Claims travel. Corrections don't.

Something arrives — forwarded by a relative, said on a podcast, half-remembered from years ago. It sounds plausible. It might even have started with a real study. By the time it reaches you it has been retold a dozen times, and whatever qualification it once carried fell off somewhere along the way.

FolksLore exists to follow that trail back.

What it does

You paste in what you were told. It separates the distinct claims inside it — most forwarded messages bundle several, typically one accurate, one real source misrepresented, and one invented outright — then researches each separately and reports what the evidence says, with links to the primary material.

Then it does the part that matters more: it tries to find where the belief came from. The original paper, the broadcast, the product listing, the post. What it actually said, and how it changed on the way to you.

The rules it works under

• No verdict without a source. Any claim that comes back without a retrievable citation is marked unresolved, whatever the conclusion was. Nothing is asserted from memory.
• No scores or percentages. "73% likely true" is a number dressed as a measurement. Each claim gets a reason instead: fabricated, misrepresented, true but beside the point, contradicted by evidence, or genuinely unsettled.
• No jargon. The words misinformation, disinformation, conspiracy theory, debunked and fake news never appear in a result. They signal a side, and the argument is over before the evidence is read.
• No topic is off limits. Politics, religion, medicine, history, conflict. A checker that avoids uncomfortable subjects is worse than useless, because the pattern of what it dodges is itself a position.
• "Nobody knows" is a real answer — and so is a flat no. Where decades of study keep coming back negative, that is a finding, not an open question.
• You can push back. Every result can be challenged. The reply can concede, clarify, and admit a verdict was stated too confidently — but it cannot change a finding because someone insisted. Only new evidence does that.

What it's useful for

• The email your parent forwarded that you're fairly sure is wrong, but can't prove
• The claim you half-remember and can't source
• Something confidently asserted on a podcast, without a citation
• Working out whether a statistic is real, misused, or invented
• Finding the study underneath a headline about a study

The result is written to be copied and sent on. That's the point: the answer needs to travel back down the same chain the claim came up.

What it can't do

It runs on Claude, an AI model made by Anthropic, working from sources retrieved at the moment you ask. It can misread a source, miss a better one, or be confidently wrong. Every finding lists what it relied on so you can open those and judge for yourself — that part is the trustworthy bit, not the verdict.

It won't look into private individuals, won't give medical, legal or financial advice, and won't supply the method for something dangerous. Beyond that, ask it anything.

Who runs it

It's a small independent project, not a company, a newsroom, or anyone's campaign. It carries no advertising and takes no money from anyone with a stake in what's true. Nobody pays for a finding, a placement, or a favourable reading, and there is nothing to sell.

It doesn't use accounts, cookies, trackers or third-party scripts, and it never stores IP addresses. Questions are kept for 90 days, unlinked to anyone, so the checking can be reviewed and improved.

Every claim is examined the same way regardless of who it came from or which direction it points. If a result looks slanted, say so — the challenge box records it, and disputed findings are the main way this gets better.

Ask a new question