Why this matters
AI and social media can strengthen democracy, or they can wear it down. Voters have long relied on clear rules in broadcast political advertising: sponsors identify themselves, candidates stand by their ads, and the public can see who paid. Social media brought only partial, mostly voluntary versions of those rules, while recent rules have further weakened financial transparency. AI systems and recommendation algorithms now shape much of what voters see, and more and more they talk to voters directly. The same basic expectations of honesty, transparency and fairness should apply to them.
AI is already being used in this election in ways that erode public trust, and much of it happens where the public can’t see. Federal rules haven’t kept up, state rules are inconsistent and shifting in the courts, and past voluntary pledges aren’t sufficient, haven’t been reliably kept, or are hard to check. Public, on-the-record answers can happen now. These principles are meant for the 2026 election and beyond.
Principles, and what we ask companies to show the public
We ask every major AI and platform company for written, public answers on each point below, this election cycle and every one after it.
1.Authentic and clearly labeled
Voters can trust that what they see and hear from a candidate, PAC, or other election-related organization is real, and can always tell when they’re dealing with AI. Synthetic media of real people is used only with their consent or as obvious and clearly labeled parody. Platforms and AI tools refuse to create or run political ads that deceptively depict real candidates or election officials, and apply extra review in the final weeks before an election. AI-generated political content is labeled where people will notice, and ad libraries show which political ads use AI. Likewise, AI personas never represent themselves as human: AI-run accounts, chatbots, influencers, or agents say clearly that they are AI and who is behind it, and confirm that whenever asked. Campaigns and election offices sign their official media so anyone can verify it.
Show the public: What AI-generated political content and ads you label, what you refuse to create or run, and how well those safeguards work.
2.Transparent funding
Platforms require political advertisers to disclose who pays for their messages, including money routed through intermediaries and payments to influencers, and show this in their ad libraries. AI and platform companies disclose their own political spending, and that of the groups they fund.
Show the public: Your company’s political spending this election cycle, and that of its affiliated groups.
3.Fair and transparent platforms
Platforms and AI systems apply the same published rules to every candidate, party and viewpoint, and don’t covertly favor or suppress any of them, whether at the direction of outside actors or of their own owners, executives or staff. Companies explain how their ranking, recommendation and AI systems handle political content, and open them to independent researchers so the public can check that they are fair.
Show the public: How you detect and stop efforts to manipulate elections through your systems, whether by foreign or domestic actors outside your company or by owners, executives, employees or contractors inside it, and who independently checks those safeguards. And what you have changed this election cycle in how your systems rank, recommend, moderate or generate political content, and who approved those changes.
4.Accurate voting information
AI products direct voters to official sources on when, where and how to vote, and work to prevent false or misleading information about voting. Companies work closely with election officials of every party.
Show the public: How your AI products answer questions about candidates and voting, and how you test those answers for accuracy and even-handedness.
5.Accountability in public
Companies report before and after each election on the threats they find and the fixes they make, and correct problems quickly when they occur.
Show the public: The election-related threats you have found and acted on, and what data independent researchers can use to check your work.
Companies we are asking
Tier 1
Companies whose products most shape what US voters see and hear about elections:
- AlphabetGoogle Search, YouTube, Gemini
- AnthropicClaude
- MetaFacebook, Instagram, WhatsApp, Threads, Meta AI, Muse
- MicrosoftBing, Copilot, LinkedIn
- OpenAIChatGPT, Sora
- SpaceX / xAIX, Grok
- TikTokTikTok
Tier 2
Companies with a growing role in how voters get information:
- AmazonAlexa+
- AppleSiri, Apple Intelligence news summaries
- PerplexityAI search
- RedditReddit, Reddit Answers
- SnapSnapchat, My AI
We also urge oversight bodies to look at widely used AI models built abroad, such as DeepSeek and Alibaba’s Qwen, which fall largely outside US oversight.
How public officials can help
Election officials, secretaries of state, attorneys general, and federal agencies such as the FEC, FCC and FTC, of every party, can:
- Request written, public answers to these questions from every major AI and platform company.
- Use their existing oversight and investigative authority, focused on security safeguards and factual records.
- Work with companies on fast, reliable channels for correcting false information about voting.
Members of Congress, state legislators and local councils of every party can also request written, public answers, and hold public hearings where company leaders explain their safeguards.
Where this came from
This statement grew out of participant discussions at a cross-partisan, cross-sector gathering at the SF Innovators & Investors for Democracy Forum on October 8, 2026, co-hosted by DemocracyWeek.net and the Center for Constitutional Democracy at UC Law SF, plus related cross-partisan discussions prior to the Forum and since. People across the political spectrum have shared real hope for what AI could do to help strengthen democracy, and real concern that careless deployment and weak governance of AI could break it. Everyone is welcome to sign. Partners and collaborators also welcomed.
Last updated: Oct 23, 2026