Can you take one claim that is racing across feeds right now and prove, on paper, whether it survives contact with its own source?

Source-Criticism Brief

Context

Your newsroom has received an article draft about the Korean deepfake epidemic and legislative response. The draft makes three testable claims: (1) South Korea passed a law in 2023 addressing deepfake pornography with named penalties; (2) the law carries a specific prison sentence (up to 5 years for possession); (3) the law was a response to a documented social problem (cite evidence it existed before legislation). You have one copy of the source article, your team roles, and a GPT window. If you publish something false, credibility dies; if you bury a true story, you miss the scoop. No hedging allowed. The question is not 'Is this article good?' but 'Can we publish this claim-by-claim without correction or retraction?'

Mission

You are delegating fact-checking to an AI assistant by writing specific, iterative prompts, breaking the work into per-role search tasks, tracing sources backward to their origins, and deciding as a team which claims pass the bar. The artefact is a claim-by-claim verification report proving your methodology: it must be clear enough for a peer auditor to replicate your work, and defensible enough for you to sign off on every confidence level.

Finish Line

A claim-by-claim verification report (one page) that lists each claim, its verdict (verified / partially verified / unverified / contradicted), confidence percentage, and the source(s) that settled it — auditable and replicable by a peer reviewer.

Deliverables

Source-Criticism Brief
lesson

A one-page dossier that takes one circulating claim apart on the table — claim, source chain, red flags, and a verdict you would defend out loud.

Team Roles

Investigator

Owns the sources log — finds real primary sources, records URLs and dates, chains reasoning from claim to evidence.
- Produce a sources log table (claim | URL | publication date | finding) with at least 3 sources per major claim; every URL must be functional and retrievable
- Document the search sequence in writing: which search did you try first, what did it return, what search did that lead to next? Show the chain of reasoning that moved from 'need to verify X' to 'found evidence at URL Y'
- Flag when a source cites another source and note whether you traced back to the original or stopped at the intermediary — explicitly state this in the sources log
Fact-Checker

Owns the claim inventory — extracts every factual assertion, runs them through AI prompts, records verdicts with evidence.
- Extract 8-12 distinct factual claims from the source article verbatim; categorise each as verifiable (can be proven true/false with evidence) or unverifiable (opinion, subjective, too vague); include the exact quote or paraphrase from the article
- For each verifiable claim, write 2-3 specific AI prompts that would yield independent confirmation (example: for 'The Korean deepfake law passed in 2023', prompt = 'When did South Korea pass legislation addressing deepfake pornography? Cite the law name and date'); do not use generic prompts
- Record the final verdict for each claim (verified | partially verified | unverified | contradicted) with the source(s) that resolved it; state the confidence level (90% = multiple independent sources; 70% = one solid source with corroboration; 50% = one source, no contradiction; below 50% = not publishable)
Source-Tracer

Owns the citation chain — traces claims backward through sources, verifies origins are real, maps how facts mutate across outlets.
- For each claim that cites a study, report, or named institution, build a citation-chain table (claim | source-A says | cites which source | original URL | what original actually says | divergence); trace backward at least one level; state whether the intermediary's summary matches the original
- Locate the primary source (the original institution, study publication, or press release, not a news article summarising it); confirm the date, author, and key figures are real; describe how to access the original
- For at least two major claims, produce a visual or written map showing how the fact travelled through 2-3 outlets, noting what was added, deleted, or reframed along the way (example: 'Original study found 15% adoption in dataset X; News outlet A cited 'growing adoption' without a number; Twitter account B posted 'AI adoption surges'; by the time it reached our source article, it became 'AI adoption is near universal')
Editor

Owns the final verdict — resolves team disagreements, writes the publishability summary, builds the one-page snapshot.
- Write a 250-350 word executive summary that answers: Can we publish this article? If yes/no/with edits, say it in the first sentence. For each major claim, give one-sentence verdict + confidence percentage + why you trust or don't trust the source(s) used
- When the investigator, fact-checker, and source-tracer disagree on a claim's status, document the disagreement in 2-3 sentences: what each role found, which evidence is stronger, and why you're calling it verified/unverified in the final report — do not hedge
- Create a one-page claim-status table (claim | verdict | confidence % | source | reasoning) scannable in 60 seconds; this is the artefact a journalism teacher audits, so every confidence level must be defensible if challenged

Exemplars

Anki — powerful, intelligent flashcards
AnkiWeb

Gold-standard personal mastery system: active recall + spaced repetition. Validity proven by repeated solo testing, not by looking things up — the capstone’s whole logic.