Claim verification checklist

Claim Verification Checklist

What This Skill Does

Generates a point-by-point checklist of every verifiable claim in a draft article, categorized by claim type and accompanied by the specific evidence needed to confirm each one.

When To Use This Skill

• A story is close to publication and needs a systematic fact-check before the editor or legal team sees it
• You are the assigning editor and want to hand a reporter a clear list of what still needs to be verified
• You have inherited a draft from another writer and need to understand what is sourced and what is asserted
• You want to ensure no factual claims slip through on deadline without a documented verification method

What You Need To Provide

Required: The full text of the draft article or a detailed section-by-section summary Optional: A note on which claims are already verified (so the checklist can mark them as such); the publication's fact-checking standard (minimal, standard, or rigorous); the story type (news, investigation, profile, feature)

How the Assistant Approaches This

1. Reads the draft and extracts every claim that is factually verifiable — distinguishing factual assertions from opinions, characterizations, and attributed quotes
2. Categorizes each claim by type: statistical, biographical, chronological, attribution (quote or paraphrase), official record, scientific, or legal
3. For each claim, specifies the minimum evidence standard needed for verification — primary source, secondary corroboration, expert confirmation, or documentary proof
4. Flags highest-risk claims: those that are specific, potentially defamatory, or that directly contradict public statements by named individuals
5. Closes with a "Next Step" note: which High-risk item to verify first, estimated time to clear the checklist, and whether libel-check-brief should be run alongside for any drafts containing allegations against named individuals

Output Format

Numbered checklist. Each item: Claim (exact quote from draft) | Claim Type | Required Evidence | Risk Level (Low / Medium / High). Grouped by section of the article. Followed by a Summary section noting total claim count, High-risk count, and items already marked as verified. Length scales with article length — typically 400–800 words for a 1,500-word article. Output ends with a "Next Step" note: which High-risk claim to verify first, how long the full check is likely to take, and whether libel-check-brief should be run on the same draft.

Quality Criteria

• Every specific, verifiable claim in the draft is captured — no omissions
• Claims are reproduced accurately from the draft, not paraphrased
• Evidence requirements are specific — not "needs a source" but "needs the original 2023 court filing" or "needs confirmation from the named agency's press office"
• High-risk claims are correctly identified — specificity, named individuals, and contradiction of official positions all elevate risk
• Summary gives an accurate picture of the draft's overall verification status
• Output includes at least one concrete next action the user can take immediately (start with the highest-risk item, run libel-check-brief, or confirm the structural concern noted in the summary)

Example

Input

Draft excerpt: A 900-word investigative feature about rising injury rates among warehouse workers at large fulfillment centers.

Key passages:

• "Warehouse injury rates have increased 32 percent over the past five years, according to federal labor statistics."
• "Workers at fulfillment centers are nearly three times more likely to be injured on the job than employees at traditional retail warehouses."
• "The company declined to comment."
• "In 2022, the company settled a workplace safety complaint with OSHA for $187,000 — one of the largest penalties in the region that year."
• "Former safety manager Diane Park, who left the company in 2021, told this reporter the injury tracking system was deliberately manipulated to underreport incidents."
• "The company had a recordable injury rate of 7.4 per 100 workers in 2023, compared to an industry average of 4.1."

Output

Claim Verification Checklist — Warehouse Injury Investigation

---

Section: Lede and Statistics

1. Claim: "Warehouse injury rates have increased 32 percent over the past five years" Type: Statistical Required Evidence: Pull the exact BLS (Bureau of Labor Statistics) data series (NAICS code for warehousing and storage, 2019–2024). Confirm the 32% figure matches the published injury and illness rate tables. Note the precise metric used (DART rate, TCIR, or recordable incidents). Risk Level: Medium — specific percentage requires exact sourcing; common error is comparing different metrics across years.

2. Claim: "Workers at fulfillment centers are nearly three times more likely to be injured than employees at traditional retail warehouses" Type: Statistical / Comparative Required Evidence: Identify the exact study or dataset making this comparison. "Nearly three times" is a specific multiplier — confirm the source, methodology, and publication date. If this comes from an advocacy group or union report rather than a government source, note the affiliation. Risk Level: High — comparative injury rate claims are frequently contested; methodology matters.

Section: Company Specifics

3. Claim: "The company declined to comment" Type: Attribution / Process Required Evidence: Document the outreach: date, method (email/phone/formal letter), recipient name or title, and absence of response or explicit decline. Keep correspondence on file. Risk Level: Low — standard process claim, but outreach must be documented.

4. Claim: "In 2022, the company settled a workplace safety complaint with OSHA for $187,000" Type: Official Record / Legal Required Evidence: Pull the OSHA inspection record from OSHA's public enforcement database (osha.gov). Confirm the settlement amount, year, and citation classification. Verify this is a settlement, not a penalty under appeal. Risk Level: High — specific dollar figure and year; legal records must match exactly. "One of the largest penalties in the region that year" requires comparative data — pull OSHA regional enforcement data for 2022 to verify.

5. Claim: "The company had a recordable injury rate of 7.4 per 100 workers in 2023" Type: Statistical / Official Record Required Evidence: Confirm whether this figure comes from OSHA 300 logs (which may be obtained by request or public filing), a regulatory filing, or a leak. If from a source, state that attribution clearly. If from public data, cite the exact filing. Risk Level: High — specific internal metric; if obtained from a leaked document, consult legal before publication.

Section: Source Testimony

6. Claim: Former safety manager "Diane Park...told this reporter the injury tracking system was deliberately manipulated to underreport incidents" Type: Attribution / Allegation Required Evidence: Confirm recording or contemporaneous notes exist for this interview. This is a serious allegation of intentional misconduct against the named company. The word "deliberately" is legally significant. Corroboration required: at least one additional source making the same or similar allegation, or documentary evidence of the manipulation. Company must be given right of reply specifically to this allegation before publication. Risk Level: High — named source making misconduct allegation; defamation exposure if not corroborated and right-of-reply not documented.

---

Summary

• Total verifiable claims: 6
• High risk: 4
• Medium risk: 1
• Low risk: 1
• Items pre-verified by reporter: 0 (assume none unless marked)

Recommend legal review before publication given the specific misconduct allegation in item 6 and the contested statistical claims in items 1 and 2.

Known Limitations

• This skill identifies verifiable claims but cannot verify them. It is a roadmap for fact-checking, not a substitute for it.
• Opinion, characterization, and attributed quotes are generally excluded from the checklist — but the skill may miss some edge cases where attributed opinions contain embedded factual assertions.
• For very long articles (3,000+ words), the checklist can become unwieldy. Consider running the skill on one section at a time.
• The skill cannot assess whether a claim is already verified from context not included in the draft text. Mark verified claims explicitly in the input to get a clean working checklist.

Related Skills

• statistics-fact-checker
• libel-check-brief
• source-credibility-brief