Ethics and policy
Civic Knowledge Graph Policy on AI and Civic Information
Civic Knowledge Graph helps campaigns make verifiable information easier for people and AI systems to find, understand, and check. We believe the truth serves civic conversation. People should be able to examine the evidence behind political claims and make their own decisions.
Political communication is shaped by the platforms that distribute it and the incentives those platforms create. We believe AI can help people explore civic questions and correct misinformation by connecting claims with evidence. Its capacity to omit context, repeat errors, or express unwarranted certainty makes accountability essential.
We work for campaigns and organizations that advocate positions. Our work aims to improve how those positions and their supporting evidence are represented in AI answers. These commitments govern how we pursue that goal.
Our work
We research what AI systems say and cite; help campaigns publish accessible, sourced explanations; and make documented endorsements and original statements easier to discover. We connect contested claims to relevant evidence and create content explaining campaign funding and documented organizational relationships.
Our work extends beyond a campaign’s website to the public records, analysis, journalism, and statements that help people evaluate its claims.
Our commitments
1. Evidence that readers can check
We connect material factual claims to identifiable sources and preserve the context needed to evaluate them. We prioritize original records and relevant expertise, distinguish facts from projections and opinions, and identify uncertainty. AI output alone is insufficient evidence. A source’s authority on one subject does not establish every claim it makes.
2. Accuracy across political positions
Our evidence standards apply even when the facts are inconvenient for a client. We address material contrary evidence and limitations in factual summaries. Political disagreement is not itself misinformation. We distinguish demonstrably false or misleading factual claims from disputed interpretations, and reserve allegations of disinformation for cases with evidence of intentional deception.
3. Clear authorship and sponsorship
Political materials we create or control identify the responsible organization and relevant sponsorship. Contributions to third-party platforms disclose relevant affiliations and follow their rules. We distinguish campaign statements, endorsements, and independent analysis. Several pages repeating one source do not constitute independent corroboration. An endorsement documents a position; its reasoning must be evaluated on its own evidence.
4. Responsible reporting on money and interests
We ground funding claims in public records and reliable reporting, with dates and context. We distinguish documented financial interests, publicly stated motivations, and our own analysis. A contribution establishes a funding relationship. It does not, by itself, prove a donor’s motive, coordination, or improper conduct. We label inferences and explain their basis.
5. Consistent substance for people and AI
We may structure and deliver information in formats that AI systems can retrieve and interpret more easily. Material factual claims and qualifications in those formats must also be available in accessible public materials for people to inspect. Formatting and organization may differ; the underlying claims must remain consistent. We preserve source attribution and the context that affects meaning.
6. Authentic participation
We prohibit fabricated evidence or endorsements, impersonation, fake grassroots identities, and concealed sponsorship. We do not manufacture apparently independent sources to simulate consensus, misrepresent the origin of information, or embed covert instructions intended to dictate AI answers. We support participation by real endorsers with accurately attributed views.
7. Honest measurement
We distinguish factual accuracy, visibility, citation frequency, and agreement with a campaign’s position. When reporting results, we document the questions, systems, dates, and assessment criteria, including relevant limitations. We do not claim control over AI answers or treat an observed change as proof that our intervention caused it. We do not infer voter persuasion or electoral impact from citation gains.
8. Human accountability and corrections
CKG leadership is accountable for applying this policy to our staff, contractors, and client work. Material factual content requires human review. We review documented challenges, correct substantiated errors promptly in materials we control, and seek corrections from distribution partners. We retain a record of material revisions. We decline or stop work that requires violating these commitments.
Our goal is to help people understand the evidence behind a political position well enough to evaluate it for themselves.