Fact-Checking Skill

Do NOT trigger for:

• General research or information gathering (use web search instead)
• Checking grammar, spelling, or writing quality
• Verifying code functionality or technical documentation
• Questions about opinions rather than factual claims

Workflow

Step 1: Understand the Content

Before beginning verification, analyze what needs to be checked:

1. Identify specific claims - Extract concrete, verifiable statements from the content
2. Note the context - Identify:
   • Geographic references (countries, regions, cities)
   • Named individuals (politicians, public figures, organizations)
   • Languages used in the content
   • Time period or dates mentioned
   • Subject matter (politics, health, science, etc.)
3. Determine user context:
   • User's native language (for selecting appropriate fact-checkers)
   • User's location if relevant

Example Analysis:

• Content: "Video claiming vaccines cause autism, mentions Andrew Wakefield, references UK study"
• Claims to verify: Vaccine-autism link, Wakefield's research
• Context: Medical/health topic, UK origin, English language
• Key entities: Andrew Wakefield, MMR vaccine, UK medical establishment

Step 2: Select Fact-Checking Services

CRITICAL: Begin by fetching the current list of fact-checking services:

Fetch: https://en.wikipedia.org/wiki/List_of_fact-checking_websites

From this list, select 3-7 relevant fact-checking services based on:

Selection Criteria

1. User's language/location - Always include fact-checkers in the user's native language

2. Content language/location - If different from user's language, also include fact-checkers in the content's language and region

3. Geographic relevance - If content mentions specific countries/regions:

   • Include fact-checkers from those countries
   • Example: Content about French politics → include French fact-checkers

4. Subject matter specialists - Some fact-checkers specialize:

   • Health/medical claims → Health Feedback, Science Feedback
   • Politics → country-specific political fact-checkers
   • General → Snopes, FactCheck.org, Full Fact

5. Person-specific - If content focuses on specific public figures:

   • Include fact-checkers from their home countries
   • Example: Claims about a US politician → include US fact-checkers

Exclusion Rule

NEVER use services listed under "Fraudulent fact-checking websites" on the Wikipedia page, regardless of how well they match other criteria.

Prioritization

When you must limit selections:

• Prioritize: User's language > Content's language > Geographic relevance
• Prefer well-established services (FactCheck.org, Snopes, Full Fact, AFP Fact Check, etc.)
• Include at least one international/general service

Example Selection:

• User: Polish speaker
• Content: English article about US vaccines
• Selected services:
  1. Demagog.pl (Polish, for user)
  2. FactCheck.org (US, for content geography)
  3. Snopes (US, general/medical)
  4. Health Feedback (health specialist)
  5. Full Fact (UK, English-speaking, general)

Step 3: Search Each Fact-Checking Service

For each selected service, conduct targeted searches:

Search Strategy

1. Extract 2-4 search terms from the content:

   • Key person names
   • Main topics/subjects
   • Specific claims or events
   • Important keywords

2. Translate terms to the fact-checker's native language if needed

3. Construct search queries using DuckDuckGo with site operator:

   Format: site:domain.com [search terms in appropriate language]
   
   Examples:
   - site:fullfact.org vaccines autism
   - site:demagog.org.pl szczepionki autyzm
   - site:factcheck.org Andrew Wakefield MMR
   - site:healthfeedback.org vaccine safety
   

4. Execute 1-3 searches per fact-checker (depending on content complexity)

Search Best Practices

• Keep queries concise (2-4 words typically)
• Start broad, then narrow if needed
• Don't repeat very similar queries
• If first search yields good results, proceed to analysis
• If first search yields poor results, try alternative terms

Step 4: Analyze Search Results

For each fact-checking service:

1. Review search results - Examine the first 5-10 results from each search

2. Select relevant articles - Choose articles where:

   • Headline directly addresses the claim being verified
   • Content appears substantial (not just brief mentions)
   • Publication date is relevant (recent for ongoing issues, any date for historical debunks)

3. Fetch and read articles - Use web_fetch to retrieve the full text of 2-4 most relevant articles per fact-checker

4. Extract key findings for each article:

   • Verdict - What did the fact-checker conclude? (True, False, Misleading, Mixed, Unproven, etc.)
   • Evidence - What evidence did they cite?
   • Context - Any important nuance or context
   • Relevance - How directly does this address the user's claim?

Step 5: Synthesize and Present Results

Organize findings into a clear, user-friendly format:

Handle Fresh Content First

Before presenting results, check if the content is very recent (3 days old or less):

1. If fact-checks found: Proceed normally with presentation

2. If no fact-checks found AND content is ≤3 days old:

   • Note that the content is too fresh for fact-checkers to have covered it yet
   • If task scheduling is available:
     • Schedule a follow-up fact-check for 3 days from now
     • Inform user: "I've scheduled a follow-up check for [date]. I'll notify you if fact-checkers have published verification by then."
   • If task scheduling is NOT available:
     • Suggest: "This content is very recent (published [date]). Fact-checkers typically need a few days to verify claims. I recommend checking back in 3 days for updated verification."
   • Offer preliminary analysis using general web search
   • Proceed with any available information from general sources

3. If no fact-checks found AND content is older:

   • Note that fact-checkers haven't specifically covered this
   • Offer general web research instead

Structure Your Response

1. Opening summary (2-3 sentences)

   • Overall consensus from fact-checkers
   • Brief answer to the user's question

2. Key findings by claim (if multiple claims)

   • Group related findings together
   • Present contradicting evidence if it exists

3. Detailed evidence (organized by fact-checker or by claim)

   • Include specific verdicts
   • Cite evidence fact-checkers used
   • Note any disagreements between fact-checkers

4. Important context (if relevant)

   • Historical background
   • Why the claim persists
   • Common misconceptions

5. Source citations

   • Provide direct links to all fact-checking articles referenced
   • Format: [Fact-Checker Name]: Article Title (Date if available) - [URL]

Presentation Guidelines

• Be objective - Present findings without inserting personal judgment
• Be nuanced - Avoid oversimplifying complex issues
• Be clear about uncertainty - If fact-checkers disagree or evidence is inconclusive, say so
• Be balanced - If some evidence supports and some contradicts, present both
• Use accessible language - Avoid jargon, explain technical terms
• Highlight consensus - When multiple fact-checkers agree, emphasize this

Formatting

• Use clear headers to organize different claims or themes
• Use natural prose, not bullet points, for the main findings
• Only use lists for: multiple similar items, source citations, or when explicitly helpful
• Include clickable citations throughout (not just at the end)

Example Response Structure

Based on verification from five established fact-checking organizations, the claim that vaccines cause autism has been thoroughly debunked. Multiple independent reviews of the evidence have found no causal link between vaccination and autism spectrum disorder.

The origins of this claim trace back to a fraudulent 1998 study by Andrew Wakefield, which was later retracted by The Lancet. Fact-checkers consistently note that Wakefield lost his medical license, and subsequent large-scale studies involving millions of children have found no connection.

[Full Fact reviewed the evidence in 2023](link), concluding "There is no link between the MMR vaccine and autism." Their analysis examined 12 major studies and found consistent results across different populations and methodologies.

[FactCheck.org's comprehensive analysis](link) explains that "The myth persists despite overwhelming scientific consensus against it" and details how the original study was not only retracted but shown to involve falsified data.

However, [Demagog.pl](link) notes that while the vaccine-autism link is false, concerns about vaccine safety in general are legitimate and should be addressed through proper scientific channels rather than dismissed.

**Important context**: The persistence of this myth has real public health consequences, as fact-checkers note declining vaccination rates in some communities. Understanding why the claim was debunked helps address ongoing concerns.

**Sources consulted:**
- Full Fact: "MMR vaccine does not cause autism" - [link]
- FactCheck.org: "Wakefield's Fraudulent Research" - [link]  
- Snopes: "Vaccines and Autism" - [link]
- Demagog.pl: "Szczepionki i autyzm - mit czy prawda?" - [link]
- Health Feedback: "Scientific consensus on vaccine safety" - [link]

Common Scenarios

Scenario 1: Single Specific Claim

User request: "Is it true that 5G causes COVID-19?"

Approach:

• Identify claim: 5G technology causes or spreads COVID-19
• Select 4-5 general fact-checkers (international scope, tech/health focus)
• Search for "5G COVID" or "5G coronavirus"
• Expected result: Multiple fact-checkers will have debunked this
• Present: Clear consensus with explanation of why the claim is false

Scenario 2: Article with Multiple Claims

User request: "Can you fact-check this article about climate change?"

Approach:

• Extract 3-5 specific verifiable claims from the article
• Select fact-checkers: user's language + climate-focused services
• Search each claim separately
• Present: Findings organized by claim, with overall assessment

Scenario 3: Complex Political Claim

User request: "Did [politician] really say/do [thing]?"

Approach:

• Identify the specific claim and context
• Select fact-checkers from politician's country + user's language
• Search politician's name + key terms
• Present: Direct answer with context, including if statement was taken out of context

Scenario 4: Viral Social Media Content

User request: "I saw this video on TikTok claiming [X], is it real?"

Approach:

• Identify what's being claimed in the video
• Select broad, well-known fact-checkers (viral content often fact-checked widely)
• Search for key terms from the claim
• Present: Whether it's been debunked, original context if misrepresented

Scenario 5: Historical Claim

User request: "Did [historical event] really happen this way?"

Approach:

• Note that this is historical verification, may need broader research
• Select fact-checkers + consider using general web search for historical records
• Present: What fact-checkers say if available, acknowledge if claim is outside typical fact-checking scope

Scenario 6: Very Fresh Content (Breaking News)

User request: "I just saw this article published today claiming [X]. Is it true?"

Approach:

• Check publication date: is it 3 days old or less?
• Search fact-checkers anyway (sometimes they work very quickly on major stories)
• If no fact-checks found:
  • With task scheduling: Schedule follow-up check for 3 days later, notify user of the scheduled check
  • Without task scheduling: Inform user that content is too fresh, suggest returning in 3 days
• Offer preliminary analysis using general web search
• Present: "This is very recent content. Fact-checkers haven't had time to verify yet. Here's what I found from general sources, but I recommend waiting for professional fact-checking."

Example response:

This article was published just [X hours/days] ago, which is too recent for professional 
fact-checkers to have verified the claims yet. They typically need a few days to conduct 
thorough research.

I've scheduled a follow-up fact-check for [date in 3 days]. I'll notify you automatically 
if fact-checkers publish verification by then.

In the meantime, here's what I found through general web research:
[preliminary findings with appropriate caveats]

Note: These are preliminary findings only. Professional fact-checkers may provide more 
thorough verification in the coming days.

Edge Cases and Limitations

When Fact-Checkers Haven't Covered the Topic

If searches return no relevant results:

1. Try broader search terms
2. Try related claims that fact-checkers may have covered
3. If still no results, check if the content is recent (3 days or less)
4. For fresh content (≤3 days old):
   • Acknowledge: "This is very recent content. Professional fact-checkers typically need a few days to verify claims."
   • If scheduling tools are available: Schedule a follow-up fact-check for 3 days later
   • If scheduling is not available: Suggest the user returns in 3 days for updated verification
   • Offer to do preliminary general web research in the meantime
5. For older content: Acknowledge "Professional fact-checkers haven't specifically addressed this claim"
6. Offer to do general web research instead
7. Consider if the claim is too obscure or too local for major fact-checkers

Contradicting Fact-Checkers

If fact-checkers disagree:

1. Present all perspectives fairly
2. Note the disagreement explicitly
3. Consider if they're addressing slightly different aspects
4. Look for consensus on specific sub-points
5. Don't force a conclusion if the evidence is genuinely mixed

Outdated Information

If fact-checks are old but the claim is current:

1. Note the publication dates
2. Search for more recent fact-checks
3. Consider if circumstances have changed
4. Acknowledge if using older sources due to lack of recent coverage

Language Barriers

If key fact-checkers are in languages you don't fully understand:

1. Use web_fetch to retrieve the content
2. Focus on verdicts, ratings, and conclusion sections which are often clear
3. Use any English summaries or abstracts
4. Acknowledge limitations if language creates uncertainty

Bias Concerns

Users may question fact-checker reliability:

1. Stick to well-established, internationally recognized services
2. Present findings from multiple fact-checkers to show consensus
3. Note if you're using fact-checkers from multiple countries/perspectives
4. Acknowledge that no source is perfect, but these are professional verification services

Quality Checklist

Before presenting results, verify:

• Checked at least 3 different fact-checking services
• Included fact-checkers relevant to the user's language/location
• Included fact-checkers relevant to the content's context
• Excluded any fraudulent fact-checking services
• Read full articles, not just headlines or snippets
• Provided direct links to all sources cited
• Presented findings objectively without adding personal judgment
• Acknowledged any uncertainty or disagreement between sources
• Organized response clearly with specific findings, not vague summaries
• Used natural prose for main findings, lists only where truly helpful
• If content is ≤3 days old with no fact-checks: Noted this and scheduled follow-up OR suggested user return in 3 days
• If providing preliminary analysis: Clearly distinguished it from professional fact-checking

Examples of Good Fact-Checking Services

International/English:

• FactCheck.org (US, general)
• Snopes (US, general)
• Full Fact (UK, general)
• AFP Fact Check (International, multilingual)
• PolitiFact (US, politics)

Regional/Language-Specific:

• Demagog.pl (Poland, Polish)
• Les Décodeurs (France, French)
• Correctiv (Germany, German)
• Maldita.es (Spain, Spanish)
• Aos Fatos (Brazil, Portuguese)
• Alt News (India, English/Hindi)
• Africa Check (Africa, multilingual)

Specialized:

• Health Feedback (health/medical claims)
• Climate Feedback (climate science claims)
• Science Feedback (general science claims)

Note: This is not exhaustive. Always fetch the current list from Wikipedia to see all available services.

Final Notes

Task Scheduling for Fresh Content

When content is very recent (≤3 days old) and hasn't been fact-checked yet:

If task scheduling tools are available:

• Automatically schedule a follow-up fact-check for 3 days later
• Store the original query, claims, and context
• When the scheduled task runs:
  • Re-search the same fact-checking services
  • Compare new findings to preliminary analysis
  • Notify user only if new fact-checks were found
  • Provide updated verification with links

If task scheduling is NOT available:

• Inform the user that the content is too fresh
• Suggest they return in 3 days for updated verification
• Provide preliminary analysis from general sources with appropriate caveats
• Make it clear that preliminary findings are not from professional fact-checkers

Core Approach

This skill focuses on using professional fact-checking organizations rather than doing original research. These organizations employ journalists and researchers who specialize in verification. Your role is to:

1. Find what they've already published
2. Synthesize their findings
3. Present them clearly to the user
4. Schedule follow-ups for very recent content when possible

If a topic hasn't been covered by fact-checkers, acknowledge this and offer to do general research instead. Don't try to replace professional fact-checking with web searches alone, but do provide preliminary information when users need it for fresh content.