How States United Democracy Center tested ChatGPT and Google AI
States United Democracy Center ran two rounds of testing in 2025 and 2026 on two widely accessible tools: OpenAI’s ChatGPT free tier and the AI interface used alongside Google Search. The tests collected nearly one thousand responses submitted by users across six swing states — Arizona, Michigan, North Carolina, Nevada, Pennsylvania and Wisconsin. For Google AI the nonprofit ran two account types: sessions in Incognito Mode and accounts with a browsing history that included election‑skeptical websites.
The 2025 round found verifiable factual errors in 6.9% of Google AI responses and 8.2% of ChatGPT responses — mistakes such as failing to list the correct candidates in a race or giving false guidance about polling‑place locations. Follow‑up tests in 2026 — limited to Arizona, Pennsylvania and Michigan — recorded error rates of zero for both models. The study notes that “this is real progress and should be acknowledged.”
Accuracy improved, but completeness remains a serious shortfall
The topline decline in simple factual errors masks persistent problems. An AI reply can sound authoritative while omitting important details. In the study, ChatGPT produced incomplete lists of current gubernatorial primary candidates 88.9% of the time when queried. The models linked to a state election website — the study’s single most important measure of voter utility — less than 40% of the time.
“It will be like ‘this person is the Republican candidate and this person is the Democratic candidate’ but it is not telling you there’s also these other third‑party candidates,” Sanchez said. That pattern risks giving voters a truncated view of races if they rely on chat interfaces as their primary source.
Audit-ready is a season. It shouldn't be.
Evidence in spreadsheets, controls drifting between audits, frameworks multiplying on flat headcount. Nubivance runs continuous compliance on Rapid7 Cyber GRC - SOC 2, HIPAA, ISO 27001, PCI, CMMC.
End the scrambleProduct churn changed available outputs between test rounds
The nonprofit’s two rounds of testing also captured how quickly UI and algorithmic choices by major tech companies can alter voter experience. In February — between the first and second rounds — Google AI abruptly shifted its treatment of election queries in Incognito Mode, moving from written summaries to providing only links for election‑related searches. The study highlights this kind of product churn as an ongoing constraint: frontier AI companies are “constantly tinkering with their models, their algorithms and the technologies they are intertwined with.”
That churn matters because election officials, by contrast, have decades of experience educating voters and typically provide authoritative, stable sources of voting information.
“Generative engine optimization”: how content creators and bad actors respond
Isabel Linzer, an elections policy analyst at the Center for Democracy and Technology, told CyberScoop that voters, campaigns and governments are all using AI more freely and with fewer restrictions. Bad actors have followed suit. “We are in a phase now of generative engine optimization” where information operations are structured to rank higher in AI model responses, Linzer said.
“We’ve moved beyond [SEO] to [Generative Engine Optimization], and that’s where we’re seeing campaigns thinking about how to structure their materials to make sure that they are in a format that AI models want to use when they’re searching the web…to develop their responses to user queries,” she added. The story also notes that political campaigns are deploying deepfakes of their opponents, and that AI systems have been developed to carry out increasingly complex hacks.
What this means for voters, campaigns, and election officials
- Voters: Chatbot adoption is growing. A June Pew Research Center survey found about half of U.S. adults reported having used chatbots at least once, up from a third in 2024, and a quarter reported daily use; the top use case was searching for information. Still, the study authors and Linzer say most people are best served by going directly to local sources for accurate election information.
- Campaigns and content creators: As Linzer described, campaigns are already thinking about how to format content to be favored by AI models — a practice the study frames as an evolution of search‑engine tactics into “generative engine optimization.” That shift can amplify both campaign messaging and disinformation operations.
- Election officials: The study underscores a simple, practical priority for tech companies and officials alike: connect chats for “high‑stakes situations like elections” directly to authoritative sources. Linzer put it bluntly — make sure chats link to “the website where you can actually register to vote.”
The research shows measurable gains: basic factual error rates fell to zero in follow‑up tests. But those gains coexist with substantive gaps in completeness, linking behavior, and a fast‑moving product environment that can change how voters encounter information overnight. For a electorate increasingly turning to chatbots for local race details, the question the study leaves on the table is concrete: will platforms and content creators make official election sources the default, or will conversational interfaces keep serving polished but partial answers?
