How to fact-check a news article in under 3 minutes

Fact-checking is a skill, not a degree. Here's the four-step process working journalists actually use, simplified for normal readers.

Step 1: Identify the load-bearing claims (30 sec)

Most news articles have 3 to 5 claims their argument depends on. Highlight them. Examples:

- "Unemployment fell to 3.7% in March." (a number, a date) - "The bill would cost taxpayers $40 billion over ten years." (a projection, a source) - "Senator Smith said X." (an attribution)

Everything else is connective tissue. Don't waste time fact-checking that.

Step 2: Trace each claim to a primary source (90 sec)

For each load-bearing claim, ask: where did this number come from? A reputable article will link to its sources. If it doesn't, that's information about the article — it's making you take it on faith.

Quick rules: - Stats → Look for a government agency (BLS, CBO, Fed) or a peer-reviewed study, not a think tank press release. - Quotes → Find the original press conference, transcript, or video. Half of misattribution is just shortened quotes that change meaning. - Projections → Who modeled it? An advocacy group's "estimate" is opinion. CBO scoring is methodology. - Anonymous sources → Real reporting still uses anonymous sources, but the article should explain why anonymity was granted and what role the source had. "Sources close to" + no explanation = soft sourcing.

Step 3: Check what's missing (30 sec)

Search the same event from a different political side. If the opposing-side coverage includes a fact that your article omitted, that omission is editorial. Prism Cross-Platform Snapshot does this in one click — you'll see the same story framed across Google News, Hacker News, Bluesky, Mastodon, Lemmy.

Step 4: Verdict (15 sec)

Three possible outcomes: - VERIFIED — Every load-bearing claim has a primary source and the framing is fair. - CHERRY-PICKED — The numbers check out, but the article hides context that would weaken its argument. - MISLEADING — A claim is technically true but presented in a way that creates a false impression.

Prism Source Receipts automates this verdict for any URL you paste. You'll get a per-claim breakdown plus an overall trust rating.

A worked example

Article: "Crime is up 20% in Big City."

- Step 1: Load-bearing claim is "up 20%." - Step 2: Trace it. Up from when? Q4 2024 to Q4 2025? Up in which category? Property crime? Violent crime? Both? - Step 3: The opposing-side coverage probably notes that the absolute number is still below the 5-year average. Article-under-review omitted this. - Step 4: VERIFIED but CHERRY-PICKED. The 20% is real. The framing isn't fair.

That's the entire process. Practice it on the next three articles you read and it becomes instinct.

Related: What is media bias · How to spot loaded language · Steel-manning explained.