U.S. political candidates are trying to influence what chatbots such as ChatGPT, Claude and Gemini say about them. Documents obtained by Reuters reveal strategies to get these tools to recommend certain names to voters.
The initiative relies on publishing content online and monitoring AI responses, which are increasingly being consulted by voters seeking information about candidates and proposals.
The strategy known as GEO seeks to influence the information about politicians used by artificial intelligence systems. – Image: Mijansk786/Shutterstock
From traditional search to AI recommendations
A presentation prepared for Mike Minogue, the Republican candidate for governor of Massachusetts against Governor Maura Healey, details how to increase the politician’s presence in virtual assistants’ responses.
The plan, developed by the consulting firm Visigent, draws on GEO, the English acronym for generative engine optimization.The plan is to influence the information that systems encounter when answering political questions.
The document recommends creating hundreds of pages about Minogue’s proposals, publishing references on sites frequently cited by AIs, and producing content for YouTube and Reddit. The strategy also calls for monitoring chatbot responses and adjusting materials published online.
“If you don’t publicize the right information, who knows what AI systems will find?” said Thomas Howell, CEO of Visigent, to Reuters.
Minogue’s campaign did not respond to requests for comment from the agency.
Voters turn to chatbots
Campaigns’ interest tracks a shift in voters’ information-seeking habits. According to Cameron Hickey, CEO of the National Conference on Citizenship, about 25% of eligible voters have already asked a chatbot about a candidate while deciding whom to vote for.
Meg Schwenzfeier, from Caucus AI, which monitors voters’ use of these tools, warns of another aspect of this shift: systems may present unverified information with considerable confidence.
“Chatbots speak with great authority and can consume large amounts of unverified information, so we need to protect ourselves against disinformation,” she told Reuters.
The concern is not limited to content published by the campaigns themselves.Information found in different sources can also influence the way candidates are presented, making the reliability of those references an important issue.
Jordan Haines, technology director at the Run for Something organization, offers a less alarmist assessment. He told Reuters that, so far, models have managed to distinguish reliable information from content of dubious provenance. According to Haines, the organization has observed information published on Reddit appearing in AI responses the following day.
Political campaigns have shifted their focus from traditional SEO to directly training AI robots such as ChatGPT, Claude and Gemini. – Image: Mehaniq/Shutterstock
How much does it cost to influence responses?
Prices vary depending on the service hired and the size of the campaign. The Reuters report presents options ranging from monitoring tools to consultancies offering paid advertising.
- Basic monitoring: Visigent software starting at US$134 per month (about R$668.66).
- Smaller campaigns: CampSight charges between US$9 and US$250 per month (approximately R$44.91 to R$1,247.50).
- Political consulting: Visigent contracts start at US$5,000 per month (about R$24,950).
- Advertising and influencer strategies: costs can reach US$50,000 per month (approximately R$249,500), with compensation of up to US$200,000 (about R$998,000) for a win in a particular category.
Joe Bonilla, managing partner of the consultancy Relentless Awareness, believes the practice will gain scale by the next American presidential election. “Campaigns that use training to influence AI are a reality we face, and this will be ubiquitous in the 2028 electoral cycle,” he said.
Valdir Antonelli is a journalist specializing in digital marketing and consumer behavior.
