US campaigns test ‘generative engine optimization’ to shape chatbot answers

From search rankings to AI recommendations

US political campaigns are beginning to treat conversational AI systems as another information channel for voters. A PowerPoint presentation prepared for Republican Massachusetts gubernatorial challenger Mike Minogue and reviewed by Reuters proposed a strategy intended to make ChatGPT, Claude, Gemini and other chatbots present him as the preferred candidate on issues including taxes, housing and public safety.

The proposal goes beyond traditional search-engine optimization. It calls for hundreds of AI-friendly policy pages and citations, placements on websites frequently cited by chatbots, and content on YouTube and Reddit. It also proposed tracking chatbot answers and changing campaign content until responses more closely reflect the candidate’s platform.

A commercial campaign service

Reuters described the approach as generative engine optimization, or GEO. Fees vary widely: small monitoring work can cost hundreds of dollars a month, while campaign strategy, paid placements and influencer outreach can raise monthly costs to tens of thousands of dollars. Visigent, the firm behind the Minogue document, told Reuters that political fees start at $5,000 a month and could reach $50,000 when paid media or other outreach is included. The presentation listed a possible $200,000 success fee for a “category win.”

CampSight, operated by Run for Something Action Fund, offers another model focused on monitoring what large language models say about candidates. Its listed prices range from $9 to $250 per month for some campaigns, with custom pricing and a dedicated strategist for statewide and federal campaigns.

Usage claims need careful interpretation

Cameron Hickey of the National Conference on Citizenship told Reuters that about 25% of eligible voters had asked a chatbot about a political candidate when deciding how to vote. That is a source’s usage estimate, not evidence that chatbot answers determined votes or that the proposed GEO tactics changed those answers.

The Minogue campaign did not respond to Reuters’ requests for comment. Joe Bonilla of Relentless Awareness said AI-influence training would become widespread by the 2028 cycle. Other practitioners were more cautious: Jordan Haines of Run for Something said his organization had seen online material appear in AI answers, but that it had not seen its earlier feared scenario fully materialize and believed language models could distinguish reputable from unreliable information.

Risks for voters and platforms

Campaigns seeking to become “the recommendation inside the AI answer” face the same problem as the systems they are trying to influence: chatbots can present unvetted information with confidence. Meg Schwenzfeier of Caucus AI warned that misinformation safeguards will matter as campaigns compete to shape the information environment.

The documents and pricing show an emerging political-services market. They do not prove that the strategy worked, that a particular chatbot adopted a candidate’s preferred framing, or that voters were persuaded by AI-generated recommendations.

Sources

• Reuters: US political candidates tailor messages to AI models used by voters