How to fact-check a viral money claim with Claude

Set Claude up to cite, not guess

1. 🔎 Turn on web search. Open a new chat at claude.ai, click the settings slider in the message box, and switch Web search on (Anthropic's walkthrough). Without it, Claude answers from memory and can produce a study title that sounds exactly right and does not exist. Web search is on every plan, free included (announcement).
2. 🧠 Add Research if you are on a paid plan. Research runs many searches and comes back with a compiled, cited answer instead of one pass. Click the + at the bottom left of the chat box and choose Research. It is available on Pro, Max, Team and Enterprise (how Research works). On free, web search alone still runs everything below.
3. 📄 Grab the transcript. Open the caption and the auto transcript on the post, or screenshot the video and let Claude read the frames. You need the claims as text, in the creator's exact wording, because the precise phrasing is what gets checked.

The full fact-check prompt

This is the one I actually run. Paste it into a fresh chat, then paste the transcript when Claude asks.

You are my fact-checker for viral money and economics claims. I am going to paste the transcript of a video making claims about wealth, income, mobility or the economy. Use web search for everything, give me a clickable link for every source, and never cite a paper you have not actually located.

Work in four labelled parts.

PART 1, THE CLAIMS. List every distinct claim as a numbered line in the video's own wording. Next to each, mark whether it is an empirical claim that research could settle, a definition or a piece of jargon, or a value judgement about how the world should work. Give me the totals for each type.

PART 2, THE SORT. Put every empirical claim into one of three buckets, and say which bucket and why. Bucket A, supported: real published research backs it, so name the paper, the authors, the year, and link it. Bucket B, contested or overstated: the underlying pattern exists but the video's version is stronger than the evidence, so explain the gap in one sentence. Bucket C, unsupported: nothing published backs it, or it is an opinion wearing the costume of a finding.

PART 3, THE OTHER SIDE. Find the strongest published work that disagrees with the video's central claim and link it. Economics has genuine live disagreements, so if two credible camps exist, show me both and tell me which currently has more support. If the evidence really is one-sided, say that plainly instead of manufacturing balance.

PART 4, THE HONEST VERSION. Rewrite the central claim as one sentence that stays true to the research, and one sentence naming what it does not establish. Then tell me in two lines whether any of this should change a decision I would actually make.

Ask me for the transcript now.

The 3 follow-ups that do the real work

The first prompt gives you the map. These three are where a five-minute check turns into actually understanding the thing. Run them in the same chat.

Take the single strongest claim in that video and search specifically for published work that reaches the opposite conclusion. Give me the title, the authors, the year, a link, and the one-line reason their result differs, whether that is a different dataset, a different definition, or a different method. If the disagreement comes down to how a category was defined, spell that definition difference out in plain language, because that is usually where these arguments actually live rather than in the headline finding.

Go back through every number in the transcript and check each one against a primary source, meaning a statistical agency, a central bank, a census, or the original paper, never a news write-up of one. For each number, tell me the figure the primary source actually reports, the year it refers to, the country it covers, and whether it is adjusted for inflation. Flag any number where the video's version and the primary source disagree, and give me the link so I can open the source myself.

For every claim you put in the supported bucket, describe the effect size in ordinary terms instead of statistical language: how big is the difference, for whom, and over what period. Then tell me honestly whether an effect that size should change anything about how I plan, given that I am [describe your situation in one line]. I would much rather hear "real, but small enough that it should not drive your decisions" than a hedge that leaves me where I started.

What the three buckets mean

1. ✅ Supported. A named paper, real authors, a dataset, a year. You can open it. These are the parts of the claim you can repeat without qualifying, and they are usually narrower than the video made them sound.
2. ⚖️ Contested or overstated. The pattern is real but the strength, the direction, or the timeframe is argued over by credible researchers. This bucket is the biggest one in money content, and it is where "the pattern has intensified" lines almost always land.
3. 💭 Unsupported. Values, framing, and vibes. "The system is rigged" and "anyone can make it" are both in here. Not worthless, and not research either, so they should not borrow authority from the sentence before them.

4 things money claims get wrong

The "the wealthy will not marry you" claim, checked

The video argued that assortative mating means high earners marry high earners, that the pattern has intensified for decades, that who you marry matters nearly as much as what you do, and that saying anyone can make it ignores the system. Four sentences, three different kinds of statement. Here is how they sorted.

✅ Bucket A, supported. Assortative mating is a real, heavily studied term. Greenwood, Guner, Kocharkov and Santos, "Marry your like: assortative mating and income inequality" (American Economic Review, 2014) estimate that if husbands and wives in 2005 had matched randomly rather than by education, the US Gini coefficient would have fallen from 0.43 to 0.34 (AER page, NBER version). So the link between who marries whom and household inequality is genuinely in the literature.

⚖️ Bucket B, contested. "The pattern has intensified over decades" is the part the video treated as settled, and it is the part economists argue about most. Gihleb and Lang published a paper called, literally, "Educational homogamy and assortative mating have not increased" (NBER working paper 22927), using the Current Population Surveys and the Census and ACS. They argue both the economists' and the sociologists' versions of the increase are sensitive to how education categories are chosen, and that they find no evidence the correlation between spouses' potential earnings changed dramatically. Two credible papers, opposite headlines. That is not Claude being unhelpful, that is what the field actually looks like.

📊 What is measured and not disputed. Pew Research found that in 29% of US opposite-sex marriages both spouses now earn about the same, up from 11% in 1972 (Pew, April 2023). And "upward mobility feels harder" turns out to be measurable rather than a mood: Chetty and co-authors found absolute mobility fell from around 90% of children born in 1940 out-earning their parents to around 50% for those born in the 1980s (The fading American dream).

💭 Bucket C, opinion. "When people say anyone can make it, they are ignoring the system" is a value judgement. It might be one you agree with, and it is still not a finding, so it should not inherit the authority of the Gini number two sentences earlier.

🌱 The honest one-liner: who people marry is linked to household income inequality in US data, whether that sorting has actually increased is genuinely disputed, and none of it establishes anything about what any one person can do with their own income.

Save it so it takes 30 seconds

1. 📁 Put it in a Project. Open Projects in the Claude sidebar, make one called "claim check", and paste the full fact-check prompt into the project instructions. Every chat inside it inherits the method, so from then on you paste a transcript and nothing else.
2. ⚡ Or package it as a skill. Ask Claude to write you a SKILL.md holding the four parts and the three follow-ups, then in the app go to Customize, then Skills, then Add, then Upload a skill (Anthropic's steps). Skills work on every plan, Free included. After that you type /claim-check and paste.
3. 🎯 Know when to skip it. This is for confident, specific, numbers-heavy claims that are about to change how you think or what you do. Clearly labelled opinions and obvious jokes do not need a literature review, and running it on everything is how a good habit turns into a chore you drop.

Proof

Greenwood, Guner, Kocharkov & Santos, "Marry your like", AER 2014
Gihleb & Lang, "Educational homogamy and assortative mating have not increased", NBER
Pew Research, equal-earning marriages, April 2023
Chetty et al., "The fading American dream", Opportunity Insights
Claude web search, available on all plans
Using Research on Claude