When machines can generate anything, the scarce human act is deciding what to believe.
- Generating convincing material became cheap. Checking it did not.
- The machine is not lying to you; it is being convincing at scale, and convincing is not the same as correct.
- Most countries mention media literacy in policy; few have a dedicated policy for teaching it.
- Five questions — claim, source, incentive, check, confidence — do most of the work.
- Judgment can be taught. How well AI-assisted teaching works is still an open question.
When machines can generate anything, the scarce human act is deciding what to believe. This page says why, and sets out the five questions that do most of the work.
1 · Why this skill, why now
For most of history, producing convincing material was expensive, so scarcity did some of our filtering for us. That subsidy just ended. Fluent text, plausible images, and confident answers now cost nothing to make — which means the filter has to live in you. Whatever your job title says, your actual job increasingly includes: editor.
2 · The trap is comfort, not lies
The classic threat model was deception. The modern one is smoother: systems that sound like authoritative experts lower our guard. Research on human–AI interaction raises exactly this — the anthropomorphism of generative AI, resembling an authoritative expert, may reduce skepticism toward the information itself, which is why the paper argues critical thinking matters more during AI use, not less. It is framed by its authors as a risk, not a measured effect. The machine is not lying to you; it is being convincing at scale, and convincing is not the same as correct.“An Assessment of Human–AI Interaction Capability in the Generative AI Era: The Influence of Critical Thinking”, Journal of Intelligence, 2025, pmc.ncbi.nlm.nih.gov, read at source 16 Sep 2026: “the anthropomorphism of GenAI—resembling an authoritative expert—may reduce human skepticism toward information”.
Treat fluency as a reason to check, not a reason to relax. An answer that sounds like an expert is exactly the kind that lowers your guard.
3 · We are not teaching it fast enough
UNESCO's 2025 global mapping of media and information literacy is blunt: while 171 of 194 member states (88%) reference these skills in national policy frameworks, only 17 countries — fewer than 9% — have adopted a stand-alone policy — and 29% of countries offer media and information literacy education limited to technical digital skills, neglecting critical thinking, analysis and ethical engagement. The gap between how fast synthetic content arrived and how slowly evaluation skills are taught is one of the quieter risks this site tracks.UNESCO, New UNESCO issue brief reveals global gaps in media and information literacy policies and education, 2025, read at source 16 Sep 2026: the brief maps “194 UNESCO Member States”; “171 countries reference MIL within national policy frameworks”; “only 17 countries have taken the next step to develop dedicated, standalone MIL policies”. The 88% and 9% are this page’s arithmetic on those counts.
4 · What critical thinking actually is (working definition)
Not cynicism, and not vibes. A repeatable loop: Claim — what exactly is being asserted? Source — who says so, and how would they know? Incentive — who benefits if I believe this? Check — does a second, independent channel agree? Confidence — how sure should I actually be, and am I comfortable saying "unclear"? Five questions, a few seconds each, run before you share, cite, or act. That loop is the whole discipline; everything else is practice.
Run the five questions before you share, cite or act. Claim, source, incentive, check, confidence: a few seconds each, and “unclear” is an acceptable answer.
5 · Where to practise
Three pages on this site apply the loop to three common situations, each built from mistakes this site made and corrected: how to check a viral screenshot, how to read a number before you repeat it, and how to check a citation an AI gave you.
For generated media and writing: spot fake images and video and how to spot AI writing. For how claims carry sources, how much of the web is bots shows a figure with its evidence and its own correction. And the correction notes left in place on this site’s own pages are the loop run on its own work.
6 · The optimistic close
Here is the part the doomers miss: judgment is learnable. Media-literacy teaching exists on that premise. The evidence on the newest approach — using AI tools themselves to teach source checking — is still thin. One 2026 study of an AI-based media-literacy programme for children reports “significant gains in students’ media literacy skills”, but this site could not open the paper and read that result only as quoted from its abstract; a 2025 paper argues that AI in classrooms “must therefore also be aligned with a critical media literacy approach”, which is a recommendation rather than a result. The machines that created this problem may help train the skill that solves it. That is worth testing, not assuming.Not read at source: “Transforming clicks into critical thinking: An AI-based media literacy program for children”, ScienceDirect, 2026, sciencedirect.com refuses automated access; result read in search results quoting the abstract. “Technology, education and critical media literacy: potential, challenges, and opportunities”, Frontiers in Human Dynamics, 2025, frontiersin.org, read at source 16 Sep 2026.
Part of the Record. Companion pages: spot fake images and video and how to spot AI writing. On what stays valuable when output is free, see the work AI cannot take.