Fact-checking prompts

Help me check this claim.

THE CLAIM: [STATE IT EXACTLY AS YOU HEARD OR READ IT]
WHERE I ENCOUNTERED IT: [SOURCE]
WHY IT MATTERS: [WHAT YOU WOULD DO WITH IT]
ANY SUPPORTING MATERIAL: [PASTE]

Produce:

1. WHAT THE CLAIM ACTUALLY ASSERTS - broken into its separate factual components. A single sentence often contains three claims of different strength, and usually only one is contested.

2. WHAT KIND OF CLAIM EACH IS:
   - A verifiable fact with a definite answer
   - A statistic, which depends entirely on definition and method
   - A prediction, which cannot be verified now
   - An interpretation or judgement, which is not a fact and cannot be fact-checked
   - A definitional claim, where the disagreement is about words
   This classification often dissolves the dispute.

3. THE DEFINITION DEPENDENCY - for any statistic, what would have to be defined for it to be checkable: what is counted, over what period, for which population. State how the answer changes under different reasonable definitions.

4. THE PLAUSIBILITY CHECK - is the claim consistent with things generally known? Do a rough order-of-magnitude estimate independently and compare. Implausible claims are usually either wrong or measuring something other than what you assumed.

5. THE ORIGIN QUESTION - where would this claim come from? Trace it as far as the material allows. Widely-repeated figures often have no traceable origin, and that is itself a finding.

6. THE COMMON DISTORTIONS to check for: a real finding stripped of its qualifications, a figure from one context applied to another, a projection reported as a measurement, an old figure presented as current, and a correlation restated causally.

7. WHAT WOULD VERIFY IT - the specific source that would settle this, and what to search for.

8. MY LIMITATION - I cannot browse or access current sources. I can tell you whether a claim is internally coherent, plausible, and what to check, but I cannot confirm it. State clearly where my assessment is reasoning rather than verification.

9. THE VERDICT - plausible and checkable / implausible / unverifiable as stated / not a factual claim. With what to do next.

If I do not have enough to assess it, say so rather than producing a confident-sounding answer.

Check this for errors before I publish it.

THE CONTENT:
"""
[PASTE]
"""

WHAT IT IS: [ARTICLE / REPORT / PRESENTATION / POST]
AUDIENCE: [WHO, AND HOW EXPERT]
WHAT I AM SURE OF: [THE CLAIMS YOU KNOW ARE SOLID]
WHAT I AM LESS SURE OF: [THE ONES YOU ARE GUESSING AT]

Produce a table: Claim | Type | Risk if wrong | How to verify | Priority.

Identify:

1. EVERY FACTUAL CLAIM - including the ones stated so confidently they read as background. The dangerous errors are in assertions nobody thought to check.

2. EVERY NUMBER - and for each: is the source stated, is the unit clear, is the period specified, is the comparison basis given? Check any arithmetic present. Verify that percentages are consistent with the figures they derive from.

3. INTERNAL CONTRADICTIONS - places where the piece says two incompatible things, or where a figure in one section does not match the same figure elsewhere.

4. OVERSTATEMENT - claims stronger than the evidence given. Look for: 'proves', 'always', 'never', 'all', 'the most', 'significantly', and causal language attached to correlational evidence.

5. UNSOURCED ASSERTIONS - anything that sounds like a fact but has no attribution, especially statistics and historical claims.

6. THE DATE PROBLEM - anything that will be wrong within a year: current figures, 'recently', 'the latest', named versions, prices, and positions people hold. Flag for either a date stamp or rephrasing.

7. THE THINGS I SAID I WAS LESS SURE OF - assess each specifically. Where my uncertainty should be visible in the text and is not, say so.

8. WHAT AN EXPERT WOULD OBJECT TO - given my stated audience, the claims a knowledgeable reader would challenge, and whether they would be right.

9. THE HIGHEST-RISK ERROR - the one that would most damage credibility if wrong. Verify this one first.

10. WHAT I CANNOT CHECK - be explicit that I cannot verify facts against current sources. This is a list of what to check, not a confirmation that anything is correct.

Help me trace this statistic.

THE STATISTIC: [THE NUMBER AND WHAT IT CLAIMS]
WHERE I SAW IT: [SOURCE]
HOW IT WAS ATTRIBUTED: [WHAT CITATION WAS GIVEN, if any]
WHAT I HAVE FOUND SO FAR: [YOUR RESEARCH]

Produce:

1. WHAT THE STATISTIC WOULD REQUIRE - for this number to exist, someone must have measured something specific. What would they have had to measure, on whom, how, and when? Often this reveals immediately that the measurement is impractical or impossible, which tells you the figure is an estimate or an invention.

2. THE DEFINITION QUESTIONS - what would need defining for this number to be meaningful. Different reasonable definitions usually produce very different figures, which explains why several incompatible versions of the same statistic circulate.

3. THE LIKELY CHAIN - how a figure like this typically propagates: an original study or estimate, a press release simplifying it, an article simplifying that, and then endless repetition detached from the source. At which stage does the version I have look like it sits?

4. THE TRANSFORMATION CHECK - common ways figures mutate in transmission:
   - A projection becomes a measurement
   - A subgroup finding becomes a general claim
   - An estimate's range becomes its midpoint stated as fact
   - A figure for one country becomes global
   - An old figure remains in circulation unchanged for years
   - A hypothetical or illustrative number becomes a citation
   Which of these would explain a discrepancy in what I have found?

5. WHO WOULD PRODUCE THIS - the kinds of organisation that measure this sort of thing: statistical agencies, regulators, industry bodies, academic researchers, or a vendor with a commercial interest. Name the type to search for.

6. THE SEARCH STRATEGY - what to search for to find the primary source, including the technical terminology likely used by whoever produced it.

7. THE RED FLAG ASSESSMENT - characteristics suggesting this figure may have no solid origin: a suspiciously round number, no date, no methodology ever mentioned, attribution to an organisation rather than a specific publication, and citation only to other articles rather than to research.

8. WHAT TO DO IF YOU CANNOT FIND IT - do not repeat it. State the alternative: use a figure you can source, describe the phenomenon without a number, or say plainly that estimates vary and no authoritative figure exists.

I cannot search for you. This is a method for finding it, not a verification.

Analyse the reasoning in this argument.

THE ARGUMENT:
"""
[PASTE]
"""

CONTEXT: [WHO IS MAKING IT AND WHY]
MY POSITION: [WHAT YOU CURRENTLY THINK - be honest]

Produce:

1. THE ARGUMENT RECONSTRUCTED - the premises and the conclusion, stated clearly and fairly. Steelman it: state the strongest version of what is being argued, not the weakest. If the reconstruction is stronger than the original, say so - that is useful information.

2. DOES THE CONCLUSION FOLLOW - if all the premises were true, would the conclusion be established? This is separate from whether the premises are true, and separating them is most of the work.

3. THE PREMISES - each one: is it a fact, an assumption, a definition, or a value judgement? Which are contested? Which would need evidence?

4. THE HIDDEN PREMISE - almost every argument relies on something unstated. Name it. It is frequently where the real disagreement lives, and surfacing it often ends the dispute.

5. THE REASONING PROBLEMS actually present - only flag what is there, and name the specific passage:
   - Correlation treated as causation
   - A false choice between two options when others exist
   - Attacking a weakened version of the opposing view
   - Generalising from unrepresentative cases
   - Circular reasoning
   - An appeal to authority outside its expertise
   - Shifting the burden of proof
   - Conflating two senses of a word across the argument
   - Selective evidence
   - A claim that cannot be falsified

6. WHAT IS STRONG about the argument. Nearly every argument that persuades people has something right in it, and identifying it is more useful than listing fallacies.

7. WHAT WOULD SETTLE IT - the evidence that would resolve the disagreement, and whether it exists or could be gathered.

8. MY OWN POSITION, EXAMINED - I told you what I think. Apply the same scrutiny to my view: what does it assume, and what would make me wrong?

9. THE HONEST VERDICT - strong / partly sound / weak, and in what respect.

Do not tell me what I want to hear. If the argument I disagree with is sound, say so.

These two sources seem to contradict each other. Help me work out what is going on.

SOURCE A:
"""
[PASTE THE RELEVANT PART]
"""

SOURCE B:
"""
[PASTE THE RELEVANT PART]
"""

WHAT I AM TRYING TO ESTABLISH: [YOUR QUESTION]

Produce:

1. WHAT EACH ACTUALLY CLAIMS - stated precisely, in your words, with the qualifications each includes. Often the contradiction disappears at this step, because one or both were being read more broadly than they were written.

2. IS THIS A REAL CONTRADICTION - can both be true simultaneously? Work through:
   - DIFFERENT DEFINITIONS: the same term meaning different things. The most common explanation by a wide margin.
   - DIFFERENT POPULATIONS: true of different groups
   - DIFFERENT PERIODS: both true, at different times
   - DIFFERENT SCOPE: one general, one specific
   - DIFFERENT MEASURES: measuring related but distinct things
   - DIFFERENT THRESHOLDS: the same data with different cut-offs
   - ONE IS WRONG
   - BOTH ARE PARTIALLY RIGHT: the truth is conditional and neither states the condition

3. THE RECONCILIATION - if they can both be true, state the fuller picture that accommodates both. This is usually more accurate than either source alone.

4. IF ONE IS WRONG - which, and on what basis: method, recency, sample, independence, or expertise. Say if you cannot tell from what I gave you.

5. THE UNDERLYING DISAGREEMENT - if this is a genuine dispute, what is it actually about? Usually one of: what the evidence shows, what the terms mean, what standard of evidence is required, or what values apply. Naming which changes how you resolve it.

6. WHAT I SHOULD CONCLUDE for my stated question. Where the honest answer is that it depends on a definition, say which definition and let me choose knowingly.

7. WHAT WOULD RESOLVE IT - the evidence or clarification needed.

8. HOW TO REPRESENT THIS if I write about it - honestly acknowledging the disagreement rather than picking the source that suits me.

Do not resolve the contradiction by averaging the two claims. That is almost always wrong.

I generated this with an AI and I need to check it before using it.

THE CONTENT:
"""
[PASTE]
"""

WHAT IT IS FOR: [PURPOSE AND AUDIENCE]
HOW IT WILL BE USED: [published / decision / internal reference]
MY OWN KNOWLEDGE OF THE TOPIC: [expert / some knowledge / none]

Produce a verification checklist, ordered by risk:

1. THE FABRICATION RISK AREAS - the things models get wrong most often, in order:
   - CITATIONS AND REFERENCES: author names, titles, dates, journals, page numbers, URLs. These are frequently plausible and wrong. Every single one needs checking against the actual source.
   - SPECIFIC NUMBERS AND STATISTICS: particularly ones with no source attached
   - QUOTES: attributed quotations are high-risk
   - NAMES, DATES AND EVENTS: especially anything recent, niche, or involving less prominent people
   - LEGAL, REGULATORY AND TECHNICAL SPECIFICS: version numbers, clause references, thresholds, deadlines
   - ANYTHING RECENT: models have a knowledge cutoff and may confidently describe outdated states of affairs

   Go through my content and list every instance of each, as a checklist with the claim quoted.

2. THE PLAUSIBLE-BUT-WRONG CHECK - claims that sound authoritative and are the kind of thing models confabulate. These are more dangerous than obvious errors because they do not prompt scrutiny.

3. THE CONFIDENCE MISMATCH - places where the text states something with more certainty than the topic warrants. Models tend to write confidently regardless of underlying uncertainty.

4. THE OMISSION CHECK - what a knowledgeable person would expect to see here that is missing: a caveat, an exception, a competing view, or a standard qualification.

5. GIVEN MY KNOWLEDGE LEVEL - if I said I have no expertise, be explicit that I cannot evaluate this content myself and need either a knowledgeable reviewer or primary sources. This is the honest answer and the most important one.

6. THE PRIORITY ORDER - what to verify first, given how this will be used. Published content and decisions need everything checked; internal reference can be more selective.

7. WHAT CANNOT BE FIXED BY CHECKING - if the content rests on a framing or structure that is subtly wrong, checking individual facts will not catch it.

I am also a model and subject to the same failure modes. Treat this checklist as a method, not a verification, and do not treat my assessment of the content as confirmation of it.

Help me stress-test something I believe.

WHAT I BELIEVE: [THE BELIEF]
WHY I BELIEVE IT: [YOUR REASONS]
HOW CONFIDENT I AM: [percentage or description]
WHAT IT WOULD MEAN IF I AM WRONG: [THE STAKES]
WHERE THIS BELIEF CAME FROM: [experience / reading / someone told you / reasoning / you are not sure]

Produce:

1. THE BELIEF, CLARIFIED - stated precisely enough to be tested. Vague beliefs cannot be wrong, which is why they survive. Name any ambiguity in how I stated it.

2. THE EVIDENCE AUDIT - for each of my stated reasons:
   - What kind of evidence is it: direct observation, someone's testimony, a statistic, an inference, or an intuition?
   - How strong is it for this conclusion?
   - Could it support a different conclusion equally well?

3. THE SOURCE PROBLEM - I said where it came from. Assess that route: personal experience is vivid and unrepresentative; something widely repeated may have no origin; reasoning from first principles is only as good as the premises.

4. WHAT WOULD I EXPECT TO SEE if this belief were true, that I would not see if it were false? If nothing distinguishes the two cases, the belief is not connected to evidence and confidence in it is unearned.

5. THE STRONGEST CASE AGAINST - written properly, as someone who disagreed and was well-informed would put it. Not a strawman.

6. THE DISCONFIRMING EVIDENCE I MIGHT BE DISCOUNTING - things I have encountered and explained away.

7. THE SELECTION PROBLEM - if this belief comes from experience, what cases would I not have seen? Survivorship and selection shape most experiential beliefs.

8. THE CONFIDENCE CHECK - is my stated confidence justified by the evidence I gave? Over-confidence relative to evidence is the normal state, not an unusual failing. Say what level would be warranted.

9. WHAT WOULD CHANGE MY MIND - stated concretely. If I cannot answer this, the belief is not functioning as a belief about the world.

10. THE HONEST VERDICT - well-supported / plausible but under-evidenced / mostly inherited without examination / probably wrong.

Be direct. I am asking to be challenged, not reassured.