The ‘Do My Own Research’ Voter Has a New Research Assistant

I spent years moderating focus groups, and I lost count of how many voters proudly said they “did their own research.” What that really meant was typing a candidate’s name into Google, glancing at ten blue links and, more often than not, closing the browser without clicking a thing.

Now, voters searching for a candidate or issue are often getting a single neatly packaged answer assembled by artificial intelligence from whatever sources it finds most credible and clearly written. And if it answers their question, their “research” is done. 

That’s great if your campaign is one of those sources. If not, it’s a problem.

AI summaries aren’t everywhere yet, but they’re showing up more often, especially for younger, high-information voters. Pew Research reported in October that 79 percent of Americans under 30 see AI summaries at least occasionally when searching for information — about half say they find them somewhat trustworthy.  And those who are still skeptical are going to continue their “own research” starting with the sources the AI cites.

The shift from getting search results to getting “The Answer” will be one of the cycle’s biggest changes in how voters engage with campaign information. A campaign’s ability to shape what appears in these AI-generated summaries is rapidly becoming as important to message discipline as earned and social media.

To see how ready campaigns are for this new reality, we ran a test. We built six criteria to evaluate how campaign websites perform in AI-driven search: discoverability, clarity, and technical structure among them. Then we examined AI summaries generated from common voter questions in the 2025 races. The New Jersey governor’s race offered a revealing case study. Keep in mind AI mode was a feature that didn’t exist when most of these campaigns started, and they have had very little time to test, adjust and broadly change assumptions about online communications. 

Technical Details Matter Once Again

Republican Jack Ciattarelli’s website earned a GEO score of 74, among the highest of all campaigns we reviewed. His site looked great: clear navigation, neatly labeled issue sections (Affordability, Education, Energy), and calls to action that create entry points built for supporters, but which also grab algorithms.

But when we asked Google’s AI, “What will Jack Ciattarelli do on healthcare?”, none of the content on Ciattarelli’s own pages were cited. Instead, the AI answer drew from older sources it had an easy time engaging with and were deemed more authoritative. Unfortunately for him that included the New Jersey AFL-CIO and Planned Parenthood, which had both endorsed Mikie Sherrill.

Looking under the hood, Ciattarelli’s site lacked the technical cues and content flags that tell AI where to look: structured data, clear Q&A formatting, and a newsroom producing fresh, quotable updates. Unlike some candidates, the content he wanted boosted existed on his site, but it was buried too deep and written in a format AI systems don’t easily read to be surfaced in our searches.

Mikie Sherril’s “Official Site Advantage

Democrat Mikie Sherrill’s campaign website scored slightly lower (66) thanks to a JavaScript-heavy design and policy materials locked in PDFs that made it harder for AI systems to index. But Sherrill had an advantage Ciattarelli lacked: a broader, well-maintained digital footprint across multiple platforms.

Her official congressional website and YouTube presence had timely statements and clips, clearly organized by issue. When AI systems looked for credible, structured content, they found her official site before her campaign site (or her opponents’). In our testing, she controlled about half of the citations in her Google AI profile, an unusually high level of message ownership for any candidate.

Sherrill’s example shows that campaigns need a good website, but also need to think of their entire online network and build a voice across channels that consistently reinforces their message in bot-friendly ways.

The New Rules of Message Discipline

Our analysis of campaign websites confirmed what a lot of corporate brands are experiencing: AI publishing rewards structure, freshness, authority and clarity, often over look and feel or visual vibes. 

That makes “boring” web fundamentals suddenly critical again. Candidates who keep their issues in clean HTML pages, fill in the alt-text for images, update press content with dates and bylines, and answer policy questions in clear, quotable language are the ones most likely to control their own narrative. So stop trying to dodge the bots with robots.txt and instead build machine readable pages that show them where to go.

Those who don’t risk letting their opponents and unaffiliated organizations fill the vacuum. In one state, we even saw AI-generated answers citing an opponent’s campaign ad as a factual source about their economic policy. 

For 2026 campaigns the lesson is clear: if you don’t fill the space with accessible, authoritative content, AI will now do it for you — and there’s a good chance you won’t like the answer.

A longtime Democratic researcher, Mike Gehrke is currently the Chief Innovation Officer at GPS Impact.