Compile a fact check of the post. Return JSON {"claim":"...","check_plan":[...],"trusted_sources":[...]}: claim - specific statement being verified; check_plan — verification steps in order; trusted_sources - Types of reliable sources for this topic. Return only JSON, no text around it.
Mechanics: AI and viral post are a draft and a rumor, not a source. Verification = find the original source OUTSIDE of them. A three-step plan: (1) highlight the statement being verified - a specific fact, figure, date; (2) find an independent primary source - the official website, the original document, two unrelated sites; (3) check the details - date, date, context, whether the phrase was taken out. Hierarchy of trust: primary source > authoritative publication > aggregator > random post. Technique: look not for confirmation, but for refutation - type “[statement] myth / refutation”, this way lies are caught faster than when you look only for “evidence” (otherwise the search will slip in what you already believe). The second trick: for AI links, check that the page actually exists and says exactly what it says, and not just that the link looks plausible. Insider: the date is the most common hidden deception: a true fact with an outdated or substituted date. A typical mistake: “check” a fact by asking the same AI again - this is an echo, not a check.
Unlock access to submit solutions for instant AI review.