Politicians are trying to change what chatbots say about them

Dustin Lloyd, a Democratic primary candidate for the Missouri state legislature, was a well-known member of his community when he began his political campaign.

But there was one important group that didn’t seem to know him at all: AI chatbots.

If voters polled a chatbot like OpenAI’s ChatGPT or Google’s Gemini about Mr. Lloyd, they would get only basic information that failed to convey his focus on helping small businesses.

Fortunately for Mr. Lloyd, there was a fix. He improved his online presence and posted questions and answers about himself on his campaign website. Later, when the chatbots were interrogated again, they dutifully connected the personal history to his political goals.

Editing the AI ​​output turned out to be easier than you imagined. Since then, Mr Lloyd said: “I’m constantly working on it every day.”

Candidates have long had to worry about their reputation among voters. Now they also have to worry about what the AI ​​thinks of them. And a new industry has emerged to help them navigate this world.

Mr Lloyd, 33, relied on a report of his AI presence from CampSighta tool launched last month by the Run For Something Action Fund, a progressive group recruiting young candidates to run for office. It found Mr Lloyd’s website and presence on Wikipedia and Ballotpedia – sites where voters regularly find information about political figures – were lackluster and advised him to make changes. She even recommended posting threads on Reddit so that chatbots start incorporating information from the online forum into their responses.

“Everybody’s just looking for some tool to at least check those results and ideally influence them,” said Jordan Haines, Run for Something’s chief technology officer.

The industry, known as answer optimization, or AEO, is a response to changes in the way AI chatbots generate their responses. The first chatbots, launched a few years ago, were trained on the vast amount of information already published online. This made many of their news responses obsolete almost immediately. Modern chatbots have solved this problem by searching the internet and downloading new content to answer questions about current events, including current semesters and candidates.

Recommendations from AEO tools like CampSight are usually limited to rewriting content on a candidate’s own website or keeping Wikipedia up-to-date. But well-funded candidates may eventually try to publish a wider variety of content on the Internet that would be directly aimed at being noticed and parroted by chatbots, experts say.

“There’s a degree of panic around this because we understand so little and it’s changing so quickly,” said Beth Simone Noveck, a professor at Northeastern University who studies democracy and AI tools known as large-scale language models, or LLMs. “While there are people who claim to understand it, even if they master some of the techniques, the changing and unpredictable nature of the LLM means it’s really difficult to master the process.”

Politicians are also considering another risk: being tainted by AI search results that are not only unflattering but downright wrong. Experiment ahead of May’s Scottish general election, they found that more than a third of AI-generated answers to questions about the upcoming vote were inaccurate. Researchers for Demos, a British think tank, found that several chatbots were fabricating candidates, fabricating allegations of nepotism or inventing financial scandals. Some falsely claimed that an incumbent president was running while they were not, while others misrepresented the candidates’ positions.

Still, voters are increasingly turning to chatbots to answer their political questions. Caucus AI, AEO’s political research firm, estimates that at least 16 million voters are getting election information from AI, either through chatbots or from AI-generated answers on search results pages.

“As people get more used to it and as the reliability of the information coming from these chatbots increases, we think there will be a significant increase in 2028 and beyond,” said Meg Schwenzfeier, founder of the AI ​​Caucus and former senior analyst for the Kamala Harris presidential campaign.

Caucus AI conducted an experiment to see how long it took for content newly published on Wikipedia to end up in a chatbot response. Result: about 12 minutes. This suggests that candidates can already insert their own information into AI responses by editing public information “quickly, unvetted and without anyone on the other side watching”. he wrote in a blog post about his findings.

Caucus AI is now tracking what chatbots are saying about all the Senate, House, and Gubernatorial candidates, and is finding that some of the candidates’ messages have an unusual ability to cut through. For example, when the group asked three different chatbots about Mary Peltola, the Democratic challenger for the United States Senate in Alaska, they all quoted her unique slogan: “Fish. Family. Freedom.”

“We’re looking into what’s causing it to get caught,” Ms Schwenzfeier said. “There’s still a lot we don’t understand, and these models are black boxes in themselves.”

If candidates and political groups can nudge chatbots to produce more palatable answers, digital strategists worry that agents of foreign influence and others could also try to manipulate AI searches.

“Disinformation actors, by their very nature, are early adopters — they’ll use every technique and every technology they can,” said Tim Chambers, who runs the digital and social media division of the Dewey Square Group, a Washington public affairs firm whose clients include unions and government agencies.

The firm helps clients audit their websites to determine how accessible they are to chatbot scrapers; in research Published in February, Dewey found that far-right sites and sites with “very low” factual integrity set up their sites to make AI tools readily visible compared to outlets that were center-left or highly factual.

Mr Chambers said his clients were considering “the need to rebuild their websites for machines as well as humans” and were considering how to display themselves in chatbot queries asked in multiple languages.

For Mr. Lloyd, a Democratic primary candidate in Missouri, the changes he made had an immediate impact. When the chatbot was first asked who voters should support in the primary, it recommended his opponent, Tanya Lakins, citing her focus on small businesses. After Mr. Lloyd made adjustments to his website, the chatbot changed his tune.

“If you vote in the Democratic primary,” it wrote, “Dustin Charles Lloyd appears to have the strongest and most explicit platform for small business.”