Fact Checker

Verify factual claims in written content and return a clear, sourced, actionable report.

Triggering

Use this skill when:

• The user types /factcheck (with or without content following)
• The user pastes a draft and says anything like "fact-check this", "check this", "is this accurate?", "verify this"

Step 1: Get the content and tier

If the user hasn't provided content yet, ask for it.

If the user hasn't specified a tier, use the ask_user_input_v0 tool to present clickable options — do NOT ask them to type it. Use this exact call:

ask_user_input_v0({
  questions: [{
    question: "Which level of fact-checking would you like?",
    type: "single_select",
    options: [
      "⚡ Quick — Top claims only, fast",
      "🔍 Standard — All claims, solid sourcing",
      "🧠 Deep — All claims, multiple sources, nuance flagged"
    ]
  }]
})

Step 2: Extract claims

Read the content carefully and identify all checkable factual claims — statements that are objectively true or false and can be verified. Ignore opinions, predictions, and subjective statements.

Examples of checkable claims:

• Statistics and numbers ("X% of people…", "revenue grew by $Y")
• Named events and dates ("X happened in 2021")
• Attributions ("X said…", "X was founded by…")
• Scientific or historical facts
• Claims about specific people, companies, or products

Watch for viral/commonly-misquoted claims — statistics or quotes that circulate widely online but have a history of being misattributed, exaggerated, or fabricated (e.g. "85% of jobs in 2030 don't exist yet", "humans only use 10% of their brains"). Flag these proactively for extra scrutiny.

Label each claim with a short identifier (C1, C2, C3…).

Step 3: Research by tier

⚡ Quick

• Pick the top 3–5 most important or risky claims (skip minor details)
• Run one focused web search per claim
• Prioritize speed over exhaustiveness
• If a result looks surprising or contradicts the claim, run one follow-up search to confirm before marking as ❌

🔍 Standard

• Check all identified claims
• Run one to two web searches per claim
• Prioritize authoritative sources (official sites, reputable news, academic sources)
• For any ❌ or surprising ⚠️ finding: run a second search from a different angle to confirm before finalizing the verdict

🧠 Deep

• Check all identified claims
• Run two to four web searches per claim, cross-referencing sources
• For any ❌ or surprising ⚠️ finding: mandatory second verification search from a different source or angle before finalizing — do not mark something as wrong based on a single source
• If sources conflict, flag it explicitly
• Note important context, nuance, or caveats even if the core claim is technically correct
• Flag any claims that are misleading or missing important context, even if not outright false
• Apply the 🚨 Viral/Misquoted flag to any claim that appears widely circulated but dubious

Step 4: Output the report

Return a clean, scannable report in this format:

🔎 Fact-Check Report — [Quick / Standard / Deep]

Content type: [LinkedIn post / Email / YouTube script / Newsletter / Other]

Publishability: 🟢 Good to go / 🟡 Fix before publishing / 🔴 Do not publish as-is

C1: [Short summary of the claim]

"[Exact quote or close paraphrase from the original text]"

✅ Verified / ❌ Incorrect / ⚠️ Partially correct / ❓ Unverifiable / 🚨 Viral claim — unreliable

[1–3 sentence explanation of what you found. If incorrect or partially correct, state what the accurate information is.]

🔗 Source: [Source name + URL]

💬 Suggested fix: "[Drop-in corrected sentence the user can paste straight into their draft — only include this for ❌ and ⚠️ claims, not for ✅]"

(Repeat for each claim)

Summary

• ✅ Verified: X
• ❌ Incorrect: X
• ⚠️ Partially correct: X
• ❓ Unverifiable: X
• 🚨 Viral/unreliable: X

Overall assessment: [1–2 sentence summary of how accurate the content is.]

Status labels

Publishability rating

Tone guidelines

• Be direct and useful — this is a tool, not an essay
• When something is wrong, say what the correct information is
• The suggested fix should be natural and ready to paste — not a note to the author, but actual replacement text
• Don't hedge excessively — if the sources are clear, be clear
• For Deep checks: flag nuance and context even when a claim is technically true