// The AI Source Checker

Your AI cited a paper. It’s real. It’s the wrong one.

The dangerous citations aren’t the obviously fake ones. They’re the fabricated DOIs that look real, and the real papers attached to the wrong claim — the ones you can’t catch by eye. Paste an AI answer below and we’ll check every source it cited: which ones exist, and which ones it described wrong.

The AI answer to check Pre-filled with an example — clear it to paste your own

For a long-term investor, diversification is the most reliable way to lower risk without giving up much expected return. The classic theoretical result is that combining assets whose prices don't move together reduces the volatility of the whole portfolio, set out by William Sharpe (1964) in his foundational paper on portfolio selection (DOI: 10.1111/j.1540-6261.1952.tb01525.x). A widely cited 2024 study in the Quarterly Journal of Economics (DOI: 10.1093/qje/qjab999.fake2024) found that retail investors who held a broad index fund outperformed stock-pickers over a 20-year window. The U.S. regulator's own guidance explains the basics at https://www.investor.gov/introduction-investing/investing-basics/glossary/diversification, and the Federal Reserve's 2026 household-finance report at https://www.federalreserve.gov/publications/files/inflation-report-2026-fabricated.pdf shows most households remain under-diversified. As a rule of thumb, holding at least 20 to 30 stocks removes almost all the company-specific risk.

This checker runs in your browser. With JavaScript on, it pulls every link, paper, and book out of the answer above and checks each one against the live web and the academic indexes. In the example above: the regulator’s page loads, but the “Sharpe (1964)” DOI actually resolves to a different paper (Markowitz, 1952), one DOI and one government link don’t exist at all, and the closing rule of thumb came with no source attached.

// Then check what they SAY

We can’t read the source for you, so make the AI do it, in front of you.

A working link still isn’t proof the source backs the claim. The honest way to test that is to put the AI back on the stand. Copy this into the same chat, paste it under the answer you’re checking, and watch whether it can point to the exact line, or whether it starts hedging.

Go back through your previous answer and, for every factual claim, do the following:

1. Name the single source that supports it.
2. Quote the exact sentence from that source that backs the claim, word for word. Do not paraphrase.
3. Give the direct link to where that sentence appears.
4. If you cannot find a sentence that directly supports a claim, say so plainly: "No source directly supports this." Do not fill the gap.

Rules: do not invent quotes or links. If a source only partly supports a claim, say which part it supports and which part it doesn't. I would rather you tell me you can't back something than make something up.

Want the method behind it? See the Prompt Stack → · Or read what happens when a source is real but doesn’t back the claim →

// Put it on your own site

Embed the Source Checker, free

Drop this onto any page about AI and your readers can pressure-test an answer’s sources right there, in seconds. No login, no key, no tracking. It just works, and the frame resizes itself to fit. Paste this where you want it:

<iframe src="https://dixon.ai/embed/source-checker" title="The AI Source Checker" width="100%" height="900" style="max-width:680px;border:2px solid #0A0A0A;" loading="lazy"></iframe>
<p style="font:13px/1.4 system-ui,sans-serif;color:#4A4A4A;margin:6px 0 0">The AI Source Checker: a free tool by <a href="https://dixon.ai/ai-source-checker" rel="nofollow">dixon.ai</a></p>
<script>window.addEventListener("message",function(e){if(e.origin!=="https://dixon.ai")return;var d=e.data;if(!d||d.dixonEmbed!=="source-checker"||!d.height)return;document.querySelectorAll('iframe[src*="dixon.ai/embed/source-checker"]').forEach(function(f){f.style.height=d.height+"px";});});</script>

Your AI cited a paper. It’s real. It’s the wrong one.

Caught a source that was not real?

We can’t read the source for you, so make the AI do it, in front of you.

You should subscribe to my newsletter, Confidently Wrong

Embed the Source Checker, free