Extract claims and facts to verify

Pull every checkable claim, number and quote from a piece into a fact-check list with a priority and source type.

LLazyPrompt(@lazyprompt)WritingExtractWorks with any model
Prompt template
Read the text below and build a fact-checking list.

Text:
{{text}}

What the text is for: {{purpose}}
Publication standard: {{standard}}

Extract every statement that can be checked: statistics, dates, named people and organisations, quotations, causal claims, superlatives ("first", "largest") and attributed opinions.

For each item provide:
- The exact wording from the text, in quotes.
- Claim type (statistic, quote, date, attribution, causal claim, superlative, other).
- Risk if wrong: High, Medium or Low (High if it could mislead readers, damage someone's reputation, or has legal implications).
- What kind of source would verify it (for example official statistics, original study, primary document, direct confirmation from the person).

Rules:
- Do not state whether claims are true. You may note that a claim looks unusually precise or is missing a source.
- Do not invent sources or links.
- Order the list by risk, High first.

Output format: a table with columns #, Claim, Type, Risk, Verify with. Then a short list of claims in the text that have no source cited.
Example output
| # | Claim | Type | Risk | Verify with | |---|---|---|---|---| | 1 | "Office work is dead" (attributed to CEO Sam Park) | Quote | High | Direct confirmation or original transcript | | 2 | "Remote work has caused a boom in small-town housing prices" | Causal claim | High | Housing market data with analysis of cause | | 3 | "A 2023 study found that remote workers are 13% more productive" | Statistic | Medium | The original study, including sample and definition | | 4 | "87% want to keep working from home" | Statistic | Medium | Survey methodology and questionnaire | | 5 | "survey of 2,000 employees" | Statistic | Low | Internal survey records | Claims with no source cited - #1 (no date or venue for the quote) - #2 (no data, and causation asserted) - #3 (study not named)

Why it works

This works as a pre-publication checklist for writers and editors. It deliberately does not judge truth, since a model cannot verify claims reliably on its own.

Licensed CC0 (public domain) Version 1 · history