Fact-Checking Prompt - AiPro Institute\u2122<\/title>\n <style>\n *{margin:0;padding:0;box-sizing:border-box}\n body{font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,'Helvetica Neue',Arial,sans-serif;line-height:1.6;color:#333;background:#fff;padding:2rem 1rem}\n .container{max-width:900px;margin:0 auto}\n .page-title{text-align:center;font-size:2.5rem;font-weight:700;background:linear-gradient(135deg,#667eea 0%,#764ba2 100%);-webkit-background-clip:text;-webkit-text-fill-color:transparent;background-clip:text;margin-bottom:2rem}\n .card{background:#fff;border-radius:12px;box-shadow:0 4px 6px rgba(0,0,0,.1);overflow:hidden;margin-bottom:2rem}\n .card-header{background:linear-gradient(135deg,#667eea 0%,#764ba2 100%);color:#fff;padding:2rem}\n .card-header h1{font-size:2rem;margin-bottom:.5rem}\n .card-header .subtitle{font-size:1.1rem;opacity:.95}\n .meta-badges,.tool-badges{display:flex;gap:.75rem;margin-top:1rem;flex-wrap:wrap}\n .badge{background:rgba(255,255,255,.2);padding:.4rem .9rem;border-radius:20px;font-size:.9rem;backdrop-filter:blur(10px)}\n .tool-badge{background:transparent;border:1px solid rgba(255,255,255,.4);padding:.4rem .9rem;border-radius:20px;font-size:.85rem}\n .card-body{padding:2.5rem}\n .section-title-container{display:flex;justify-content:space-between;align-items:center;margin:2.5rem 0 1.25rem 0}\n .section-title-container:first-child{margin-top:0}\n .section-title{font-size:1.75rem;color:#764ba2;border-left:4px solid #764ba2;padding-left:1rem;margin:0}\n .copy-button{background:linear-gradient(135deg,#667eea 0%,#764ba2 100%);color:#fff;border:none;padding:.6rem 1.5rem;border-radius:6px;cursor:pointer;font-size:.95rem;font-weight:500;transition:opacity .3s}\n .copy-button:hover{opacity:.9}\n .prompt-box{background:#f8f9fa;border:1px solid #dee2e6;border-radius:8px;padding:1.5rem;margin:1.25rem 0;font-family:'Courier New',monospace;font-size:.95rem;line-height:1.6;white-space:pre-wrap;overflow-x:auto}\n .placeholder{color:#fd7e14;font-weight:bold}\n .tip-box{background:#fff9e6;border-left:4px solid #ffc107;padding:1.25rem;margin:1.25rem 0;border-radius:4px}\n .tip-box strong{color:#f57c00}\n h3{color:#764ba2;font-size:1.35rem;margin:2rem 0 1rem 0}\n p{margin-bottom:1rem;line-height:1.8}\n ul,ol{margin-left:2rem;margin-bottom:1rem}\n li{margin-bottom:.5rem;line-height:1.8}\n .example-output{background:#f0f8ff;border:2px solid #4a90e2;border-radius:8px;padding:1.5rem;margin:1.25rem 0}\n .example-output h4{color:#4a90e2;margin-bottom:1rem}\n .chain-step{background:#f8f9fa;border-left:4px solid #667eea;padding:1.5rem;margin:1.5rem 0;border-radius:4px}\n .chain-step h4{color:#667eea;margin-bottom:.75rem}\n .footer{background:#f8f9fa;padding:2rem;margin-top:2rem;border-radius:8px;display:flex;justify-content:space-around;align-items:center;flex-wrap:wrap;gap:1.5rem}\n .footer-stat{text-align:center}\n .footer-stat-value{font-size:1.75rem;font-weight:700;color:#764ba2}\n .footer-stat-label{color:#666;font-size:.95rem}\n @media (max-width:768px){.page-title{font-size:1.75rem}.card-header h1{font-size:1.5rem}.card-body{padding:1.5rem}.section-title{font-size:1.35rem}.section-title-container{flex-direction:column;align-items:flex-start;gap:1rem}.footer{flex-direction:column}}\n <\/style>\n<\/head>\n<body>\n <div class=\"container\">\n <h1 class=\"page-title\">Fact-Checking Prompt<\/h1>\n\n <div class=\"card\">\n <div class=\"card-header\">\n <h1>Fact-Checking Prompt<\/h1>\n <p class=\"subtitle\">AI Safety & Governance<\/p>\n <div class=\"meta-badges\">\n <span class=\"badge\">\u23f1\ufe0f 20-30 minutes<\/span>\n <span class=\"badge\">\ud83d\udcca Intermediate<\/span>\n <\/div>\n <div class=\"tool-badges\">\n <span class=\"tool-badge\">ChatGPT<\/span>\n <span class=\"tool-badge\">Claude<\/span>\n <span class=\"tool-badge\">Gemini<\/span>\n <span class=\"tool-badge\">Perplexity<\/span>\n <span class=\"tool-badge\">Grok<\/span>\n <\/div>\n <\/div>\n\n <div class=\"card-body\">\n <div class=\"section-title-container\">\n <h2 class=\"section-title\">The Prompt<\/h2>\n <button class=\"copy-button\" onclick=\"copyPrompt()\">\ud83d\udccb Copy Prompt<\/button>\n <\/div>\n\n <div class=\"prompt-box\" id=\"promptContent\">You are a rigorous fact-checker and verification analyst. Fact-check the following content and produce an evidence-based verification report.\n\n<span class=\"placeholder\">[CONTENT_TO_FACT_CHECK]<\/span> (paste article, post, transcript, or claims)\n\n<span class=\"placeholder\">[TOPIC_DOMAIN]<\/span> (e.g., \"health\", \"finance\", \"politics\", \"science\", \"law\", \"product claims\")\n\n<span class=\"placeholder\">[REQUIRED_RIGOR]<\/span> (e.g., \"journalistic\", \"academic\", \"legal-grade\")\n\n<span class=\"placeholder\">[SOURCE_CONSTRAINTS]<\/span> (e.g., \"must use primary sources\", \"web sources allowed\", \"only internal docs\")\n\nUse the V.E.R.I.F.Y. FRAMEWORK:\n\n**V - Verify Claims List** \u2192 Extract all factual claims and turn them into checkable statements\n**E - Evidence Collection** \u2192 Identify what evidence is needed and where to find it\n**R - Reliability Scoring** \u2192 Rate sources (primary, secondary, partisan, anonymous, outdated)\n**I - Inconsistency Detection** \u2192 Spot contradictions, missing context, misleading framing\n**F - Findings & Verdicts** \u2192 Label each claim: True\/False\/Misleading\/Unproven\n**Y - Yield a Corrected Summary** \u2192 Provide corrected version with citations and uncertainty notes\n\nDELIVER 10 COMPONENTS:\n\n\u2713 1. Claim Extraction (list 10-30 claims)\n\u2713 2. Claim Prioritization (top 5 high-impact claims)\n\u2713 3. Evidence Needed (per claim)\n\u2713 4. Source Plan (primary vs secondary sources)\n\u2713 5. Verification Results (per claim: verdict + evidence)\n\u2713 6. Confidence Level (high\/medium\/low)\n\u2713 7. Missing Context & Misleading Framing\n\u2713 8. Corrected Summary (rewrite with verified facts only)\n\u2713 9. Open Questions (what remains unknown)\n\u2713 10. Appendix (sources used + quotes)\n\nOUTPUT FORMAT:\n\n## Claim List\n[numbered claims]\n\n## Priority Claims\n[top 5]\n\n## Verification Table\n[table: claim \u2192 verdict \u2192 confidence \u2192 evidence (quotes + citations) \u2192 notes]\n\n## Missing Context \/ Misleading Framing\n\n## Corrected Summary (Verified Only)\n\n## Open Questions\n\n## Sources & Quotes Appendix\n\nConstraints:\n- If you cannot verify a claim, mark it UNPROVEN (do not guess)\n- Distinguish \u201cno evidence found\u201d from \u201cevidence shows false\u201d\n- Quote sources directly for key claims\n- Use clear verdict labels and confidence levels\n<\/div>\n\n <div class=\"tip-box\"><strong>\ud83d\udca1 Pro Tip:<\/strong> Treat every number, date, and named entity as a separate checkable claim. Most misinformation hides in \u201csmall specifics\u201d rather than the main narrative.<\/div>\n\n <div class=\"section-title-container\"><h2 class=\"section-title\">The Logic<\/h2><\/div>\n\n <h3>1. Claim Extraction Prevents \u201cVibes Checking\u201d and Forces Measurable Verification<\/h3>\n <p><strong>WHY IT WORKS:<\/strong> Many so-called fact checks evaluate an article\u2019s overall tone or political alignment rather than verifying specific claims. Breaking content into discrete, checkable statements converts an amorphous task (\u201cis this true?\u201d) into a structured verification workflow. This also prevents the \u201ctruthy narrative\u201d problem: a story can feel plausible while containing several wrong specifics. Claim extraction ensures each number, date, entity, and causal statement is tested. It also improves transparency: readers can see exactly what was checked and what wasn\u2019t. In professional workflows, claim lists are the backbone of verification because they create auditability and reproducibility.<\/p>\n <p><strong>EXAMPLE:<\/strong> A post says \u201cCountry X cut emissions 40% since 2010, saving $2B in healthcare costs, according to a UN report.\u201d That\u2019s at least three claims: (1) emissions change magnitude and baseline, (2) healthcare savings amount, (3) source attribution (\u201cUN report\u201d). A claim-based fact check will verify each separately, often finding that one is accurate while another is overstated. When teams adopt claim extraction, they reduce error rates in published summaries because they stop repeating unverified numbers. It also helps prioritize: a wrong \u201c$2B\u201d claim might matter more than a minor date error.<\/p>\n\n <h3>2. Evidence Hierarchies Reduce Misinformation Amplification by Favoring Primary Sources<\/h3>\n <p><strong>WHY IT WORKS:<\/strong> The internet is full of circular reporting: one article cites another which cites a tweet. An evidence hierarchy prioritizes primary sources (official data, filings, peer-reviewed papers, direct transcripts) over secondary commentary. This reduces the risk of amplifying unverified claims. Reliability scoring forces explicit judgment about source quality: primary vs. secondary, conflict of interest, outdatedness, and methodology transparency. This makes the fact check defensible and reduces \u201cappeal to authority\u201d mistakes where a reputable outlet is treated as proof without confirming its underlying sources.<\/p>\n <p><strong>EXAMPLE:<\/strong> Financial claim: \u201cCompany Y revenue grew 50% last quarter.\u201d The correct primary source is the company\u2019s earnings report (10-Q), not a blog summary. By forcing a primary-source-first plan, you avoid errors where a blog confuses revenue with bookings. In health claims, peer-reviewed studies and government datasets (CDC, WHO) outrank influencer posts. When the evidence hierarchy is explicit, you can also mark uncertainty: \u201cOnly secondary sources exist; claim remains unproven.\u201d This prevents overconfident conclusions.<\/p>\n\n <h3>3. Verdict Labels Separate Unknown From False, Preventing Overconfident Claims<\/h3>\n <p><strong>WHY IT WORKS:<\/strong> Many fact checks wrongly equate \u201cwe couldn\u2019t verify it\u201d with \u201cit\u2019s false,\u201d or they assume \u201csounds false\u201d without evidence. A disciplined system uses verdict labels: True, False, Misleading (partly true but missing context), Unproven (insufficient evidence), and Unclear (ambiguous claim). This reduces unjustified certainty. It also helps users: knowing something is unproven suggests caution, not dismissal. In governance contexts, this precision is essential because decisions can depend on evidence quality.<\/p>\n <p><strong>EXAMPLE:<\/strong> A claim: \u201cA new law bans encryption.\u201d If evidence is unclear because the bill is proposed, not passed, the right label is Misleading or Unproven depending on context. Another claim: \u201cStudy proves coffee causes cancer.\u201d The evidence might show correlation in one cohort but not causation; verdict: Misleading. By using clear labels, you avoid sensationalism and maintain credibility. This also supports automation: systems can route \u201cUnproven\u201d items to further research and treat \u201cFalse\u201d items differently (e.g., correction required).<\/p>\n\n <h3>4. Inconsistency Detection Catches Framing Tricks That Avoid Direct Lies<\/h3>\n <p><strong>WHY IT WORKS:<\/strong> Much misinformation is not outright falsehood but misleading framing: cherry-picked time windows, base rate neglect, omitted denominators, correlation-causation confusion, and quote mining. An inconsistency module looks for internal contradictions (numbers that don\u2019t add up), missing context (per-capita vs total), and rhetorical sleights (anonymous sourcing presented as fact). This is crucial because you can \u201cfact check\u201d individual sentences as true while the overall implication is misleading. Inconsistency detection reveals the gap between literal truth and honest interpretation.<\/p>\n <p><strong>EXAMPLE:<\/strong> \u201cCrime increased 20%\u201d may be true for one month vs previous month, but misleading if year-over-year crime decreased. Or \u201cvaccine adverse events doubled\u201d may be true due to reporting changes, not actual harm. In financial reporting, \u201cprofits fell 30%\u201d might omit that profits were unusually high last year (base effect). Inconsistency detection forces the report to include denominators, time frames, and alternative baselines. This reduces manipulation and improves reader understanding.<\/p>\n\n <h3>5. Corrected Summaries Prevent the \u201cDebunk Without Replacement\u201d Problem<\/h3>\n <p><strong>WHY IT WORKS:<\/strong> Simply labeling a claim false leaves a vacuum; readers often retain the original misinformation due to repetition and lack of an alternative coherent narrative. A corrected summary provides a replacement story using verified facts only, with uncertainty notes. This supports better memory encoding: people remember the corrected explanation rather than the debunk label. It also enables downstream reuse: journalists, analysts, and teams can publish the corrected version as an accurate update.<\/p>\n <p><strong>EXAMPLE:<\/strong> If a claim \u201cPolicy X caused unemployment to rise\u201d is unproven, the corrected summary might say: \u201cUnemployment rose from A to B between dates; economists cite multiple contributing factors (inflation, interest rates); no direct causal evidence links the change to Policy X.\u201d This avoids leaving readers with \u201ceverything is wrong\u201d confusion. In corporate settings, corrected summaries reduce rework because teams can copy verified text into reports without repeatedly re-checking the same claims.<\/p>\n\n <h3>6. Open Questions Lists Create Responsible Uncertainty and Next-Step Research<\/h3>\n <p><strong>WHY IT WORKS:<\/strong> Some claims are not currently verifiable (ongoing investigations, unpublished data, conflicting sources). Listing open questions clarifies what is unknown and what evidence would resolve it. This prevents speculation and guides further research. It also improves transparency: stakeholders can see limitations rather than assuming completeness. In high-stakes environments, this protects credibility because you don\u2019t overpromise certainty.<\/p>\n <p><strong>EXAMPLE:<\/strong> In a breaking news event, casualty numbers vary. The correct approach: verify what is confirmed, list discrepancies, and specify what would resolve uncertainty (official updates, hospital reports). In product claims, list what needs lab testing or third-party certification. Teams that explicitly track open questions avoid reintroducing rumors later and create a clear research backlog. This also helps governance: unresolved high-impact claims can trigger \u201cdo not publish\u201d or \u201clabel as unverified\u201d policies until resolved.<\/p>\n\n <div class=\"section-title-container\"><h2 class=\"section-title\">Example Output Preview<\/h2><\/div>\n\n <div class=\"example-output\">\n <h4>Sample: Fact-Checking a Viral Post About a \u201cNew Tax Law\u201d<\/h4>\n <p><strong>Claim List (Excerpt):<\/strong> (1) \u201cA new federal law imposes a 5% tax on all digital transactions starting July 1.\u201d (2) \u201cCongress passed it unanimously.\u201d (3) \u201cIt applies to Venmo, PayPal, and credit cards.\u201d (4) \u201cIt will raise $200B per year.\u201d<\/p>\n <p><strong>Verification Results (Excerpt):<\/strong> Claim (1): UNPROVEN\u2014no bill number or official text provided; no primary source identified. Claim (2): FALSE\u2014unanimous passage not supported by congressional records. Claim (3): MISLEADING\u2014transaction reporting rules exist, but not a flat \u201c5% tax.\u201d Claim (4): UNPROVEN\u2014no budget analysis provided; figure inconsistent with baseline digital payment volume.<\/p>\n <p><strong>Corrected Summary:<\/strong> No verified evidence supports a new 5% federal tax on all digital transactions starting July 1. Existing regulations may require reporting for certain transaction types, but that is different from a tax. Readers should verify using official congressional records and bill text before sharing.<\/p>\n <p><strong>Open Questions:<\/strong> If the claim references a specific bill, what is the bill number? What official budget scoring exists? Which agencies are involved?<\/p>\n <\/div>\n\n <div class=\"section-title-container\"><h2 class=\"section-title\">Prompt Chain Strategy<\/h2><\/div>\n\n <div class=\"chain-step\">\n <h4>Step 1: Structured Fact-Check Report<\/h4>\n <p><strong>Prompt:<\/strong> Use the main Fact-Checking Prompt with the content to verify.<\/p>\n <p><strong>Expected Output:<\/strong> A claim-based verification report with verdicts, evidence quotes, corrected summary, and open questions.<\/p>\n <\/div>\n\n <div class=\"chain-step\">\n <h4>Step 2: Expand Evidence Collection for Priority Claims<\/h4>\n <p><strong>Prompt:<\/strong> \u201cFor the top 5 priority claims, propose a deeper evidence plan: primary sources to retrieve, search queries, who to contact, and how to validate. Then produce a revised verification table with stronger evidence.\u201d<\/p>\n <p><strong>Expected Output:<\/strong> A deeper research plan and improved confidence levels for high-impact claims.<\/p>\n <\/div>\n\n <div class=\"chain-step\">\n <h4>Step 3: Build a Publishable \u201cCorrection + Context\u201d Article<\/h4>\n <p><strong>Prompt:<\/strong> \u201cTurn the corrected summary into a publishable article: explain what\u2019s true, what\u2019s false, why confusion happened, and how readers can verify. Include a checklist for spotting misinformation.\u201d<\/p>\n <p><strong>Expected Output:<\/strong> A reader-friendly correction piece that can be shared publicly to counter misinformation.<\/p>\n <\/div>\n\n <div class=\"section-title-container\"><h2 class=\"section-title\">Human-in-the-Loop Refinements<\/h2><\/div>\n\n <h3>Maintain a \u201cSource Reliability\u201d Rubric for Your Organization<\/h3>\n <p>Create a rubric: primary documents (highest), peer-reviewed research, official data, reputable journalism, commentary, social media (lowest). <strong>Technique:<\/strong> require at least one primary source for high-stakes claims.<\/p>\n\n <h3>Build a Claims Database to Avoid Re-Fact-Checking the Same Meme<\/h3>\n <p>Store verified claims and citations so teams can respond quickly when the same misinformation returns. <strong>Technique:<\/strong> tag by topic and verdict for fast retrieval.<\/p>\n\n <h3>Use \u201cPrebunking\u201d for Predictable Misinfo Cycles<\/h3>\n <p>For recurring topics, publish verification guides ahead of time. <strong>Technique:<\/strong> anticipate claims and prepare primary sources and explainer templates.<\/p>\n\n <h3>Require Reviewer Sign-Off for High-Risk Topics<\/h3>\n <p>Health, finance, and legal content should require a subject matter reviewer. <strong>Technique:<\/strong> define SLA (e.g., 24h) and escalation if reviewer unavailable.<\/p>\n\n <h3>Track Corrections and Measure \u201cTime to Correction\u201d<\/h3>\n <p>Measure how quickly misinformation is corrected after detection. <strong>Technique:<\/strong> set a target (e.g., <48h) and run retrospectives on misses.<\/p>\n\n <h3>Teach Writers to Cite as They Write<\/h3>\n <p>Most verification pain comes from missing citations. <strong>Technique:<\/strong> require citations per paragraph during drafting, not after.<\/p>\n\n <div class=\"footer\">\n <div class=\"footer-stat\"><div class=\"footer-stat-value\">4.8\u2605<\/div><div class=\"footer-stat-label\">Average Rating<\/div><\/div>\n <div class=\"footer-stat\"><div class=\"footer-stat-value\">2,304<\/div><div class=\"footer-stat-label\">Times Copied<\/div><\/div>\n <div class=\"footer-stat\"><div class=\"footer-stat-value\">191<\/div><div class=\"footer-stat-label\">Reviews<\/div><\/div>\n <\/div>\n <\/div>\n <\/div>\n <\/div>\n\n <script>\n function copyPrompt(){\n const promptContent=document.getElementById('promptContent').innerText;\n navigator.clipboard.writeText(promptContent).then(()=>{\n const button=document.querySelector('.copy-button');\n const originalText=button.innerHTML;\n button.innerHTML='\u2713 Copied!';\n setTimeout(()=>{button.innerHTML=originalText;},2000);\n }).catch(err=>console.error('Failed to copy text: ',err));\n }\n <\/script>\n<\/body>\n<\/html>\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<\/div>","protected":false},"excerpt":{"rendered":"<p>Fact-Checking Prompt – AiPro Institute\u2122 Fact-Checking Prompt Fact-Checking Prompt AI Safety & Governance \u23f1\ufe0f 20-30 minutes \ud83d\udcca Intermediate ChatGPT Claude Gemini Perplexity Grok The Prompt \ud83d\udccb Copy Prompt You are a rigorous fact-checker and verification analyst. Fact-check the following content and produce an evidence-based verification report. [CONTENT_TO_FACT_CHECK] (paste article, post, transcript, or claims) [TOPIC_DOMAIN] (e.g.,…<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[170],"tags":[],"class_list":["post-5621","post","type-post","status-publish","format-standard","hentry","category-ai-safety-governance"],"acf":[],"_links":{"self":[{"href":"https:\/\/teen.aiproinstitute.com\/zh\/wp-json\/wp\/v2\/posts\/5621","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/teen.aiproinstitute.com\/zh\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/teen.aiproinstitute.com\/zh\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/teen.aiproinstitute.com\/zh\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/teen.aiproinstitute.com\/zh\/wp-json\/wp\/v2\/comments?post=5621"}],"version-history":[{"count":4,"href":"https:\/\/teen.aiproinstitute.com\/zh\/wp-json\/wp\/v2\/posts\/5621\/revisions"}],"predecessor-version":[{"id":5645,"href":"https:\/\/teen.aiproinstitute.com\/zh\/wp-json\/wp\/v2\/posts\/5621\/revisions\/5645"}],"wp:attachment":[{"href":"https:\/\/teen.aiproinstitute.com\/zh\/wp-json\/wp\/v2\/media?parent=5621"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/teen.aiproinstitute.com\/zh\/wp-json\/wp\/v2\/categories?post=5621"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/teen.aiproinstitute.com\/zh\/wp-json\/wp\/v2\/tags?post=5621"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}

\n\t\t\t\t\t\t

\n\t\t\t\t\t

\n\t\t\t

\n\t\t\t\t\t\t

\n\t\t\t\t\t\n\n\n \n \n Fact-Checking Prompt - AiPro Institute\u2122<\/title>\n <style>\n *{margin:0;padding:0;box-sizing:border-box}\n body{font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,'Helvetica Neue',Arial,sans-serif;line-height:1.6;color:#333;background:#fff;padding:2rem 1rem}\n .container{max-width:900px;margin:0 auto}\n .page-title{text-align:center;font-size:2.5rem;font-weight:700;background:linear-gradient(135deg,#667eea 0%,#764ba2 100%);-webkit-background-clip:text;-webkit-text-fill-color:transparent;background-clip:text;margin-bottom:2rem}\n .card{background:#fff;border-radius:12px;box-shadow:0 4px 6px rgba(0,0,0,.1);overflow:hidden;margin-bottom:2rem}\n .card-header{background:linear-gradient(135deg,#667eea 0%,#764ba2 100%);color:#fff;padding:2rem}\n .card-header h1{font-size:2rem;margin-bottom:.5rem}\n .card-header .subtitle{font-size:1.1rem;opacity:.95}\n .meta-badges,.tool-badges{display:flex;gap:.75rem;margin-top:1rem;flex-wrap:wrap}\n .badge{background:rgba(255,255,255,.2);padding:.4rem .9rem;border-radius:20px;font-size:.9rem;backdrop-filter:blur(10px)}\n .tool-badge{background:transparent;border:1px solid rgba(255,255,255,.4);padding:.4rem .9rem;border-radius:20px;font-size:.85rem}\n .card-body{padding:2.5rem}\n .section-title-container{display:flex;justify-content:space-between;align-items:center;margin:2.5rem 0 1.25rem 0}\n .section-title-container:first-child{margin-top:0}\n .section-title{font-size:1.75rem;color:#764ba2;border-left:4px solid #764ba2;padding-left:1rem;margin:0}\n .copy-button{background:linear-gradient(135deg,#667eea 0%,#764ba2 100%);color:#fff;border:none;padding:.6rem 1.5rem;border-radius:6px;cursor:pointer;font-size:.95rem;font-weight:500;transition:opacity .3s}\n .copy-button:hover{opacity:.9}\n .prompt-box{background:#f8f9fa;border:1px solid #dee2e6;border-radius:8px;padding:1.5rem;margin:1.25rem 0;font-family:'Courier New',monospace;font-size:.95rem;line-height:1.6;white-space:pre-wrap;overflow-x:auto}\n .placeholder{color:#fd7e14;font-weight:bold}\n .tip-box{background:#fff9e6;border-left:4px solid #ffc107;padding:1.25rem;margin:1.25rem 0;border-radius:4px}\n .tip-box strong{color:#f57c00}\n h3{color:#764ba2;font-size:1.35rem;margin:2rem 0 1rem 0}\n p{margin-bottom:1rem;line-height:1.8}\n ul,ol{margin-left:2rem;margin-bottom:1rem}\n li{margin-bottom:.5rem;line-height:1.8}\n .example-output{background:#f0f8ff;border:2px solid #4a90e2;border-radius:8px;padding:1.5rem;margin:1.25rem 0}\n .example-output h4{color:#4a90e2;margin-bottom:1rem}\n .chain-step{background:#f8f9fa;border-left:4px solid #667eea;padding:1.5rem;margin:1.5rem 0;border-radius:4px}\n .chain-step h4{color:#667eea;margin-bottom:.75rem}\n .footer{background:#f8f9fa;padding:2rem;margin-top:2rem;border-radius:8px;display:flex;justify-content:space-around;align-items:center;flex-wrap:wrap;gap:1.5rem}\n .footer-stat{text-align:center}\n .footer-stat-value{font-size:1.75rem;font-weight:700;color:#764ba2}\n .footer-stat-label{color:#666;font-size:.95rem}\n @media (max-width:768px){.page-title{font-size:1.75rem}.card-header h1{font-size:1.5rem}.card-body{padding:1.5rem}.section-title{font-size:1.35rem}.section-title-container{flex-direction:column;align-items:flex-start;gap:1rem}.footer{flex-direction:column}}\n <\/style>\n<\/head>\n<body>\n <div class=\"container\">\n <h1 class=\"page-title\">Fact-Checking Prompt<\/h1>\n\n <div class=\"card\">\n <div class=\"card-header\">\n <h1>Fact-Checking Prompt<\/h1>\n <p class=\"subtitle\">AI Safety & Governance<\/p>\n <div class=\"meta-badges\">\n <span class=\"badge\">\u23f1\ufe0f 20-30 minutes<\/span>\n <span class=\"badge\">\ud83d\udcca Intermediate<\/span>\n <\/div>\n <div class=\"tool-badges\">\n <span class=\"tool-badge\">ChatGPT<\/span>\n <span class=\"tool-badge\">Claude<\/span>\n <span class=\"tool-badge\">Gemini<\/span>\n <span class=\"tool-badge\">Perplexity<\/span>\n <span class=\"tool-badge\">Grok<\/span>\n <\/div>\n <\/div>\n\n <div class=\"card-body\">\n <div class=\"section-title-container\">\n <h2 class=\"section-title\">The Prompt<\/h2>\n <button class=\"copy-button\" onclick=\"copyPrompt()\">\ud83d\udccb Copy Prompt<\/button>\n <\/div>\n\n <div class=\"prompt-box\" id=\"promptContent\">You are a rigorous fact-checker and verification analyst. Fact-check the following content and produce an evidence-based verification report.\n\n<span class=\"placeholder\">[CONTENT_TO_FACT_CHECK]<\/span> (paste article, post, transcript, or claims)\n\n<span class=\"placeholder\">[TOPIC_DOMAIN]<\/span> (e.g., \"health\", \"finance\", \"politics\", \"science\", \"law\", \"product claims\")\n\n<span class=\"placeholder\">[REQUIRED_RIGOR]<\/span> (e.g., \"journalistic\", \"academic\", \"legal-grade\")\n\n<span class=\"placeholder\">[SOURCE_CONSTRAINTS]<\/span> (e.g., \"must use primary sources\", \"web sources allowed\", \"only internal docs\")\n\nUse the V.E.R.I.F.Y. FRAMEWORK:\n\n**V - Verify Claims List** \u2192 Extract all factual claims and turn them into checkable statements\n**E - Evidence Collection** \u2192 Identify what evidence is needed and where to find it\n**R - Reliability Scoring** \u2192 Rate sources (primary, secondary, partisan, anonymous, outdated)\n**I - Inconsistency Detection** \u2192 Spot contradictions, missing context, misleading framing\n**F - Findings & Verdicts** \u2192 Label each claim: True\/False\/Misleading\/Unproven\n**Y - Yield a Corrected Summary** \u2192 Provide corrected version with citations and uncertainty notes\n\nDELIVER 10 COMPONENTS:\n\n\u2713 1. Claim Extraction (list 10-30 claims)\n\u2713 2. Claim Prioritization (top 5 high-impact claims)\n\u2713 3. Evidence Needed (per claim)\n\u2713 4. Source Plan (primary vs secondary sources)\n\u2713 5. Verification Results (per claim: verdict + evidence)\n\u2713 6. Confidence Level (high\/medium\/low)\n\u2713 7. Missing Context & Misleading Framing\n\u2713 8. Corrected Summary (rewrite with verified facts only)\n\u2713 9. Open Questions (what remains unknown)\n\u2713 10. Appendix (sources used + quotes)\n\nOUTPUT FORMAT:\n\n## Claim List\n[numbered claims]\n\n## Priority Claims\n[top 5]\n\n## Verification Table\n[table: claim \u2192 verdict \u2192 confidence \u2192 evidence (quotes + citations) \u2192 notes]\n\n## Missing Context \/ Misleading Framing\n\n## Corrected Summary (Verified Only)\n\n## Open Questions\n\n## Sources & Quotes Appendix\n\nConstraints:\n- If you cannot verify a claim, mark it UNPROVEN (do not guess)\n- Distinguish \u201cno evidence found\u201d from \u201cevidence shows false\u201d\n- Quote sources directly for key claims\n- Use clear verdict labels and confidence levels\n<\/div>\n\n <div class=\"tip-box\"><strong>\ud83d\udca1 Pro Tip:<\/strong> Treat every number, date, and named entity as a separate checkable claim. Most misinformation hides in \u201csmall specifics\u201d rather than the main narrative.<\/div>\n\n <div class=\"section-title-container\"><h2 class=\"section-title\">The Logic<\/h2><\/div>\n\n <h3>1. Claim Extraction Prevents \u201cVibes Checking\u201d and Forces Measurable Verification<\/h3>\n <p><strong>WHY IT WORKS:<\/strong> Many so-called fact checks evaluate an article\u2019s overall tone or political alignment rather than verifying specific claims. Breaking content into discrete, checkable statements converts an amorphous task (\u201cis this true?\u201d) into a structured verification workflow. This also prevents the \u201ctruthy narrative\u201d problem: a story can feel plausible while containing several wrong specifics. Claim extraction ensures each number, date, entity, and causal statement is tested. It also improves transparency: readers can see exactly what was checked and what wasn\u2019t. In professional workflows, claim lists are the backbone of verification because they create auditability and reproducibility.<\/p>\n <p><strong>EXAMPLE:<\/strong> A post says \u201cCountry X cut emissions 40% since 2010, saving $2B in healthcare costs, according to a UN report.\u201d That\u2019s at least three claims: (1) emissions change magnitude and baseline, (2) healthcare savings amount, (3) source attribution (\u201cUN report\u201d). A claim-based fact check will verify each separately, often finding that one is accurate while another is overstated. When teams adopt claim extraction, they reduce error rates in published summaries because they stop repeating unverified numbers. It also helps prioritize: a wrong \u201c$2B\u201d claim might matter more than a minor date error.<\/p>\n\n <h3>2. Evidence Hierarchies Reduce Misinformation Amplification by Favoring Primary Sources<\/h3>\n <p><strong>WHY IT WORKS:<\/strong> The internet is full of circular reporting: one article cites another which cites a tweet. An evidence hierarchy prioritizes primary sources (official data, filings, peer-reviewed papers, direct transcripts) over secondary commentary. This reduces the risk of amplifying unverified claims. Reliability scoring forces explicit judgment about source quality: primary vs. secondary, conflict of interest, outdatedness, and methodology transparency. This makes the fact check defensible and reduces \u201cappeal to authority\u201d mistakes where a reputable outlet is treated as proof without confirming its underlying sources.<\/p>\n <p><strong>EXAMPLE:<\/strong> Financial claim: \u201cCompany Y revenue grew 50% last quarter.\u201d The correct primary source is the company\u2019s earnings report (10-Q), not a blog summary. By forcing a primary-source-first plan, you avoid errors where a blog confuses revenue with bookings. In health claims, peer-reviewed studies and government datasets (CDC, WHO) outrank influencer posts. When the evidence hierarchy is explicit, you can also mark uncertainty: \u201cOnly secondary sources exist; claim remains unproven.\u201d This prevents overconfident conclusions.<\/p>\n\n <h3>3. Verdict Labels Separate Unknown From False, Preventing Overconfident Claims<\/h3>\n <p><strong>WHY IT WORKS:<\/strong> Many fact checks wrongly equate \u201cwe couldn\u2019t verify it\u201d with \u201cit\u2019s false,\u201d or they assume \u201csounds false\u201d without evidence. A disciplined system uses verdict labels: True, False, Misleading (partly true but missing context), Unproven (insufficient evidence), and Unclear (ambiguous claim). This reduces unjustified certainty. It also helps users: knowing something is unproven suggests caution, not dismissal. In governance contexts, this precision is essential because decisions can depend on evidence quality.<\/p>\n <p><strong>EXAMPLE:<\/strong> A claim: \u201cA new law bans encryption.\u201d If evidence is unclear because the bill is proposed, not passed, the right label is Misleading or Unproven depending on context. Another claim: \u201cStudy proves coffee causes cancer.\u201d The evidence might show correlation in one cohort but not causation; verdict: Misleading. By using clear labels, you avoid sensationalism and maintain credibility. This also supports automation: systems can route \u201cUnproven\u201d items to further research and treat \u201cFalse\u201d items differently (e.g., correction required).<\/p>\n\n <h3>4. Inconsistency Detection Catches Framing Tricks That Avoid Direct Lies<\/h3>\n <p><strong>WHY IT WORKS:<\/strong> Much misinformation is not outright falsehood but misleading framing: cherry-picked time windows, base rate neglect, omitted denominators, correlation-causation confusion, and quote mining. An inconsistency module looks for internal contradictions (numbers that don\u2019t add up), missing context (per-capita vs total), and rhetorical sleights (anonymous sourcing presented as fact). This is crucial because you can \u201cfact check\u201d individual sentences as true while the overall implication is misleading. Inconsistency detection reveals the gap between literal truth and honest interpretation.<\/p>\n <p><strong>EXAMPLE:<\/strong> \u201cCrime increased 20%\u201d may be true for one month vs previous month, but misleading if year-over-year crime decreased. Or \u201cvaccine adverse events doubled\u201d may be true due to reporting changes, not actual harm. In financial reporting, \u201cprofits fell 30%\u201d might omit that profits were unusually high last year (base effect). Inconsistency detection forces the report to include denominators, time frames, and alternative baselines. This reduces manipulation and improves reader understanding.<\/p>\n\n <h3>5. Corrected Summaries Prevent the \u201cDebunk Without Replacement\u201d Problem<\/h3>\n <p><strong>WHY IT WORKS:<\/strong> Simply labeling a claim false leaves a vacuum; readers often retain the original misinformation due to repetition and lack of an alternative coherent narrative. A corrected summary provides a replacement story using verified facts only, with uncertainty notes. This supports better memory encoding: people remember the corrected explanation rather than the debunk label. It also enables downstream reuse: journalists, analysts, and teams can publish the corrected version as an accurate update.<\/p>\n <p><strong>EXAMPLE:<\/strong> If a claim \u201cPolicy X caused unemployment to rise\u201d is unproven, the corrected summary might say: \u201cUnemployment rose from A to B between dates; economists cite multiple contributing factors (inflation, interest rates); no direct causal evidence links the change to Policy X.\u201d This avoids leaving readers with \u201ceverything is wrong\u201d confusion. In corporate settings, corrected summaries reduce rework because teams can copy verified text into reports without repeatedly re-checking the same claims.<\/p>\n\n <h3>6. Open Questions Lists Create Responsible Uncertainty and Next-Step Research<\/h3>\n <p><strong>WHY IT WORKS:<\/strong> Some claims are not currently verifiable (ongoing investigations, unpublished data, conflicting sources). Listing open questions clarifies what is unknown and what evidence would resolve it. This prevents speculation and guides further research. It also improves transparency: stakeholders can see limitations rather than assuming completeness. In high-stakes environments, this protects credibility because you don\u2019t overpromise certainty.<\/p>\n <p><strong>EXAMPLE:<\/strong> In a breaking news event, casualty numbers vary. The correct approach: verify what is confirmed, list discrepancies, and specify what would resolve uncertainty (official updates, hospital reports). In product claims, list what needs lab testing or third-party certification. Teams that explicitly track open questions avoid reintroducing rumors later and create a clear research backlog. This also helps governance: unresolved high-impact claims can trigger \u201cdo not publish\u201d or \u201clabel as unverified\u201d policies until resolved.<\/p>\n\n <div class=\"section-title-container\"><h2 class=\"section-title\">Example Output Preview<\/h2><\/div>\n\n <div class=\"example-output\">\n <h4>Sample: Fact-Checking a Viral Post About a \u201cNew Tax Law\u201d<\/h4>\n <p><strong>Claim List (Excerpt):<\/strong> (1) \u201cA new federal law imposes a 5% tax on all digital transactions starting July 1.\u201d (2) \u201cCongress passed it unanimously.\u201d (3) \u201cIt applies to Venmo, PayPal, and credit cards.\u201d (4) \u201cIt will raise $200B per year.\u201d<\/p>\n <p><strong>Verification Results (Excerpt):<\/strong> Claim (1): UNPROVEN\u2014no bill number or official text provided; no primary source identified. Claim (2): FALSE\u2014unanimous passage not supported by congressional records. Claim (3): MISLEADING\u2014transaction reporting rules exist, but not a flat \u201c5% tax.\u201d Claim (4): UNPROVEN\u2014no budget analysis provided; figure inconsistent with baseline digital payment volume.<\/p>\n <p><strong>Corrected Summary:<\/strong> No verified evidence supports a new 5% federal tax on all digital transactions starting July 1. Existing regulations may require reporting for certain transaction types, but that is different from a tax. Readers should verify using official congressional records and bill text before sharing.<\/p>\n <p><strong>Open Questions:<\/strong> If the claim references a specific bill, what is the bill number? What official budget scoring exists? Which agencies are involved?<\/p>\n <\/div>\n\n <div class=\"section-title-container\"><h2 class=\"section-title\">Prompt Chain Strategy<\/h2><\/div>\n\n <div class=\"chain-step\">\n <h4>Step 1: Structured Fact-Check Report<\/h4>\n <p><strong>Prompt:<\/strong> Use the main Fact-Checking Prompt with the content to verify.<\/p>\n <p><strong>Expected Output:<\/strong> A claim-based verification report with verdicts, evidence quotes, corrected summary, and open questions.<\/p>\n <\/div>\n\n <div class=\"chain-step\">\n <h4>Step 2: Expand Evidence Collection for Priority Claims<\/h4>\n <p><strong>Prompt:<\/strong> \u201cFor the top 5 priority claims, propose a deeper evidence plan: primary sources to retrieve, search queries, who to contact, and how to validate. Then produce a revised verification table with stronger evidence.\u201d<\/p>\n <p><strong>Expected Output:<\/strong> A deeper research plan and improved confidence levels for high-impact claims.<\/p>\n <\/div>\n\n <div class=\"chain-step\">\n <h4>Step 3: Build a Publishable \u201cCorrection + Context\u201d Article<\/h4>\n <p><strong>Prompt:<\/strong> \u201cTurn the corrected summary into a publishable article: explain what\u2019s true, what\u2019s false, why confusion happened, and how readers can verify. Include a checklist for spotting misinformation.\u201d<\/p>\n <p><strong>Expected Output:<\/strong> A reader-friendly correction piece that can be shared publicly to counter misinformation.<\/p>\n <\/div>\n\n <div class=\"section-title-container\"><h2 class=\"section-title\">Human-in-the-Loop Refinements<\/h2><\/div>\n\n <h3>Maintain a \u201cSource Reliability\u201d Rubric for Your Organization<\/h3>\n <p>Create a rubric: primary documents (highest), peer-reviewed research, official data, reputable journalism, commentary, social media (lowest). <strong>Technique:<\/strong> require at least one primary source for high-stakes claims.<\/p>\n\n <h3>Build a Claims Database to Avoid Re-Fact-Checking the Same Meme<\/h3>\n <p>Store verified claims and citations so teams can respond quickly when the same misinformation returns. <strong>Technique:<\/strong> tag by topic and verdict for fast retrieval.<\/p>\n\n <h3>Use \u201cPrebunking\u201d for Predictable Misinfo Cycles<\/h3>\n <p>For recurring topics, publish verification guides ahead of time. <strong>Technique:<\/strong> anticipate claims and prepare primary sources and explainer templates.<\/p>\n\n <h3>Require Reviewer Sign-Off for High-Risk Topics<\/h3>\n <p>Health, finance, and legal content should require a subject matter reviewer. <strong>Technique:<\/strong> define SLA (e.g., 24h) and escalation if reviewer unavailable.<\/p>\n\n <h3>Track Corrections and Measure \u201cTime to Correction\u201d<\/h3>\n <p>Measure how quickly misinformation is corrected after detection. <strong>Technique:<\/strong> set a target (e.g., <48h) and run retrospectives on misses.<\/p>\n\n <h3>Teach Writers to Cite as They Write<\/h3>\n <p>Most verification pain comes from missing citations. <strong>Technique:<\/strong> require citations per paragraph during drafting, not after.<\/p>\n\n <div class=\"footer\">\n <div class=\"footer-stat\"><div class=\"footer-stat-value\">4.8\u2605<\/div><div class=\"footer-stat-label\">Average Rating<\/div><\/div>\n <div class=\"footer-stat\"><div class=\"footer-stat-value\">2,304<\/div><div class=\"footer-stat-label\">Times Copied<\/div><\/div>\n <div class=\"footer-stat\"><div class=\"footer-stat-value\">191<\/div><div class=\"footer-stat-label\">Reviews<\/div><\/div>\n <\/div>\n <\/div>\n <\/div>\n <\/div>\n\n <script>\n function copyPrompt(){\n const promptContent=document.getElementById('promptContent').innerText;\n navigator.clipboard.writeText(promptContent).then(()=>{\n const button=document.querySelector('.copy-button');\n const originalText=button.innerHTML;\n button.innerHTML='\u2713 Copied!';\n setTimeout(()=>{button.innerHTML=originalText;},2000);\n }).catch(err=>console.error('Failed to copy text: ',err));\n }\n <\/script>\n<\/body>\n<\/html>\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<\/div>","protected":false},"excerpt":{"rendered":"<p>Fact-Checking Prompt – AiPro Institute\u2122 Fact-Checking Prompt Fact-Checking Prompt AI Safety & Governance \u23f1\ufe0f 20-30 minutes \ud83d\udcca Intermediate ChatGPT Claude Gemini Perplexity Grok The Prompt \ud83d\udccb Copy Prompt You are a rigorous fact-checker and verification analyst. Fact-check the following content and produce an evidence-based verification report. [CONTENT_TO_FACT_CHECK] (paste article, post, transcript, or claims) [TOPIC_DOMAIN] (e.g.,…<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[170],"tags":[],"class_list":["post-5621","post","type-post","status-publish","format-standard","hentry","category-ai-safety-governance"],"acf":[],"_links":{"self":[{"href":"https:\/\/teen.aiproinstitute.com\/zh\/wp-json\/wp\/v2\/posts\/5621","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/teen.aiproinstitute.com\/zh\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/teen.aiproinstitute.com\/zh\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/teen.aiproinstitute.com\/zh\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/teen.aiproinstitute.com\/zh\/wp-json\/wp\/v2\/comments?post=5621"}],"version-history":[{"count":4,"href":"https:\/\/teen.aiproinstitute.com\/zh\/wp-json\/wp\/v2\/posts\/5621\/revisions"}],"predecessor-version":[{"id":5645,"href":"https:\/\/teen.aiproinstitute.com\/zh\/wp-json\/wp\/v2\/posts\/5621\/revisions\/5645"}],"wp:attachment":[{"href":"https:\/\/teen.aiproinstitute.com\/zh\/wp-json\/wp\/v2\/media?parent=5621"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/teen.aiproinstitute.com\/zh\/wp-json\/wp\/v2\/categories?post=5621"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/teen.aiproinstitute.com\/zh\/wp-json\/wp\/v2\/tags?post=5621"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}