05 — Making search and AI omissions observable, explainable and contestable

2026 research edition · Research topic 5/10 · CognitiveLiberties.com

Editorial position and scope

Proposed mission: Cognitive liberty means that people can think and inquire freely, seek and receive information, voluntarily share their views and evidence, encounter disagreement, and challenge opaque decisions that prevent lawful information from being found. Access does not equal endorsement: freedom to inspect a claim is compatible with labeling its evidence weak, correcting falsehoods, and applying proportionate lawful restrictions. Individual privacy remains an option and a safeguard, not the central objective or a demand that people stop sharing with knowledge systems.

Status of this document: Independent desk research as of October 11, 2026. It uses public sources and the existing public website. It is not an internal site audit, a measured search-ranking experiment, an inspection of proprietary AI training datasets, or legal advice. Numbered [S##] citations resolve to SOURCES.md.

Research question

How can users and publishers test whether lawful content is excluded, demoted or misrepresented by search and AI systems, without presuming that every low ranking is censorship?

Executive finding

Accountability requires diagnosing pipeline stages, repeatable visibility audits, publisher tools and context-specific appeals. European DSA rules create concrete transparency duties for covered services, but they do not promise all viewpoints equal exposure and do not apply identically worldwide.

Evidence from public sources

1. The EU DSA contains recommender-system transparency provisions, while other obligations differ according to platform category and size. [S13]
2. The European Commission describes a public DSA transparency database and research data-access channels to facilitate oversight of covered moderation and systemic risks. [S14]
3. NIST AI RMF provides voluntary guidance for accountable, transparent, valid, reliable AI and for evaluating harms. [S15]
4. CognitiveLiberties.com currently publishes an evidence ledger and directories, which can support a public, reproducible case register. [S21][S22]

Analysis: what these findings actually mean

1. A useful audit records discovery pipeline transitions: URL accessible → crawled → indexed → retrieved for test query → ranked → cited → accurately described → clicked. Loss of visibility at one stage has a different remedy from failure at another.

2. Compare pairs of ideas with similar source quality and wording, across query paraphrases, locations, devices and repeated dates. Pre-register the reference set and allow independent reviewers to challenge the benchmark’s own ideological framing.

3. When a platform provides no appeal, a third-party reporting system can still publish transparent observations with screenshots, request dates, technical diagnostics, confidence intervals and documented corrections. Do not infer motives from outcomes alone.

4. Public accountability should aim for notice, explanations, auditing access, due process where appropriate, and meaningful exit via independent links. Disclosure should not require revealing detailed anti-spam rules useful to attackers.

Application to CognitiveLiberties.com

The goal is to make a meaningful discovery path between a visitor's curiosity, multiple independently operated sources, original evidence, fair representation of dissent and an explanation of how a source was selected. This is not merely a traffic-acquisition campaign. The current site includes usable public navigation and source records, including its Resource Navigator, network and evidence ledger. It should not be described as having broken endpoints based on a past unverified report. [S18][S19][S20][S21]

1. Action 1: Develop a public “Why is this source missing?” troubleshooting page.
2. Action 2: Publish a downloadable visibility audit dataset with neutral topic coding and replicable scripts.
3. Action 3: Draft a corrections and challenge process for errors in the project’s own directory.

How to test whether the idea works

Reproducible test design

Unit: a documented question → source list → human interpretation task, not a single isolated search screenshot. Start with a preregistered set of ten to thirty representative tasks, including questions where expert consensus is strong and others involving substantial interpretive dispute. Keep question phrasing, language, device, time, geography, tools and source eligibility criteria recorded. Include at least two independently maintained competing sources where an actual debate exists, and assign evidence-strength labels through blinded review.

Baseline: test the current public site's navigation and at least one ordinary general-purpose search/AI route. Compare to a link-first trail and, only if built, an AI-assisted source mapper. Log whether a participant reaches an original source, recognizes disagreements accurately and can find a rebuttal without re-querying a centralized intermediary. Audit the alternative for false balance and omitted mainstream evidence as carefully as for omission of minority perspectives.

Interpretation: report descriptive rates, uncertainty and failure cases. Avoid causal claims from observational cross-platform comparisons. Record opt-outs and consent for human research; do not infer ideological categories from users' queries.

Measurements specific to this research question

• Appeal completion and error-correction rates, where observable.
• True omission against a documented qualified-reference set.
• Inter-reviewer agreement and reproducibility of reported visibility changes.

Counterarguments, risks and uncertainty

Ordinary relevance ranking, lawful moderation and technical crawling defects must be distinguished from discriminatory exclusion. Many model internals are not public.

The opposite failure is also possible: an indiscriminate commitment to showing more voices can reward spam, coordinated influence, fabricated evidence, or false balance. A sound approach distinguishes being accessible, being relevant, being accurate, being recommended, and being included in model training. None automatically implies the others. Users should have ways to examine unconventional claims and to scrutinize the evidence against them.

Next steps

In the first thirty days, publish the topic definition, inspect relevant public pages, collect the primary sources in a transparent manifest and create a small reproducible benchmark. In ninety days, build at least one functioning editorial source trail and run a qualitative usability test. At six months, evaluate publisher participation and source representation. At twelve months, publish the results—including failures, disputes, maintenance burden and user feedback—without converting contested findings into universal claims.

Sources cited in this report

• [S13] Digital Services Act – recommender systems and user transparency — European Union Regulation 2022/2065: https://eur-lex.europa.eu/eli/reg/2022/2065/oj. Relevance: Transparency and platform obligations depend on service classification; not a general equal-ranking law.
• [S14] Digital Services Act transparency database and data access — European Commission: https://digital-strategy.ec.europa.eu/en/policies/dsa-brings-transparency. Relevance: Description of decisions databases and researcher access.
• [S15] AI Risk Management Framework — NIST, 2023: https://www.nist.gov/itl/ai-risk-management-framework. Relevance: Voluntary risk framework: validity, transparency, harms, fairness, contestability.
• [S21] Cognitive Liberty Research Evidence Ledger — CognitiveLiberties.com, observed 2026-10-11: https://cognitiveliberties.com/research-evidence. Relevance: Accessible public evidence ledger; disproves older claim of broken page.
• [S22] Cognitive Liberty Organizations Directory — CognitiveLiberties.com, observed 2026-10-11: https://cognitiveliberties.com/cognitive-liberty-organizations. Relevance: Accessible organizations directory; disproves older claim of broken page.