Fact-Check with AI: How to Verify Any Claim

You are scrolling through your feed and read: 'New study proves: AI can detect cancer five years earlier than any doctor.' Sounds impressive. You are about to share it -- but then you pause. Which study? From which university? How large was the sample size? Has it been confirmed by other researchers? You do not know. And you are not alone.

The problem: We are drowning in claims that sound convincing.

Every day you encounter dozens of statements that start with 'Studies show...' or 'Experts say...' On social media, in news, in conversations with colleagues. Most sound plausible. Some are true. Some are half-true -- a grain of truth surrounded by exaggeration. And some are simply false, but packaged so well that you do not notice.

The real problem: you have neither the time nor the expertise to research every claim yourself. So you do what most people do: you believe what fits your worldview and ignore the rest. That is human -- but dangerous. Because that is exactly how half-truths and misinformation spread.

AI as a fact-checking tool -- with one important caveat

AI models like ChatGPT, Claude, or Gemini are remarkably good at analyzing claims:

- Check logic: Is the argument internally consistent? Does the conclusion follow from the premises?
- Evaluate sources: Is the cited source reputable? Does the referenced study actually exist?
- Detect manipulation: Are common tricks being used -- false causality, cherry-picked data, appeals to authority?
- Provide counterarguments: What do critics say? Which perspective is missing?
- Explain complexity: If the claim involves a specialized topic, AI can make the background understandable.

But -- and this is crucial -- AI is not an oracle. AI models can hallucinate: they sometimes invent studies that do not exist or misquote sources. That is why AI is an analysis tool, not a truth detector. You use AI to ask the right questions and dissect arguments -- but the final verification is yours, by checking the sources mentioned.

This is what makes this challenge especially valuable: you learn not only how to use AI for fact-checking, but also how to critically question AI answers themselves.

The task (25 minutes, 3 phases):

Phase 1 -- Collect claims and quick-check (8 min)

Gather 3 claims you have encountered in the past few days. These could be:

- A headline from social media
- A statistic someone mentioned in a meeting
- A statement from a podcast or YouTube video
- A health tip someone sent you
- A claim about AI, technology, or the economy

If nothing comes to mind, use one of these:
- 'AI-generated text is now indistinguishable from human-written text.'
- 'Remote work reduces productivity by 20 percent.'
- 'In 10 years, 50 percent of all current jobs will be replaced by AI.'

Copy this prompt:

'You are an experienced fact-checker and media analyst. You work according to the standards of reputable fact-checking organizations like Snopes, Full Fact, or PolitiFact. Your approach: no claim is true or false until it has been verified. You analyze neutrally, thoroughly, and transparently.

I have collected three claims I want to check. Do a quick check on each:

Claim 1: [Insert first claim here]
Claim 2: [Insert second claim here]
Claim 3: [Insert third claim here]

For each claim:

1. Plausibility score (1-10): How likely is this claim to be true? 1 = very unlikely, 10 = very likely.

2. Quick analysis (3 sentences): What supports it, what contradicts it? Are there obvious warning signs?

3. Source evaluation: If a source is mentioned -- is it generally trustworthy? If no source is mentioned -- is that already a warning sign?

4. Manipulation check: Is any of these techniques being used?
- False causality (A happens after B, so B must cause A)
- Cherry-picking (only data supporting the thesis is shown)
- Appeal to authority (an expert says it, so it must be true)
- Emotionalization (fear or outrage instead of facts)
- False dichotomy (either X or Y, as if no other option exists)
- Exaggeration (a small effect is presented as revolutionary)

5. Verdict:
- Probably true
- Partially true (with caveats)
- Unverifiable (too vague or no sources)
- Probably not true
- False

6. Most important open question: What would need to be checked to definitively evaluate this claim?

Be honest: if you are not sure about something, say so. If you cannot verify information, explicitly point that out.'

Read through the quick checks and choose the claim that seems most interesting or questionable -- that is the one you will use for Phase 2.

Phase 2 -- Deep analysis of one claim (12 min)

Now you go deep on one claim. This is the core of the challenge: you learn how to systematically take a claim apart. Copy this prompt:

'Analyze the following claim in detail:

Claim: [The claim from Phase 1 that you selected]

Context: [Where did you find this claim? e.g. social media post, news article, conversation, advertisement]

Conduct a complete fact-check analysis:

1. Break down the claim:
- What exactly is being claimed? Formulate the core assertion in one sentence.
- What hidden assumptions are embedded in the claim?
- Is the claim verifiable? Or is it worded so vaguely that it can neither be proven nor disproven?

2. Source verification:
- If a study is cited: who conducted it? How large was the sample? Was it published in a peer-reviewed journal?
- If experts are cited: are they experts in this specific field? Or are they being used as authorities on a topic outside their expertise?
- If numbers are mentioned: where do they come from? Are they current? Are they presented in the right context?
- IMPORTANT: only cite sources you are confident actually exist. If you are unsure about a source, say so explicitly.

3. Logic check:
- Is the argument logically sound? Does the conclusion follow from the premises?
- Are there logical fallacies? (False causality, circular reasoning, straw man argument, etc.)
- Is it generalizing from a single case to the whole?

4. Context check:
- Is important context missing that would put the claim in perspective?
- Is a trend being exaggerated? (e.g. a small change presented as dramatic)
- Does the claim only hold under specific conditions that are not mentioned?

5. Counter-position:
- What do critics or skeptics say?
- Are there studies or data that contradict the claim?
- What alternative explanation exists?

6. Who benefits?
- Who has an interest in this claim being spread?
- Is there a financial, political, or ideological background?
- Is the claim being used to promote a specific product, agenda, or opinion?

7. Final verdict:
- True / Mostly true / Partially true / Mostly not true / False / Unverifiable
- Justification in 2-3 sentences
- What would need to change for the verdict to be different?

8. How to verify my analysis:
- Give me 2-3 specific search terms I can enter in Google Scholar, PubMed, or a news search engine to verify the claim myself.
- If you cited sources: how can I check whether these sources actually exist and say what you claim?'

Important: verify the AI answer!

After receiving the analysis, take 3 minutes to actually search for the mentioned sources or search terms. This is the crucial step: you use AI to ask the right questions -- but you verify the answers yourself.

Ask yourself:
- Did the AI cite real studies I can find?
- Are the numbers the AI mentioned accurate?
- Is there something the AI overlooked?

If you find inconsistencies, confront the AI: 'You cited a study by [author/institution]. I cannot find this study. Does it actually exist, or did you fabricate it?'

Phase 3 -- Build your personal fact-check toolkit (5 min)

Finally, you create a tool you can use again and again. Copy this prompt:

'Create a compact fact-check toolkit I can use whenever I encounter a questionable claim.

Part 1 -- Quick test (30 seconds):
Give me 5 yes/no questions I can ask myself about ANY claim to decide whether it deserves deeper scrutiny. The questions should cover the most common warning signs.

Part 2 -- Warning signs checklist:
List the 10 most common warning signs that indicate a claim may be false or misleading. For each warning sign: a concrete example of what it looks like in practice.

Part 3 -- Fact-check prompt (reusable):
Write a compact, reusable prompt (maximum 10 lines) I can copy every time I want to check a claim. The prompt should contain the key verification steps from Phase 2 in condensed form.

Part 4 -- Trustworthy sources:
Give me 5-8 fact-checking websites and research tools (international and well-known) I can use for cross-verification. For each source: what it is particularly good at.

Part 5 -- The golden rule of fact-checking:
Formulate in one sentence the most important takeaway from this challenge.'

Three examples of how fact-checking with AI works in practice:

Example 1 -- Social media claim:
Claim: 'ChatGPT uses 10 times as much energy per query as a Google search.'
Quick check: Sounds plausible -- AI models do require more computing power than a simple search. But 10 times as much? The number varies widely depending on the study and model. Some estimates cite 3 to 10 times as much, others significantly less with optimized systems. The '10 times' is a simplification that may hold in some cases but is misleading as a blanket statement.
Verdict: Partially true -- the general trend is correct, but the specific number is context-dependent and often cited without a source.

Example 2 -- Health claim:
Claim: 'Intermittent fasting extends life expectancy by up to 30 percent.'
Quick check: A 30 percent life extension would be a sensational result. Such effects have been observed in laboratory experiments with mice, but mouse studies cannot be directly transferred to humans. Human studies show positive effects on weight and metabolism, but no robust data for 30 percent life extension.
Verdict: Mostly not true -- the effect in mice is presented as fact for humans. A classic case of transferring animal studies to humans without qualification.

Example 3 -- Business claim:
Claim: 'Remote work reduces productivity by 20 percent.'
Quick check: There are studies showing reduced productivity with remote work, and studies showing increased productivity. The 20 percent likely comes from one specific study with specific conditions (e.g. particular industry, particular country, during the pandemic). As a blanket statement, it is wrong -- the effect depends heavily on the type of work, company culture, and individual circumstances.
Verdict: Partially true -- a single study is being generalized despite mixed research findings.

Why this works:

Fact-checking is not an innate ability -- it is a technique with clear steps: break down the claim, check sources, test logic, assess context, seek counter-positions. Most people do not skip these steps out of laziness, but because they do not know which questions to ask.

AI is a powerful analysis tool for this: it can structure arguments, identify logical errors, and provide counter-positions you would not have thought of on your own. But -- and this is the most important lesson -- AI does not replace your critical thinking. It supports it. The final verification stays with you.

The greatest benefit: you develop an instinct for when your inner alarm should ring. After this challenge, you will no longer simply accept claims at face value. You will automatically ask: which source? What context? Who benefits? And that changes how you consume information -- permanently.

Get even more out of it:
- Analyze a news source: 'Analyze this news source [name of source]: how reputable is it? What is its political leaning? Have there been cases of misinformation in the past?'
- Statistics translator: 'This statistic is cited in an article: [insert statistic]. Explain to me as a layperson: what does this number actually say? What does it NOT say? How could it be interpreted misleadingly?'
- Viral check: 'This social media post went viral: [insert post]. Analyze: why is this post spreading so fast? What psychological triggers are being used? Is the content accurate?'
- Check your own biases: 'I firmly believe that [your belief]. Search for the strongest arguments AGAINST my position. I want to find my own blind spots.'

Your learning outcome: You now know how to break a claim into its components: core assertion, hidden assumptions, sources, logic, context, and interest. You experienced that AI is excellent at asking the right questions, but that you must verify the answers yourself. You have a reusable toolkit with a quick test, warning signs checklist, and fact-check prompt that you can use for any questionable claim. The most important insight: the claim itself is not the problem -- the problem is accepting it without checking.