A new ISD report showed that the top AI chatbots are often providing outdated, unclear and inaccurate responses to questions about elections and voting procedures.

In a recent article, NBC News heavily covered the methodology and findings of this report and spoke to its authors: ISD Director of Civic Innovation Max Read and Digital Research Analyst Valeria de la Fuente.

“These are sources or tools that are serving information to users beyond what they are searching for and doing it in an overconfident way that sometimes misrepresents the information that they’re providing,” Max said.

The outlet also highlighted ISD’s findings that the models performed notably worse when answering Spanish-language questions.

“The gap in performance between English and Spanish results, especially with the generic prompts, was on average 16%,” Valeria said, explaining that the gap was largely due to models’ incomplete or ambiguous answers in Spanish.

Max and Valeria explained that this research conveys the importance of ensuring official, election-related resources online are kept up to date, and AI companies must improve their products’ election accuracy.

“The developers are saying that these are tools for gathering information that will be more complete and more accessible than Googling or going to find it yourself,” Max said. “If that is the standard that they’re setting for themselves, then it’s incumbent on them to make sure that they’re not giving overconfident answers or not missing or misrepresenting exceptions to rules.”

Additionally, The Hill extensively featured ISD’s research methodology and results, quoting the report in its coverage. The New York Times also cited the reports and highlighted its findings.

Time interviewed Valeria, who discussed where the chatbots fell short, such as in its responses to Spanish-language prompts and ability to debunk harder-to-refute election-related concerns.

“We know that voters are going to turn to chatbots more and more,” Valeria said. “So the quality of the responses that we found is concerning.”

Further, Valeria spoke to Factchequeado on the differences between interacting in English and Spanish with the chatbots. The outlet emphasized ISD’s finding that the models performed worse in Spanish; Valeria explained the possible reasons behind the gap, including models’ lack of precision in Spanish due to fewer available sources.

POLITICO also covered the report in its Weekly Cybersecurity newsletter, quoting Max’s caution to non-English speakers when getting their information from AI.

The full articles are available on NBC News, The Hill, The New York Times, Time, Factchequeado and POLITICO.