Texas Votes matches candidates to your values, not your party. Every design decision is made to keep the experience fair for all voters — regardless of where you fall on the political spectrum.
Candidates and answer options are shuffled every time so position on screen never creates bias.
Republican and Democratic ballots are generated simultaneously with identical AI prompts and formatting. For undecided voters, even the loading order is randomized.
Party affiliation is intentionally hidden from candidate cards so you evaluate candidates on their positions, not partisan identity.
Recommendations are based on your stated issues, priorities, and candidate qualities — not party registration.
Every question is framed neutrally. Answer options are shuffled. The spectrum picker says "No wrong answers." All deep-dive policy labels use strictly descriptive, symmetric language — reviewed and normalized through independent AI audits to eliminate rhetorical heat from both sides of every issue.
Goes beyond left/right: Progressive, Liberal, Moderate, Conservative, Libertarian, and Independent. Moderate and Independent voters are never auto-assigned a party.
Every proposition shows supporters AND opponents, fiscal impact, outcomes if it passes or fails, and includes a "Your Call" option for genuinely contested issues.
Every AI prompt includes an explicit NONPARTISAN instruction block requiring: factual, issue-based reasoning only; no partisan framing or loaded terms; equal analytical rigor for all candidates regardless of position; and recommendations connected to the voter's specific stated values — never to party-line assumptions. Disclaimers appear on every recommendation screen. Confidence levels (Strong Match, Good Match, Best Available) prevent false certainty.
All candidates are cross-referenced against official filings from the Texas Secretary of State, county clerks, and Ballotpedia. Candidate data includes positions, endorsements, strengths, and concerns — presented with equal depth for every candidate in a race.
Endorsements include context labels identifying the type of each endorsing organization (e.g., labor union, editorial board, advocacy group, elected official). This helps voters understand who is behind each endorsement without having to research each organization separately.
A daily automated updater re-verifies candidate data using AI-powered web research. Each ballot page and candidate profile displays a "Data last verified" timestamp so you can see exactly when the information was last checked.
Every candidate profile includes per-candidate source URLs so voters can verify claims independently — sources are attributed to individual candidates, not just per-race, ensuring every claim can be traced to its origin. Sources are visible on candidate profile pages and in the PWA ballot view, linking directly to official filings, news articles, and campaign materials. All source citations include the URL, title, and the date the source was last accessed.
Each candidate recommendation includes "Why this match?" factors — short phrases citing the specific voter priorities that drove the recommendation. Instead of a generic "Good Match" label, voters can see exactly which of their stated values and issue priorities the AI connected to each candidate. This makes the recommendation logic auditable at the individual level.
All AI research prompts include an explicit 7-tier source ranking policy. When the AI uses web search to gather candidate data, it is instructed to prefer official government filings (Texas Secretary of State, county election offices) over campaign websites, nonpartisan references (Ballotpedia, VoteSmart) over news outlets, and to avoid blogs, social media, and opinion sites. When sources conflict, official filings override campaign claims, and campaign claims override news reporting. This hierarchy is documented in our methodology export.
When a candidate has sparse public information — fewer than two of key positions, endorsements, strengths, or concerns populated — the app displays a "Limited public info" label. This prevents information asymmetry from looking like favoritism: if one candidate has a detailed profile and another doesn't, the label makes it clear the gap is a data limitation, not an editorial choice.
Texas has 254 counties, and local race data is added county by county. When a voter's county doesn't yet have local race data, the app clearly labels this: "Local races for [County] County are not yet available. Your ballot shows statewide and district races only." This appears on both the ballot page and the printable cheat sheet, so voters always know the scope of their guide — rather than silently omitting races and leaving voters unaware.
All Spanish translations are reviewed for partisan bias. Proposition titles and descriptions use identical grammatical structures for both parties. Data terms (candidate names, positions) stay in English for accuracy; only the display layer is translated.
Every screen says "Do your own research before voting." The app is a starting point, not the final word.
All data stays on your device. We collect only anonymous page-view counts (via Cloudflare Web Analytics — no cookies, no personal data) and anonymous usage event counts (like "guide generated") to improve the app. No tracking, no ads. Your political views are never stored on our servers.
The full prompt sent to the AI and every design decision is documented. Nothing is hidden. View the source code, architecture details, and independent AI code reviews.
We've submitted our full AI prompts, data pipeline, and methodology to four independent AI systems (ChatGPT, Gemini, Grok, and Claude) for bias review. Audit findings directly informed improvements, including normalizing all deep-dive answer labels to use neutral, descriptive language with symmetric phrasing across opposing positions. Read the full audit results and methodology export.
An automated balance check system scores the symmetry of pros and cons across all candidates in every race. If one candidate has significantly more strengths listed than their opponent, or if the detail level is uneven, the system flags it for review. Balance scores are published on the Data Quality Dashboard and available as a public API. This ensures no candidate receives disproportionate praise or criticism.
Every candidate card in the app includes a "Flag this info" button. If you spot something that looks biased, inaccurate, or outdated, you can report it directly. Reports are sent to flagged@txvotes.app and reviewed by our team. This voter-driven accountability mechanism ensures the data stays honest even when automated checks miss something.
Our Data Quality Dashboard shows live metrics on ballot coverage, candidate completeness, and county data availability. See exactly how fresh and complete our election data is at any time.