Who Decides What AI Believes? (www.aivojournal.org)

AI assistants are no longer neutral conduits of web content but invisible editors: the newly proposed AIVO Standard introduces the Trust Ratio (Tᵣ) and ties it to a Prompt-Space Occupancy Score (PSOS™) to quantify how internal “trust layers” privilege some sources over others. These retrieval systems rank provenance by factors like licensing, editorial oversight, freshness and moderation lineage—ChatGPT favoring timestamped licensed feeds, Gemini weighting author reputation and cross-corroboration, Claude leaning on its constitutional corpus—creating algorithmic editorial boards whose credibility weightings can be reshuffled by retrains. AIVO’s audits (Q3 2025 across ChatGPT 5, Gemini 2.5 Pro, Claude Sonnet 4.5) found verified domains in 73% of high-visibility answers versus 48% for factually equivalent non-verified sources; that 25-point Trust Ratio gap preceded a 6.2% average PSOS decline in the next retrain, evidence of a self-reinforcing “Trust Loop Effect.” Technically, Tᵣ = Verified Source Appearances / Total Assistant-Visible Appearances, with sector baselines typically ≥ 0.65; AIVO prescribes an Audit → Monitor → Alert → Verify cycle to detect Tᵣ drift and link it to PSOS decay. The governance gap is striking: ISO/IEC 42001, EU AI Act article 10 and NIST AI RMF emphasize datasets and bias mitigation but don’t require trust-layer disclosure, leaving boards and CMOs exposed to epistemic and financial risk. AIVO reframes discoverability as a measurable asset linked to compliance and valuation, urging independent assurance and trust-weighted visibility metrics for marketing, finance and corporate governance.