# AI Search Is Fixing the Down-Ballot Problem. Now Pay the Sources. *Voters are using ChatGPT to compare candidates nobody covers, campaigns are publishing plainer policy pages to get cited, and a manipulation attempt was caught in six weeks because the sources were visible. The midterms show the upside of AI search in elections. One piece is missing: paying the local publishers who make it work.* Published: 2026-11-01 | Read time: 5 min read | Author: Smalk AI Research | Source: https://www.smalk.ai/blog/ai-search-elections-down-ballot-upside-midterms --- AI search in elections is doing something the open web never managed: giving voters a usable comparison of candidates in races no newsroom covers. The midterms are the first cycle where a voter in a news-desert county can ask one question and get a cited answer about a city council race, and the same visibility that let a fake newsroom slip into ChatGPT is what let NPR catch it within weeks. Read the cycle as a search veteran and the pattern is familiar: a new medium arrives, the first abuse is public, and the market that follows makes it both safer and more valuable. - Pew's October 23 survey finds 8% of U.S. adults use chatbots for election information, and most of them use it to compare candidates' positions and fact-check claims. - Campaigns are responding by publishing clearer, structured policy content designed to be cited, which raises the floor of information available in low-coverage races. - NPR's discovery of 37 disguised political sites proves the medium is auditable: AI answers show their sources, and Google's systems excluded the network entirely. - Local publishers are the input engines reach for first, which turns a shrinking audience into leverage in the answer economy. - The one unbuilt piece is a disclosed, paid lane on cited pages, so the upside for voters funds the newsrooms that supply it. ## What AI search in elections does for voters AI search in elections is the use of conversational engines such as ChatGPT, Gemini and Claude to compare candidates, verify claims and find voting logistics in a single synthesized answer with cited sources. Pew's October survey of 5,273 adults shows what the 8% actually do with it: 76% look up candidates' positions, 71% check the accuracy of things they heard elsewhere, 65% research candidates' records, and 40% use it for registration or polling-place details. That is the profile of an informed-voter tool, not a persuasion machine. ## The search-ad election versus the answer election In the open-web election, political information reached voters through ranked links and targeted ads, and the richest races got the most coverage while down-ballot contests disappeared below the fold. In the answer election, the engine assembles a comparison on demand from whatever credible pages exist, so a school-board race gets the same format as a Senate race. The open web rewarded whoever bought the most impressions. The agentic web rewards whoever publishes the clearest, most verifiable source. For campaigns that is a better incentive, and for voters it is a better product. ## What NPR reported, and what the data adds NPR's October 30 investigation, built on a VisionHeight report, found 37 websites registered within eight minutes on September 15, ten of them posing as local news, promoting mostly Democratic candidates; ChatGPT cited the network in 19 of 37 test prompts while Gemini and Google Search cited none. Reuters reported on October 26 that consultants now charge $5,000 to $50,000 a month to shape chatbot answers, and that at least one pitch proposed hundreds of plain-language policy pages. NPR describes some voters turning to chatbots; Pew's figure is 8% of adults, the AAPC Foundation's September survey says about one in five voters, and we use Pew. The Center for Campaign Innovation's September 16 study across 80 races adds a reassuring detail: campaign sites appeared in 55% of answers without any measurable favorability lift, meaning presence informs more than it persuades. ## Why the fake-newsroom story shows the medium working ### The strongest case for alarm Georgetown's Renée DiResta told NPR that posing as a news outlet crosses into deception, and she is right. The network exploited a data void, the gap in credible local coverage that Medill's October 13 report puts at 238 news-desert counties, and OpenAI's fix was still incomplete when NPR re-ran its queries. A medium that cites impostors in the races with the least scrutiny is a medium with a real vulnerability, and watchdogs are correct to say so. ### Why visible citations beat invisible feeds The abuse was caught because AI answers expose their inputs. A dark-posted social ad in 2016 reached its target and vanished; a citation in a ChatGPT answer is a public, repeatable, auditable claim that a journalist could reproduce, trace to an eight-minute domain registration, and publish six weeks after launch. Google's ranking systems excluded the sites outright, proving the quality layer transfers. The structural fix is not fewer machines but more legitimate supply: engines want the real Kalamazoo page, and the void exists only because nobody has paid for it yet. ## What this means for campaigns and for local publishers ### For campaigns, advocacy groups and the brand teams watching: clarity is the new spend The winning tactic in Reuters' reporting is the least glamorous one: write clear, structured, verifiable policy pages and get them onto sources engines already trust. Monitoring is now cheap enough for a school-board candidate, with CampSight starting at $9 a month, which narrows the gap between well-funded and shoestring campaigns. Every CMO should read this as a preview: the answer economy pays for being the clearest source, and it penalizes disguise the moment a reporter asks what the citation is. Build on legitimate pages and ask for a disclosed lane. ### For local publishers: the void is a demand signal with your name on it Every fake Beacon is evidence that engines are searching for a local page and not finding one. Medill reports traffic to the 100 largest local papers down nearly 50% since 2022, yet the same report shows local reporting is the input AI answers reach for first in a political question. That is leverage, not loss. Open the page to crawlers, structure it so the answer is extractable, measure which pages get cited, and price that citation as inventory. The newsroom that covers the council race becomes the source of record for every voter who asks about it. ## Three ways the upside compounds before 2028 First, adoption: Pew already finds 11% of adults under 50 using chatbots for election information, and down-ballot comparison is the use case with no incumbent competitor. Second, supply quality: as campaigns publish structured policy content to be cited, the baseline of verifiable candidate information rises in exactly the races that had none. Third, a disclosed market: OpenAI's political ad ban keeps the answer free of paid slots, and the next step is a labelled, paid lane on the publisher pages engines cite, so the civic value voters are getting finally funds the local journalism it depends on. ## Conclusion Hold on to this: AI search in elections is raising the floor of what a voter can know about a down-ballot race, and the medium's transparency is why its first abuse was caught, not why it succeeded. Generative Engine Advertising, native ads for AI agents placed openly on the legitimate pages engines cite, is the positive-sum model because it gives campaigns and brands a disclosed way to be present and gives local publishers revenue for being the trusted input; that is the rail Smalk AI is building for both sides of the market. What to watch next: the first local newsroom to report citation revenue from a political cycle, because that is the moment the civic upside starts paying for itself. ## FAQ ### How is AI search helping voters in the 2026 midterms? AI search gives voters a single cited comparison of candidates, including in down-ballot races with little or no news coverage. Pew finds that among the 8% of adults using chatbots for election information, most compare candidates' positions, fact-check claims, and look up registration or polling details. ### Is ChatGPT a reliable source for election information? It is a useful starting point with visible sources, not a final authority. NPR found ChatGPT citing disguised political sites in races with thin coverage, while Gemini excluded them, so voters should open the citations, and election offices remain the authority on voting logistics. ### Why is AI search better than social feeds for political information? AI answers show their sources, which makes manipulation reproducible and auditable. NPR traced a fake local-news network from a ChatGPT citation to an eight-minute domain registration within weeks, something a dark-posted social ad never allowed. ### What are campaigns doing to show up in AI answers? They are publishing clear, structured policy pages and monitoring how chatbots describe them, a practice called generative engine optimization. Reuters reports fees from $9 a month for monitoring tools to $50,000 a month for full-service work, and the content that performs is plain, verifiable and placed on trusted sources. ### How can local newsrooms benefit from AI search in elections? Local pages are the input AI engines reach for in political questions, so a newsroom that keeps its pages open, structured and measurable becomes the cited source for its district. The next step is pricing those citations, so being the trusted source generates revenue rather than only visibility.