Confirmation Bias & Conspiracy Beliefs
A research reading guide to selective evidence evaluation and conspiracy beliefs, with clear limits on diagnosing people or judging the truth of a claim.
Read dossier →Misinformation, deliberate deception, misleading context and confident AI errors—with source limits and practical questions for evaluating claims.
How can false or misleading material appear credible, and what should a reader check before accepting or sharing a claim?
WHO distinguishes misinformation shared without an intention to mislead from disinformation knowingly used to deceive and cause harm. Its explanation emphasizes the sender’s knowledge and intent. A false claim alone does not establish that its author knowingly lied. [1]
Editorial distinction: clearly labeled fiction, personal belief, testimony, interpretation and speculation should not be presented as interchangeable with verified evidence. An unsupported claim remains unverified; lack of verification alone does not prove it false.
UNESCO identifies fabricated, impostor, manipulated and misleading content, false context, and false connections among relevant categories. These distinguish inventing material from misrepresenting authentic material. [2]
Illustrations of these categories—not documented incidents: an old photograph captioned as a new event; a fabricated quotation attributed to an expert; or a headline implying more than the linked report establishes. These examples explain mechanisms and do not accuse any particular creator.
NIST describes confabulation: generative AI can produce confidently stated errors, invented citations and misleading explanations. This creates risks when people accept or circulate the output because it sounds authoritative. The profile describes a risk, not a finding that every AI answer is false. [3]
Practical inference: an AI’s assertion about itself needs evidence just as another factual assertion does. A polished answer or a natural conversation does not establish factual accuracy. Style alone also does not establish who selected a topic, supplied instructions or edited the result.
Vosoughi, Roy and Aral studied approximately 126,000 verified true and false stories circulated on Twitter from 2006–2017. Classification drew on six fact-checking organizations. False stories spread farther and faster in the analyzed data; the authors also found greater novelty among false stories. [4]
This observational study concerns a particular platform, period and selected fact-checked stories. It does not establish that every falsehood outperforms every true report, or measure present-day TikTok or generative-AI channels. Its findings should not be treated as a universal law.
Editorial reading prompts: What exactly is being claimed? Can I reach the original source? Does it support this claim, population and date? Is a quotation complete? Is the image’s context established? Are apparently separate accounts repeating one original source? What evidence could change my conclusion?
For AI-assisted work, open cited sources and compare them with the claim. Where available, examine prompts, full exchanges, production disclosures and edits. Missing production records leave uncertainty; they are not proof of deception. These are practical recommendations, not a validated detection test.
Admitting fallibility does not guarantee accuracy. Our editorial standard is to acknowledge specific mistakes, correct the affected claim, and make substantive changes visible. Apply the same scrutiny to human and AI contributions.
Holy Wars explores convictions and their consequences through authored characters and events. This dossier offers real-world background for reading claims critically; it does not establish that fictional events happened or that a character represents everyone sharing a conviction.
Project standard: distinguish sourced findings, personal beliefs, testimony, interpretation, speculation and fictional invention. Research context does not validate game statistics or abilities. No new character event or gameplay effect is established by this entry.
Prepared November 11, 2026 by Rowan with William’s direction and publication approval. Version 1.0. The linked institutional pages, NIST confabulation discussion and study abstract were inspected for the specific claims above. This is a bounded source synthesis, not an exhaustive literature review or independent expert review. No individual social-media video was reviewed or used as evidence. Sources 1–3 are institutional guidance; source 4 reports an empirical study.
Explore each character or card, then follow its research tags back here.