Fact-Check Mode: What to Verify and How

The Limits of AI for Fact-Checking

Let’s start with what you need to know: AI cannot reliably fact-check. This isn’t a limitation we’ll overcome soon — it’s structural. AI models generate responses based on patterns in training data, not real-time verification against authoritative sources.

This creates a dangerous situation: AI can confidently state incorrect information. It can cite sources that don’t exist. It can present outdated data as current. When you ask AI “is this true?”, the answer you get might be wrong — and you’d have no way of knowing without checking yourself.

So why use AI for fact-checking at all? Because AI excels at a different task: generating verification checklists. AI can identify which claims in an article are verifiable, prioritize them by importance, and suggest where to find authoritative sources. AI does the organizing; you do the checking.

Building a Verification Checklist

The first prompt (PR022 — Source Interrogator) generates research questions. Its final output — “what questions should I research independently” — is your verification checklist. But not all claims deserve equal attention.

Prioritize central claims. If the article’s main argument depends on a specific statistic being accurate, that statistic goes to the top of your list. Peripheral details matter less.

Flag surprising claims. Information that seems too convenient, too dramatic, or too perfectly aligned with the author’s thesis deserves extra scrutiny.

Check attributed statements. When an article quotes someone or cites a study, verify both the existence of the source and the accuracy of the representation.

Verify numbers first. Statistics, percentages, dates, and quantities are the easiest claims to verify — and the most commonly wrong.

Where to Look: Matching Claims to Sources

Different claim types require different verification sources:

Government statistics: Go to official databases. For economic data, the source is usually a government statistics bureau. For health data, it’s the health ministry or WHO.

Scientific claims: Peer-reviewed papers are the standard. Google Scholar, PubMed, and university library databases help. Check if the cited study actually supports the claim.

Quotes and statements: Look for original transcripts, video recordings, or official press releases. Be wary of quotes that appear only in secondary sources.

Company and financial data: Public companies file mandatory disclosures (SEC filings in the US). Press releases come from company newsrooms.

The Two Prompts in Action

PR022 (Source Interrogator) operates at the article level — it evaluates the source, identifies potential biases, and generates research questions. Run this first.

PR023 (Evidence Evaluator) operates at the claim level — it examines specific evidence supporting specific claims. Use this when you’ve identified a central claim and want to understand how well it’s supported.

A typical workflow: Run PR022 on the entire article. Extract the verification checklist. For the highest-priority claims, run PR023 to assess evidence quality. Then verify externally using suggested sources.

This connects to other critical reading skills in this pillar. The Argument Map Prompt (C043) separates claims from evidence structure. The What’s Missing Prompt (C044) identifies gaps in coverage.