Verification: The Epistemology of News
Critical Media
Three things that spread fast. Each was believed by thousands. Your job, before reading on: how would you check each one — without trusting that it "looks right"?
A Fake AI Image of an "Explosion Near the Pentagon" Briefly Moved the US Stock Market
On 22 May 2023 an AI-generated image showing a large explosion near the Pentagon spread across X, amplified by verified accounts. For a few minutes the S&P 500 dipped as traders reacted. There was no explosion. A glance at any news wire — or the image's own melted, inconsistent details — would have killed it. Nobody looked; they reacted.
"It didn’t have to be real for ten seconds to move billions."
Source ↗AI Deepfakes of South Korean Schoolgirls Spread Through Telegram — National Police Investigation
In 2024, networks on Telegram circulated AI-generated explicit deepfakes targeting South Korean women and schoolgirls, built from ordinary social-media photos. The images were fake; the harm was real. “It looks like her” was enough to do damage — a reminder that synthetic media is convincing precisely because it borrows a real face.
"Fake source. Real victim."
Source ↗A True Statistic, Stripped of Its Context, Becomes a Lie
The most durable misinformation is often *technically true*: a real number from a real study, quoted with the one sentence that gave it meaning removed. “Crime up 40%” — over what period, what baseline, which crime, measured how? Trace the claim back to its origin and the scary headline frequently shrinks to something ordinary, or reverses.
"The dangerous lies are the ones with a real receipt attached."
For each claim: what is the single fastest move that would tell you whether to trust it? Where would you look FIRST?
Source Analysis Reading
Verify Before You Believe. Four Moves.
You cannot tell a real photo from an AI one by staring harder. Verification isn't about the thing in front of you — it's about everything around it. Mike Caulfield's method is four moves, and a skilled checker runs them in under a minute.
S — Stop. The instant you feel the urge to believe or share, stop. Strong emotion is the tell: misinformation is engineered to make you react before you think. The Pentagon image worked because people reacted in seconds. Your first move is to not move.
I — Investigate the source. Who is actually telling you this? Not what the post says — who posted it, and what are they? A blue tick is not a credential. Open a new tab and look the source up laterally (what do OTHERS say about them) rather than reading their own "About" page.
F — Find better coverage. Don't evaluate the claim where you found it — go find the best coverage of it. If a real explosion hit the Pentagon, every wire service on earth would have it. If only one anonymous account does, that absence IS the answer.
T — Trace claims, quotes and media to the original. Follow it home. Reverse-image-search the photo, find the original study behind the statistic, locate the full quote. Most misinformation dies the moment you reach its origin — because the origin doesn't exist, or says something different.
These moves work on an AI deepfake and a miscaptioned photo and a true-but-gutted statistic alike, because they all fail the same test: they cannot survive contact with their own source.
- 1
The Pentagon image was fake but moved markets. Which SIFT move — done in 10 seconds — would have stopped it?
Reveal answer
Find better coverage: a real Pentagon explosion would be on every wire service instantly. Its absence everywhere else exposes the single fake post.
- 2
What does it mean to “investigate the source LATERALLY”?
Reveal answer
Instead of reading the source’s own About page, open a new tab and see what OTHER, independent sources say about them — their reputation, not their self-description.
- 3
Why is a TRUE statistic sometimes more dangerous than a made-up one?
Reveal answer
Because it carries a real receipt — stripped of context (timeframe, baseline, definition) a true number can imply something false while surviving a lazy “is it real?” check. Only Tracing it to origin restores the meaning.
How a fact-checker works
SIFT — The Four Moves
You can’t verify a claim by staring at it. Verification is about everything AROUND the claim. Four moves, under a minute: Stop · Investigate · Find · Trace.
There’s often more than one valid move; mark the fastest decisive one. Emotional spike → Stop. Unknown author → Investigate (lateral). Alone-on-the-internet → Find better coverage. A number or quote or image → Trace to origin. The skill is reaching for a SOURCE move instead of judging the claim by how real it looks.
Source Criticism Brief
This is the source-criticism evidence in your portfolio — proof you can verify AI and media output instead of trusting it.
Run SIFT on a claim from your own feed.
Find one real post, headline, image, or video from your own feed in the last day — something you might have shared. Run the four moves and write the verification trail (200–300 words):
- The claim — paste/describe it, and your honest first reaction.
- Stop — what emotion was it engineered to trigger?
- Investigate the source — who posted it; what did lateral reading reveal?
- Find better coverage — who else reports it, or is it alone?
- Trace — where does the claim/image actually originate?
- Verdict — real / misleading / fake, and the ONE move that decided it.
You pass when your verdict rests on a SOURCE move (investigate/find/trace), not on whether the thing "looked real."
- ✓ [
- ✓ ]