- New research has found that leading AI chatbots frequently give incorrect election information.
- Researchers say systems like ChatGPT, Gemini, Claude, and Grok still struggle with their provisioning and policy accuracy.
- AI chatbots become reliable information tools even before they are reliable in elections
AI companies want to make their chatbots a source of information. But a new study suggests that election information remains an area where that trust could outpace technology readiness.
NewsBench, a project created by Forum AI studying how AI systems handle journalism and news, found that leading AI chatbots repeatedly struggled when asked election-related questions. These findings add to growing evidence that conversational AI systems remain unreliable in one of the highest-stakes categories: helping people understand democracy itself.
“Ask one of the leading AI chatbots about the upcoming midterm elections, and there is a 90% chance that the answer will be wrong in some way: a factual error, an obvious partisan bias, a quote to a foreign state-controlled media outlet, or some combination of all three,” Forum wrote in a summary of the study.
The broader concern extends beyond a single chatbot or company. ChatGPT, Gemini, Claude, Grok and other major chatbots all have problems.
NewsBench researchers specifically focused on how AI systems retrieve and present factual information from journalistic sources. Their findings point to a recurring weakness. The problem is often not the reasoning. It’s recovery.
Confident issues
AI systems often fail because they surface weak sources, incomplete information, or incorrect elements before they even begin to generate a response. The researchers found that retrieval failures accounted for more than 70% of the observed errors. When systems successfully retrieved reliable information, they often responded correctly. Obtaining the right information consistently remained the most difficult challenge.
This problem becomes particularly uncomfortable during elections. The chatbot appears confident, regardless of its accuracy. Answers are clearly written with citations and authoritative language. Even incorrect information can appear trustworthy when presented with enough confidence.
Election-related studies increasingly show how dangerous this combination can become. Chatbots often mix precise details with inaccuracies in a way that appears transparent to users. The result is unlike the disinformation sites of earlier Internet eras. This sounds like expertise.
This distinction is important because people increasingly treat chatbots as experimental software and more as infrastructure. OpenAI, Google, Anthropic and other developers regularly encourage users to independently verify important information. Companies continue to sign licensing agreements with publishers, partly in hopes of improving the quality of their supply and reducing factual failures.
Electoral pressure
Even high-performing models struggled when source selection failed. Accuracy often declined further when questions contained subtle inaccuracies or misleading assumptions, the same way real users often phrase their questions online.
Timing creates additional pressure. AI companies offer increasingly sophisticated information tools while regulators continue to evolve unevenly depending on the country. Europe has placed greater emphasis on transparency requirements. Other governments stay earlier in the process. Meanwhile, adoption continues to grow.
The long-term answer could involve stronger source attribution, more transparent retrieval systems, better provenance technology, and a stronger editorial infrastructure under AI products. The challenge is that elections don’t wait until tech companies have finished improving their systems. Voters are using the tools available today, and it is clear that those tools need improvement.
Follow TechRadar on Google News And add us as your favorite source to get our news, reviews and expert opinions in your feeds.
The best business laptops for every budget
