On Oct. 26, 2026, Berliner Zeitung reported, citing a Reuters investigation, that a Visigent presentation proposed tactics to shape chatbot answers for Mike Minogue, a Republican candidate for Massachusetts governor. The pitch described a plan, while a separate audit by the Institute for Strategic Dialogue (ISD) measured election-answer quality rather than campaign influence.
What Visigent reportedly proposed
The presentation reportedly outlined hundreds of AI-oriented policy pages, paid placements on websites frequently cited by large language models, and content for YouTube and Reddit. These tactics fall under generative engine optimization (GEO), the practice of structuring online content in an effort to make it more likely to appear in chatbot answers.
How CampSight monitors candidate descriptions
Berliner Zeitung also described CampSight, operated by Run for Something Action Fund, as a service that checks chatbot descriptions of candidates daily and recommends changes to campaign websites and communications. That is a separate monitoring approach from the Visigent proposal.
What ISD’s election-answer audit measured
ISD collected 2,400 responses in June 2026 from six models across 10 U.S. state contexts. For generic questions about election procedures, it rated 71% of English responses and 55% of Spanish responses as accurate and complete, a difference of 16 percentage points.
| Language of prompt | Accurate and complete responses | Prompt type and collection period |
| English | 71% | Generic election-procedure questions, June 2026 |
| Spanish | 55% | Generic election-procedure questions, June 2026 |
Five models were queried through APIs, while Meta’s Muse Spark was queried manually. ISD noted that API responses can differ from consumer-facing apps and websites. The study collected one response for each prompt, model, language, and state combination.
The audit evaluated answers to voter-procedure questions and adversarial claims. It did not test campaign interventions or whether candidate recommendations changed, so its accuracy rates measure election-answer quality, not GEO’s effect on campaign visibility or recommendations.
