Hi readers,

Natasha here — Statistical Journalist at The Markup.

As we draw closer to the 2026 elections, we are again confronted with the question of how, and if, artificial intelligence should interact with our democratic institutions. 

Despite widespread concern leading up to the 2024 elections, AI did not appear to have as consequential of an impact as many feared — at least not in the way some predicted.

Fast forward to 2026. AI has been increasingly integrated into political campaigns. AI-generated misinformation attacks continue to make headlines. President Trump regularly posts AI images on social media. And the world has now witnessed AI wielded in dozens of countries’ elections (for better and for worse). Yet questions about exactly how AI will shape the upcoming elections continue to linger. 

I spoke with political scientist Thessalia Merivaki, the director of Washington State University’s Foley Institute. Merivaki studies how the digital information ecosystem impacts voter behavior and trust in election integrity. Her team built the Election Officials Communications Tracker, a tool that monitors social media posts from election offices across the county, which they use to study the effects of voter education strategies. 

Merivaki and I talked about what happens when more voters than ever are using AI to get election information, what election officials have done to make sure accurate information cuts through the noise, and how we can work toward a more resilient electorate in the age of AI. 

This conversation has been edited for length and clarity.

---

Natasha Uzcátegui-Liggett: What are some of the ways in which AI plays a role in our elections?

Thessalia Merivaki: Voters increasingly use AI tools and AI chatbots to search for information, replacing a traditional Google search. Even when voters do use Google, they’re frequently served AI-generated summaries rather than a list of links to review and evaluate themselves. This changes the nature of the interaction. Instead of selecting which sources you want to prioritize, voters are presented a curated answer with little visibility on where it came from and how reliable it is. Even if there are links, it is very rare that somebody is going to click on them. 

Election officials, political campaigns, and other actors that aim to disseminate information to the public are also exploring AI tools, whether it’s drafting outreach content, managing social media, or even scaling a whole communications operation — especially in jurisdictions where there’s no dedicated staff on communications.

Uzcátegui-Liggett: How does AI impact the election information ecosystem and shape voter confidence? 

Merivaki: AI is reshaping the ecosystem less through generated content aimed for direct human consumption and more through “algorithmic manipulation” — or poisoning — which refers to efforts to corrupt the infrastructure that mediates what voters find when they use chatbots or other AI-powered searches. 

We observed this in 2025 in the Australia and Moldova elections, where Russian influence networks flooded the internet with content that was designed not for voters to review, but for AI crawlers to pool and use for their training data, which ends up getting fed to users as they are using those LLMs. 

This really matters for the whole information ecosystem. It shows a structural vulnerability. In the US, we have over 10,000 local election jurisdictions and although we have somewhat uniform federal and state policies, there are different ways that this information is communicated to voters at the local level. The depth of the official information on any single jurisdiction is very thin, even where the breadth across the country is high. 

This creates data voids — gaps in credible coverage that can be manipulated or exploited by bad actors to fill with low-quality information. This also explains why, in some specific cases, AI outputs are very unreliable, even when there’s no coordinated effort to pollute the information ecosystem. We see that with context-specific and geography-specific questions about voting procedures. The AI systems have little authoritative material to draw from, so they surface low-quality information. 

Clearly, there is a voter confidence problem, because voters will not trust the information. But it also affects how election officials communicate, because they are trying to direct voters to authoritative sources about how and when to vote. We have an asymmetry. It is very difficult, given this infrastructure, for official authoritative content to dominate that AI retrieval environment.

Uzcátegui-Liggett: As a part of your research, you built the Election Officials Communications Tracker to collect data on communications from state and local election officials on social media. What have you learned? 

Merivaki: Social media is not the only tool to communicate with voters; it’s part of a toolkit. But we see that social media has become a core communications channel for election officials. It is a useful and cost-effective way to reach a large body of voters. That said, there’s significant variation in usage, presence, and substance across and within the states. 

Every state has an official elections-related account on mainstream platforms. Local election officials’ presence is very spotty and concentrated mainly on Facebook. That unevenness matters because being absent from some media doesn’t mean that there’s no voter education or interaction with voters that’s happening, the thinner presence can result in less authoritative and high quality information being out there. 

We know that every election jurisdiction prioritizes education — how to register to vote, how to vote by mail, etc. But they also use social media for trust building — explaining how elections are kept secure, what safeguards are embedded in the election process, who they should consider as an authoritative source and also some education about what misinformation is, what AI is, and things like that. 

Using data from our tracker, we found that explicit, focused messaging on how exactly elections are run and the security safeguards integrated into every step in the election process is associated with higher voter confidence, not only among the broader electorate, but also among those who have expressed skepticism in election outcomes. Those communications build trust.

Uzcátegui-Liggett: Can you expand on the ways in which trust-building campaigns by election officials were successful? And what challenges do they still face?

Merivaki: The headline is that, yes, they are effective and successful. I guess the subhead is that it’s complicated. When we test those interventions in an experimental setting, the effects are very high and we get excited. But in real life, there are many more factors we cannot control that make things a little bit complicated. Yes, there’s a substantive impact. No, it’s not long term, because there are other factors that shape how voters think about elections and candidates. So, the question is: How do we build resilience with those communications? 

One of the most prominent trust-building campaigns that we use as an example is a trusted info campaign that the National Association of Secretaries of State put together in 2020 and its core messages have been adopted by other states. The three components are:

1. When it comes to information about elections, your election officials are your authoritative source of information. 
2. We are following rules and procedures to ensure that the elections are secure. 
3. Building trust in the human component of elections — election officials are professionals, they're trained, and they're part of your community. 

Among those three frames, we see a strong relationship between reiterating that we [the election officials] are your sources and voter confidence. The more this is communicated to voters, not only from election officials, but also from other partners that amplify that message, voters will include the election officials as their top information source. In 2024, a quarter of voters reported that their primary source of information was their election official. It’s not 60%, but it’s not zero, so we’ll take that. 

Where these campaigns fall short has to do with timing. When these communications are in the pre-election phase, we see the highest gains. The most critical time, however, is after you cast your vote, especially after election night when you know who won the election. That’s when feelings of bitterness — the loser effect — come into play, and that’s where we see the largest declines in confidence and across partisan lines. 

Add to that some delays in counting because every state has different rules and candidates themselves coming out and saying they think the election is stolen, then you create conditions for significant declines in trust. 

Those declines are the most difficult to restore, because they are grounded in factors that have nothing to do with how elections are really run. That is a major limitation of trust-building campaigns, because they cannot address those dynamics. 

Uzcátegui-Liggett: So how do we work to restore voter confidence in elections in a time when trust-building campaigns and access to reliable information aren’t always enough?

Merivaki: The mindset so far has been rapid response, trying to debunk and fact check. Our research is focusing on the effects of those communications. 

It’s not just being exposed to misinformation or being bitter about your candidate losing. When voters don’t know whether their state follows certain procedures or whether policies are present, that lack of familiarity with election rules can have significant implications for the voter’s ability to be resilient when they come across misinformation. And to step back and think “Yeah, my preferred candidate says that there’s all this fraud with mail voting, but I know how this process works because I have seen it or read about it … I don’t think that’s really true.”

Uzcátegui-Liggett:  It sounds like part of the answer is that we need to do more research. What governance solutions or regulatory frameworks are needed to ensure AI is integrated responsibly into elections and election administration? 

Merivaki: Election officials are not a tech company — they don’t have engineers, not every jurisdiction has a communications director or an IT — but they have to serve all these different roles in their limited capacity and that is obviously a challenge. At the local level, they’re often very small, underfunded and at the same time the demands are high. So that context has to be at the starting point for any governance framework that is actually going to work. 

Operationally, the most critical requirement I would say is how do we nail down having human-in-the-loop oversight at every stage where AI interacts with any public facing output. Internally AI-assisted communications, however routine, also have to be reviewed by humans. That kind of check has to be there and this is the mechanism that maintains institutional legitimacy because in elections, every service is run by humans. We need to retain that and any deviation from that can have significant reputational and trust costs for the public and of course for elections and integrity. 

What we see right now is that we need very airtight, documented prompt libraries for election officials to pull from. We need protocols for vendors, we need more transparency in how those models are built, how they are audited, and who approves the output. We also need a lot of chain of custody features as we are endeavoring to integrate AI tools. 

On the data and privacy side, we need to be very explicit as to what is considered sensitive information and to what extent those apply to internal communications, which mirrors the conversations we have about AI regulation at the federal and the state level.

---

Thanks for reading,
Natasha Uzcátegui-Liggett
Statistical Journalist
The Markup/CalMatters