What Happened When Voters Used AI Search in the Texas Senate Primaries
AI-generated search answers increasingly appear before campaign websites and news coverage. In the Texas primary, those systems are already shaping how candidates are described to voters.
This analysis is based on research conducted by GPS Impact on public facing websites and AI search behavior from during the Texas Senate primary. (February 3-March 3, 2026)
If you’re tracking tech in politics professionally, you’re seeing a growing number of campaign products branded as “AI.” Most of these are “easy button” projects designed to help make a consultant or campaign’s job smoother by producing ads or simplifying analysis. Some of these have real potential, but an immediate, and much less shiny object is the change happening in search. AI-generated answers now appear before the links voters used to click, and in the last year, this has completely changed how many voters find information and make first impressions about candidates.
In 2024, two-thirds of adults reported getting news from search engines at least sometimes, and 44% of voters ran at least one Google search for information before voting. Today AI answers appear above traditional links, ads, and news carousels. On mobile, where roughly 70% of election-related searches occur, the AI Overview often fills the entire first screen.
The private sector has already responded to this shift. When AI answers appear, some sites see click-through rates drop by 18–40 percent. Companies such as Walmart and Home Depot have begun adjusting their digital strategies to account for AI answer engines and agentic shopping. In politics, the implications are only beginning to appear in campaign strategy.
The Texas Race as a Test Case
GPS Impact audited campaigns in the November 2025 elections in New Jersey and Virginia and in the December Tennessee special election. We found simple steps could move the needle in the control campaigns had over their message in these AI windows. Sometimes the advantage came from basic technical access, such as allowing AI systems to crawl and index content. How content was written and presented also made a big difference. And campaigns that used multiple platforms, including official websites, social media, and video, appeared more frequently in AI-generated answers.
The Texas Senate primary offers another opportunity to examine how AI search shapes campaign visibility. Searches related to both the Republican and Democratic primaries surged in the final weeks of the race, reaching an estimated 1.2 million queries over the past month. Studies estimate that AI-generated answers appeared in roughly 60 percent of those searches, and at that scale, relatively small shifts in citation share can translate into significant exposure. A ten-point increase in campaign-controlled citations within AI summaries would correspond to roughly 72,000 additional impressions among voters actively searching for information.
The analysis below examines how each campaign’s web presence influenced what AI systems surfaced in response to common voter searches. We reviewed code structure, content accessibility, and the depth and format of issue material, then compared those factors with the sources AI systems actually cited when generating answers to questions about a candidate’s bio, issue positions and campaign controversies and scandals.
How AI Search Treated The Democrats
James Talarico’s campaign website presents content that is well-suited for answer extraction. The site consistently describes him as a Texas state representative, former middle school teacher, Presbyterian seminarian, and U.S. Senate candidate, and this language was echoed frequently in AI-generated summaries.
Much of the issue content, however, sits behind links that require JavaScript to load. Those pages are harder for AI systems to crawl, increasing the likelihood that the systems rely instead on external media coverage, Wikipedia entries, or opponent messaging.
And, in fact, AI systems rely only lightly on Talarico’s owned content. Approximately 87 percent of the sources appearing in AI responses come from independent outlets. That share rises even higher for issue-specific searches, such as questions about his plans to address affordability. Voters encountering AI summaries are therefore more likely to see descriptions written by journalists than language originating from the campaign.
Jasmine Crockett’s campaign does a better job placing citations from campaign-controlled and official sources. Her campaign website is fully crawlable and clearly structured, making it easier for AI systems to identify and reuse topical information. She also benefits from a broader official digital footprint. Her congressional website and YouTube channel both account for a significant portion of the controlled information appearing in AI responses. This mirrors our previous studies where AI showed a preference for .gov and longer-established, official sites, even when the content on the two is similar.
Overall, Crockett controls roughly 15 percent of AI citations about her and nearly 30 percent of citations tied to issue-related searches. A strong showing on positive messaging, but her campaign faced a different challenge: Republican-leaning outlets account for a meaningful share of the sources appearing in AI answers about controversies or political disputes. Sites such as Fox News and the Free Beacon appear regularly in these summaries, and in general there is little in terms of response from the campaign or allied sources for AI systems to reference.
Talarico also appears in right-leaning sources, but at a lower rate: roughly one-third the level seen in Crockett’s results. Those citations come from outlets including Fox News, Premier Christian News, and several conservative podcasts. It does not appear to be the case that there were significantly fewer negative stories about Talarico, but right-wing websites were not working as hard (or at least were less successful) in driving them to AI search.
Both Democratic campaigns also maintained a “What Voters Need to Know” page on their sites. These pages were not prominently linked and appear to function as internal “redbox” message guidance rather than public-facing content. Ironically, the pages contain many of the phrases AI systems look for when answering common voter questions, and both were indexed by Google Search and surfaced in AI-generated summaries.
Republican’s Runoff Digital Strategies
The Republicans appear to view their websites differently than the Democrats did, and it showed up in the AI results. John Cornyn’s campaign website performs well in AI search from both a technical and editorial standpoint. The site includes a structured issues section, record-focused content, and several pages using myth-versus-fact formatting. Those formats are often easy for AI systems to extract. It is a solid deep information hub that leverages well with a large official web presence from his US Senate office.
Ken Paxton’s campaign, meanwhile, takes a different approach with his website. It prioritizes donations and data collection, with relatively little issue or biographical content available for indexing. Persuasion and voter education are secondary. The technical side is more difficult to measure, because there isn’t much content to surface. However, the page source on Paxton’s site includes several sections of “lorem ipsum” placeholder text. While not visible to human web users, they are viewable to crawlers, and those elements weaken credibility signals AI systems evaluate when parsing a page.
Despite having a campaign website that is less information-rich, Paxton’s official presence makes up the difference, and matches Cornyn’s overall share of voice. Roughly 20 percent of the sources appearing in AI answers about Cornyn come from outlets he controls, compared with about 19 percent for Paxton.
The differences between the campaigns show up in issue-related searches. Cornyn controls roughly 33 percent of those citations, compared with about 24 percent for Paxton. Much of that advantage reflects the larger volume of policy information available on Cornyn’s campaign site and his active social media and video channels.
Outside organizations also appear in the Republican results. Paxton’s issue-related citations include references from groups such as the Federalist Society and the Republican Attorneys General Association. Cornyn’s results show a different dynamic. Democratic-leaning organizations capture roughly six percent of the message space, with statements and posts from the Democratic Senatorial Campaign Committee appearing in answers about Cornyn’s positions on several issues.
The Sources AI Systems Are Reading About The Texas Senate Race
Stepping back to see which sources show up overall when answering questions about the Texas race, AI systems drew on a surprising range of publications. AI responses frequently included a mix of national reporting, regional outlets, and public media coverage. And many of the outlets cited were not the largest news organizations in the state.
Articles that had been republished or syndicated across multiple websites also appeared regularly in AI answers, sometimes alongside pieces from larger outlets. Among the sources most frequently cited were the Texas Tribune, Houston Public Media, Politico, Fox News, and Austin public radio station KUT.
Several individual articles appeared repeatedly in AI-generated summaries and carried a high level of influence in the AI search narrative:
An April 2025 ProPublica investigation reporting that the U.S. Department of Justice had closed its corruption probe into Ken Paxton as he prepared to run for the Senate.
A Houston Public Media report covering Rep. Wesley Hunt’s announcement that he was entering the Republican Senate primary.
A Texas Monthly profile of John Cornyn from June 2000, (yes, a 26-year-old article, not a typo) written as he was Texas Attorney General preparing to run for U.S. Senate.
A February 2026 Austin Independent article discussing James Talarico’s “mediocre” comment about former opponent Colin Allred.
Houston Public Media’s coverage of the January 2026 debate between Talarico and Crockett.
Video sources also appear frequently in AI search answers. Clips of Crockett questioning Attorney General Pam Bondi, along with interviews with Talarico on The View and The Late Show with Stephen Colbert, surfaced repeatedly in search results.
The AI Search Diet: Top 10 Cited Sources
In the Texas primary, AI overviews heavily favored an unconventional mix of local public media, national heavyweights, and international SEO giants.
Texas Tribune
Houston Public Media
Politico
The Hill
New York Times
Austin Independent
KUT News (Austin)
Fox News (National)
Fox 7 Austin
The Times of India
What Happens When You Ignore AI?
The Texas race offers another look at how AI search is shaping the information environment in the 2026 midterms. AI-generated answers draw from whatever material is easiest to crawl and interpret: campaign websites, news coverage, advocacy groups, and older online profiles. When campaigns publish clear, accessible information, they are more likely to appear in those summaries. When they do not, other sources step in, and those sources often define how the candidate is described to voters.
I suspect the downstream GEO impacts of Trump administration usage of .gov and official gov't social media accounts can only increase the proliferation of US-based propaganda. Ultimately Big Tech risks further eroding user trust in AI models, which furthers the question of what the future may hold -- perhaps a renewed vision of walled gardens of trusted information and journalism will emerge.