Research & Scoring Methodology
How AI Doomsday Guide evaluates evidence, handles uncertainty, explains calculations, labels scenarios and chooses trustworthy sources.
Source hierarchy
For technical or contested claims, we prefer primary research, security disclosures, regulators and named institutions. High-quality reporting is used to establish events and context. Audience and market estimates are labelled by provider because third-party measurement services estimate rather than count the whole web.
The capability-access-consequence model
Every extreme AI claim is tested against three questions: what can the system demonstrably do, what can it actually reach, and what consequence follows if safeguards fail? A fourth question — recovery — asks whether access can be revoked, systems restored or harm reversed.
Tool methodology
Our interactive tools are browser-side educational worksheets. They make assumptions visible and do not call a hidden model to produce authoritative-looking answers. Job exposure is not a redundancy probability. The p(doom) explorer is multiplication of user-selected conditional assumptions, not a scientific estimate. Scam tools assess process red flags, not media authenticity.
How we decide what to investigate
Reader questions, recurring misconceptions and public interest all help shape what we cover. Search-volume estimates can show which questions are being asked most often, but they never determine the answer. When we quote third-party audience or market estimates, we record the source, date and geography and treat them as estimates rather than exact counts.