Civic AI Audit · NCoC · Experiment 02 · July 2026
Does AI know who represents you? All 435 districts, mapped.
We asked “Who is my U.S. representative?” for every congressional district —
2,773 graded answers so far across eight product configurations. Each square below is one
district. Pick a configuration to see where it tells voters the truth, where its answer is
stale, where it names the wrong person — and where it refuses to answer at all.
The map
Squares are districts, grouped into state tiles (numerically ordered,
at-large states have one square). Hover any square for the district, its representative,
and every configuration’s verdict.
Accurate
Partially accurate
Outdated (former rep)
Wrong person
Declined to answer
Not yet captured
Every configuration
| Surface | Accurate | Partial | Outdated | Wrong person | No answer | Graded |
| GPT-5 | 394 | 0 | 27 | 11 | 0 | 436 |
| GPT-5 + search | 430 | 1 | 0 | 2 | 0 | 436 |
| Claude Opus 4.8 | 40 | 3 | 3 | 0 | 390 | 436 |
| Claude + thinking | 46 | 1 | 2 | 0 | 387 | 436 |
| Claude + search | 425 | 4 | 0 | 4 | 1 | 436 |
| Gemini 3.1 Pro | 65 | 2 | 8 | 5 | 0 | 80 |
| Gemini + search | 73 | 3 | 0 | 0 | 0 | 77 |
| Grok 4 | 348 | 3 | 55 | 30 | 0 | 436 |
Coverage note. Gemini 3.1 Pro’s run hit its Google project’s daily
request quota mid-experiment; its two configurations cover ~78 districts here and are
completing as quota windows reopen. Verdicts for the four currently vacant seats (CA-14,
FL-20, GA-13, TX-23) are pending human review of the vacancy ground truth. All other
configurations cover all 435 districts + DC’s delegate.
Method. Captured 2026-07-10 via official APIs, one question per district per
configuration. Each response judged by an LLM (Claude Opus 4.8) against ground truth
triangulated from six sources (senate.gov, House Clerk, Wikipedia, ballotinfo.org,
caucus-ai.com, candidata.space), time-versioned and human-reviewable. “Wrong person”
means the answer named someone who is not and was not recently the district’s
representative.
National Conference on Citizenship · Civic AI Audit. Companion page:
Experiment 01 — “Do AI chatbots know your senators?”