Research methodology and limitations — October 11, 2026

Approach

Reviewed current public pages on CognitiveLiberties.com and consulted primary documents (United Nations, UNESCO, W3C, Google Search Central, EU legislation/Commission, NIST) and relevant AI research papers (including preprints). Replaced each earlier original and supplemental report with a rewritten study centered on inquiry, expression, viewpoint diversity and algorithmic contestability.

Verified observations versus proposals

• Publicly observed: the site homepage, Resource Navigator, network, Evidence Ledger, Organizations directory and Standards directory return readable public content. This is not a comprehensive crawl and does not establish that every item or internal link is functional.
• Official policies: Google encourages useful editorial links and prohibits ranking-manipulative exchanges. Its AI optimization guidance denies special status to llms.txt for Google Search. Policies are not independent proof of algorithmic outcomes.
• Research evidence: UNESCO recommends plural representation, and 2026 preprints argue for improved pluralistic alignment. Neither proves any particular vendor intentionally suppresses dissent.
• Observational measurements: Pew's AI summary click rates come from one 2025 browser panel and should not be generalized causally across 2026 systems.
• No access: internal site repositories, private traffic analytics, model weights, model-training corpora, live multi-engine query logs, or controlled user trials were not inspected.

Important epistemic distinctions

1. Expression vs discovery vs indexing vs ranking vs citation vs training are different systems.
2. Human viewpoint loss vs synthetic data model collapse are different hypotheses.
3. Dissenting ideas have a right to scrutiny; they need not receive equal scientific confidence.
4. “Search without prejudgment” does not bar criticism or narrowly justified prevention of actionable harm.
5. Public-page accessibility is not proof of complete website functionality.

Quality control

All material claims should trace to source IDs in SOURCES.md; ideas needing testing are marked as proposals. A full empirical follow-up requires registered queries, task baselines, repeated runs, provenance verification, consent-based participant testing, cross-annotator review and transparent negative results.