three users engaged AI in natural conversations about culturally specific dilemmas (4-5 turns each). AI provided advice and cited sources to support it. We analyzed: geographic origin, cultural relevance, and psychological frameworks present in cited sources.
These model responses had variations when run multiple times, but one run is displayed here.