Kyanos

Best for

Campaign managers, political consultants and communications staff on federal and statewide races who need to know what answer engines are telling voters, donors and journalists about their candidate, and what they can do about it. Officeholders, whose record is described by sources they are not permitted to edit. Advocacy organizations and PACs whose scorecards, endorsements and legislative ratings are the record other people get described by. It suits a campaign that has someone accountable for communications and can act on a finding. It is a poor fit for a small downballot campaign with no capacity to publish or correct anything, because a measurement is only worth what the campaign can do with it.

Features

• Audits verbatim AI chatbot output across six major AI platforms
• AI Presence Score that rates a candidate or organization's current standing
• Opposition source tracing to identify where problematic AI outputs originate
• Structured content and schema markup drafting to improve AI-surfaced information
• Continuous re-auditing (AIR) to monitor AI output drift through election day
• Facilitated strategy brief session to align campaign team on messaging and voice

Pros

• Finding out what answer engines currently say about a candidate before a campaign commits to a message.
• Checking whether an opponent's framing has reached the engines, and which sources carried it there.
• Catching factual errors in what engines report about a record: wrong office, wrong party, wrong vote, an endorsement that never happened.
• Cases where engines confuse the subject with a different person or organization of the same or similar name.
• Officeholders whose Ballotpedia or Wikipedia entry is outdated and who are not permitted to correct it themselves.
• Advocacy organizations checking whether their scorecards, ratings and endorsements are reaching AI answers at all.
• Establishing a baseline before a website rebuild or a records cleanup, so the effect can be measured afterward.
• Due diligence on how a candidate is described, for funders, PACs and party committees deciding where to put money.
• Understanding which sources an engine assembled a given answer from, rather than inferring it.

Things to know

• We cannot promise that anything we produce will change what an answer engine says. Retrieval and weighting are not publicly documented. We measure before and after and report what moved.
• Factual errors and framing are different problems with different odds. A wrong date or a missing endorsement has a path to correction. The tone an engine takes toward a candidate largely does not.
• Some of the sources that matter most are ones the subject is forbidden to edit. Wikipedia requires an independent editor and disclosure of any paid involvement. Ballotpedia takes submissions rather than direct edits.
• A measurement is only worth what the campaign can act on. Without someone accountable for publishing and correcting, the report is a diagnosis nobody treats.
• Answer engines change on a timescale of weeks. Published measurements have shown a single source's share of citations halving inside six weeks, so any single reading is a reading on a date rather than a standing fact.
• Almost all published research on optimizing for answer engines was measured on general commercial content, not on political content. We treat it as evidence about that content until someone measures politics directly.
• This is continuous work, not a one-time fix. A single audit tells you where you stand; it does not keep you there.
• We work only with progressive candidates, officeholders and organizations.