Take Precautions When Using AI for Election Research

AI chatbots are increasingly used to gather information for election research, but they can also spread misinformation. The source notes that hundreds of races are happening in the US midterm elections, and many people turn to tools like ChatGPT, Claude, or Gemini to learn about candidates and ballot measures. However, generative AI carries risks such as bias, sycophancy, hallucinations, and the ability to draw false conclusions from bad data. The article advises users to take several precautions before relying on a chatbot for political research.

Risks of AI in Election Research

• Sycophancy: AI may tailor answers to please the user, potentially giving false or misleading information.
• Hallucination: Chatbots can invent facts, especially when asked about polling locations or ballot layouts.
• Bias: Models are trained on human‑written data, so they inherit societal biases and can reinforce stereotypes.

How to Verify Information

• Check official sources: Use state or local election websites for polling places and ballot designs.
• Inspect the AI’s citations: Look for video clips, news interviews, or reputable articles rather than random social‑media posts.
• Cross‑reference: Compare the chatbot’s answer with multiple independent sources before drawing conclusions.

Practical Steps for Researchers

1. Start with primary sources – official government pages, candidate websites, and reputable news outlets.
2. Use the chatbot as a helper – it can organize information but should not be the sole reference.
3. Verify claims – if a statement seems surprising, search for the original source or ask a fact‑checking organization.
4. Document your process – keep notes on which AI tool was used and where the information came from.

The article also mentions that a 2023 New York Times analysis found Elon Musk’s tweaks to Grok made its answers more conservative, and that political campaigns sometimes try to “poison” search results by creating flattering content for a candidate. Users should be aware that external actors can influence AI outputs, and that the technology is not neutral.