CampSight: The AI Tool Letting Political Campaigns See What Voters Hear from Chatbots

The Problem Nobody Was Watching

When a voter types “who should I vote for governor?” into ChatGPT or Gemini, they get an answer. That answer is shaped by whatever the model was trained on — Reddit threads, news articles, Ballotpedia entries, LinkedIn posts. Not necessarily your candidate’s carefully worded policy page.

Tom Steyer learned this the hard way. Despite spending over $200 million of his own money in California’s gubernatorial primary, AI chatbots were ranking him sixth on cost-of-living issues. Sixth. The models weren’t citing his campaign website. They were citing Reddit‘s r/California.

He placed third. Money didn’t fix the chatbot problem.

What CampSight Actually Does

<!--

Built by Run for Something — a progressive political organization — CampSight runs real AI chat sessions inside actual browser environments. It mimics how a real voter would interact with ChatGPT, Claude, or Gemini, then compiles those responses so campaigns can see exactly what the models are saying.

Think of it as a listening tool for a channel you didn’t know you needed to monitor.

The gap it surfaces is stark. As Amanda Litman, cofounder of Run for Something, put it:

We have already seen there is so much discrepancy between how a candidate describes themselves and how AI is describing them.

Why Chatbots Say What They Say

Here’s the uncomfortable truth CampSight’s research surfaced: LLMs are not reading your candidate’s website the way Google does. They’re shaped by what’s abundant, structured, and accessible in their training data.

A few things that matter more than you’d expect:

• Reddit and LinkedIn get cited far more than Facebook or Instagram
• Paywalled publications like The New York Times are often blocked from AI scraping entirely
• News outlets that sued AI companies show up less in chatbot responses
• Ballotpedia turned out to be a surprisingly powerful source for candidate information
• Poorly structured candidate websites get ignored — or misrepresented

In Mississippi, CampSight found that chatbots weren’t even listing the Democratic Senate candidate when asked who was running. Not malicious. Just absent. There wasn’t enough well-structured information for the model to find.

The Practical Upside

Dustin Lloyd, an Army veteran running for Missouri’s House District 39, started using CampSight and noticed something concrete: his social media reach grew after he applied the tool’s recommendations. CampSight suggested better keyword choices for his policy pages and specific technical coding to improve how chatbots pick up his site.

For a grassroots candidate running lean, that kind of targeted guidance is worth more than a generic SEO audit.

The Bigger Picture

Most campaigns — even sophisticated ones — haven’t thought seriously about LLMs as a messaging channel. Pat Dennis of American Bridge 21st Century said it plainly:

Nobody has written that playbook yet on the LLM stuff. So a lot of people just aren’t doing it.

CampSight is essentially writing that playbook in public, for candidates who can’t afford the internal tools that presidential campaigns quietly use and never share.

Run for Something built a waitlist of over 60 campaigns and organizations before the full launch, with plans to extend it to the roughly 400 candidates they endorse annually.

What This Means for Anyone Watching AI Tools

CampSight is a narrow, specific tool doing one thing well: auditing what AI says about you and helping you close the gap. It’s not trying to be a full campaign platform. That focus is exactly what makes it useful.

The broader lesson here isn’t just political. Any brand, organization, or public figure that cares about how they’re perceived should be asking the same question: what is the chatbot telling people about me?

Most aren’t asking yet. That’s the window.