What leading AI models tell three Senate candidates about housing, energy, immigration and Medicaid
Campaigns and their staffs increasingly turn to AI models to test positions, draft messages and prepare for debates. So we asked a simple question: what do those models actually say?
Using AI Frame, Signal’s digital intelligence platform, we modeled three 2026 Senate candidates (James Talarico, D-TX; Abdul El-Sayed, D-MI; and Michele Tafoya, R-MN) and put the same four issues to five leading models: ChatGPT, Claude, Gemini, Grok and Perplexity. We analyzed 160 responses and 1,468 cited sources.
What we found:
- On data centers, the advice converged. None of the 45 energy responses recommended a ban, moratorium or pause. All 15 stance responses told candidates to welcome the growth and make data centers pay their own way, Democrat or Republican.
- Local news and think tanks feed the answers. State outlets are cited at about twice the rate of national outlets, and no national news outlet is among the most-cited sources.
- The model you ask matters. Citation habits, source recency and coverage of key facts differ widely from one model to the next.
Download the full report for the findings on all four issues, what each model told each candidate, how that advice compares with their public positions, and the sources behind the answers.