AI Fact-Checking in Political Discussions

Manual vs Automatic Approaches

Why Fact-Checking Matters

  • Political discussions shape public opinion
  • Misinformation spreads quickly (especially online)
  • Trust in democracy depends on accurate information

Can You Decide What Is True?

  • Critical thinking takes effort and time
  • It requires trustworthy sources and the right method
  • Everyone is susceptible to cognitive biases

The SIFT Method

A practical framework for evaluating information:

  • Stop — pause before sharing or believing
  • Investigate the source — who is behind it?
  • Find better coverage — look for other reporting
  • Trace claims — go back to the original source

Related metacognitive framework: CRAAP (Currency, Relevance, Authority, Accuracy, Purpose)

Manual Fact-Checking

  • Performed by human experts
  • Uses verified, credible sources
  • Evaluates context, intent, and nuance

Fact Checking databases

Facts from the political discussions are created by volunteers and stored in the public repositories.

Strengths:

  • High accuracy
  • Deep understanding of context
  • Transparent reasoning and sourcing

Weaknesses:

  • Incomplete
  • Scattered
  • Difficult for non-specialists

Example: Fico on Russian State TV (Rossija 1)

  • Statement analyzed by Demagog.sk
  • Example of political messaging in foreign media

🔗 demagog.sk — Fico v ruskej štátnej televízii

  • Who said it? (authority)
  • What is the motivation (intent) of the speaker?
  • Is the message supported by evidence?

Example: EU Fact-Checking

Initiative of the European Journalism Training Association (EJTA)

EU Fact Check - A ban on new internal combustion engine cars achieves nothing for the climate

Automatic Fact-Checking (AI)

  • Uses machine learning and large language models
  • Can process large volumes of data quickly
  • Works in real-time or near real-time

Strengths:

  • Fast and scalable
  • Can monitor social media continuously
  • Cost-effective at scale

Techniques in AI Fact-Checking

  • Simple web search
  • Prompting an LLM with a claim
  • Deep Research / AI-powered web search (e.g. Perplexity, ChatGPT Search)
  • Dedicated fact-checking APIs and services

Can AI Decide the Truth?

  • AI predicts based on training data
  • Training data comes from the internet — including misinformation
  • The prediction process is not fully transparent

👉 AI can assist, but requires human oversight

Limitations of AI Fact-Checking

  • Can hallucinate (generate plausible but false information)
  • Struggles with nuance, sarcasm, and cultural context
  • Depends on the quality of data sources
  • May reflect biases present in training data

Conclusion

  • Neither manual nor AI fact-checking is perfect
  • The most effective approach combines both
  • As citizens: use SIFT, check sources, stay skeptical

👉 AI assists — humans verify

Practical Exercise

Build your own AI fact-checking agent:

  1. Take a political claim
  2. Use an LLM with web search to investigate it
  3. Compare with a manual fact-check (Demagog, EUvsDisinfo)
  4. Reflect: where did AI help? Where did it fall short?
Reload?