VotingVibe

How VotingVibe works

Your values. Your decision.

VotingVibe helps you understand your ballot and compare candidates against your own priorities, with evidence you can read for yourself.

We do not endorse candidates or tell you what to care about. You set the priorities; we show how the candidates compare.

How we stay neutral

  • Sources you can check

    Election information and candidate positions link to their sources. We keep a copy of the pages we cite so you can check the quotes. Information without a citation is not shown.

  • Every candidate gets the same treatment

    We use the same research process for everyone in a race. You can compare their positions, read the case for the runner-up and see the arguments for and against each measure.

  • Your priorities guide the match

    You choose what matters and how much it matters. Party is never an input to your match, and we never fill a gap with what a candidate's party usually believes.

  • Clear limits on what we know

    Missing evidence stays missing. We distinguish a match from a lean or a limited comparison, and say when there is too little information to compare candidates.

What you can do here

  1. Find your elections

    Enter your address, or browse by state and county. We show the location we are using and give you a way to change it. Your street address is discarded after the lookup.

  2. Explore your ballot

    See candidates and measures, open candidate profiles and compare their positions. Ballot previews assembled from official lists are labeled as reconstructed. Measure explanations show what a yes vote means and the case for and against, with sources for each sentence.

  3. Compare candidates with your views

    Start by telling us what matters to you, in your own words or from a list, then answer questions about your priorities. You can skip questions, change answers or stop at any time. Your results explain each comparison and link to the evidence. Copy, download, print or email yourself a ballot summary. You can also download a detailed research file to use in your own AI chat.

  4. Get ready to vote

    Check polling places, early-voting hours, registration deadlines, mail-ballot dates and ID rules where available. Each item includes its official source and when we checked it. Always confirm the details with your election office.

A closer look at the method

The same answers and evidence produce the same result. AI helps read sources and explain findings; a fixed formula determines the match.

How questions are chosen

We first ask what matters most to you. You can write it in your own words, pick from a list that is the same for every voter, or skip it. We read back what we understood so you can correct it or set something aside. Anything you name stays on your list, even when no candidate on your ballot has said anything about it. Naming a concern never sets a position for you: we ask which way you lean only where candidates have spoken on it, and those questions come before the rest.

Then we ask about differences between candidates on your ballot. Questions about reported records and campaign funding come first, when relevant. Broad policy questions follow, then more specific questions where your answers leave a race undecided. Endorsements come last, when they could still affect a comparison.

Broad questions come from a fixed list with wording reviewed by a person and changes recorded. They apply to a race only through issues that office actually decides. The question card lists those issues and races. If you answer a specific question, it counts on its own and is no longer included in the broad answer, so the same evidence is not counted twice.

The same issue is asked once even if several offices decide it. If duplicate issue names are merged, your saved answer carries over; if you answered both, one answer is kept. Candidate qualities, such as experience or local roots, can also be questions. Memberships in churches, clubs or professional bodies are shown on profiles but are never interview questions.

You can open background on an issue, including the strongest case for each side. We research it for the election's state, prefer sources about that state and share it with other voters there. Questions that name a state must name yours, never another; when no source about your state is available, the question uses general wording instead. The background is cited sentence by sentence, and both sides receive the same space. An endorsement or contribution is never treated as evidence of a policy position.

You can skip any question. When no further answer can distinguish the candidates, we stop asking and show their records side by side so you can choose for yourself. Reaching the end of the questions does not settle your concerns: any that no candidate has addressed are listed as still open.

What a match, lean or “closest to you” means

Each race is scored on the issues its candidates have addressed, weighted by how much you say those issues matter. An answer about one office does not count against candidates for another. Silence counts as neither agreement nor disagreement; one agreeable statement does not automatically outweigh a fuller record.

Each race also shows what applies to it from what you answered or named, judged by what that office decides rather than by what candidates happened to say. It separates the questions two or more candidates have addressed, those only one has, and those nobody in the race has addressed. Questions nobody has addressed are never scored. Where a candidate has not spoken on some of your answers, the score breakdown shows how far their fit could move if those unknowns went either way; that range is shown for context and never changes the ranking.

A match means candidates have addressed the same question and one is clearly closer to your answers. A smaller lead, or a comparison based on different questions, is a lean. “Closest to you” is a more limited comparison: it may reflect the least disagreement, a single shared view or the only candidate we can compare so far. These results are grouped separately and are not counted as matches.

Equal scores are shown as a tie. When only one candidate has a public record we can compare, we say so and do not call it a match. Unopposed races are labeled, included in your summary and not scored or researched. A race with no usable positions still shows every candidate and the search record.

The results include a contents menu, a race picker and links from your ballot summary to each race. Race cards, score breakdowns, funding, shared positions and your priorities can be expanded when you want the detail. Questions that could resolve a race are called out near the top; their links open the relevant card and take you straight to the question. Where no answer could settle a race, you can make your own choice; it is labeled as yours and does not change any other race.

Results appear before the longer explanations are written. Loading indicators identify explanations still being prepared, including on collapsed race cards. The initial reasons cite candidate claims; a plain-language explanation replaces them when ready, with a source for every sentence. After an answer, your previous results stay visible with an updating indicator until the new calculation finishes, and the race you acted on reopens. Research status remains available, and you can refresh results to include newly available evidence. Your ballot summary can be copied, downloaded, printed or emailed once. An email address used to send that summary is not kept.

Candidate qualities, dealbreakers and voting records

We research experience, local roots, independence, endorsements and reported arrests, charges, convictions or official findings. These are sourced facts, including the reported outcome, rather than labels. Independence is read from what candidates do and say about their party, such as votes against its majority, never from the party label on the ballot; older readings that used the label are replaced as their pages are read again. The same search is used for every candidate in a race. If we find no record or tie, that means unknown, not proof that none exists.

You decide whether a quality matters and which direction you prefer. A dealbreaker is a filter applied before scoring: candidates with a matching record or tie remain visible with the source, but cannot be your match. Endorsements can matter to you without being treated as a candidate's policy position.

For current members of Congress, state legislators and governors whose voting records we read, we compare stated positions with recorded votes. Once enough votes are available, votes carry more weight than statements. Consistency is calculated in code and only affects your match if you say it matters.

A consistency comparison requires at least two issues, each with a stated position and at least three counted votes. We show whether words and votes agree, partly agree or conflict, with the counts and issues. Candidates without a voting record we read are not scored on this quality. A missing record is never described as a failure to vote; an issue with no counted votes in a record we read is labeled “no votes on this.”

Campaign funding and outside spending

For federal candidates, we read Federal Election Commission filings for totals, committee contributions, donors' reported employers and independent spending for and against them. Names appear as filed. We do not infer connections between committees or keep donor addresses. Money sent through a conduit is counted as individual contributions, as the FEC counts it.

We match candidates to FEC records by seat, last name and first name, without guessing nicknames. When there are two records for one person, we use the one they filed under in this election's cycle. If we cannot resolve the records, we show no filings rather than choose one.

State and local coverage depends on the disclosure records we support. Some records provide totals; others provide committee contributions without totals. If records do not identify which contributors are committees, we do not guess from their names. Each candidate page explains what was and was not read. We do not yet cover every state's filings.

Funding questions use the categories in the filings, not committees selected by name. You can browse every committee recorded in a category for your ballot and search by name in your results. We ask separately about money a campaign accepted and money others spent independently, which the campaign cannot refuse. A dealbreaker about accepted contributions never includes independent spending. Spending against a candidate is shown but never treated as a tie to them. Missing entries are unknown, not a penalty.

Our goal is to help you vote on your values rather than on how much a candidate spent to reach you. The homepage shows total federal candidate spending for the election cycle when the FEC figure is available, with a citation and a record of the requests behind it. This is context for your research, not an endorsement or a scoring input. Federal finance data is read at most once a day.

Where the information comes from

Election information comes first from state and county election offices, official candidate lists, polling-place files and the Google Civic Information API. Where no feed lists candidates, we read the election office's own pages when someone first opens that ballot. This information is labeled as reconstructed, and the published result is shared with everyone.

Candidate positions come first from their own sites, then questionnaires they answered, interviews, news that quotes them and official records. Sources must name the candidate and race. A guide covering several candidates is read for each candidate it names. When one has addressed an issue and another has not, we search again for the other candidate on that issue using the same process. When a concern you confirm applies to a race and nobody there has addressed it, we search every candidate in that race for it the same way. That request names only the race and the question, never your words, and what it finds is shared with every voter; if nothing usable turns up, the concern is shown as searched and still not scored.

Before searching for a candidate's site, we check links on relevant pages already found for other candidates. Each site must still pass the same checks for the candidate and race. When we check sources again, we reuse readings of unchanged pages when the candidate, issue definitions and checking rules are unchanged, and read changed pages again.

Reported statements are labeled and count for less than direct evidence. Candidate directories and editable encyclopedias are used last and labeled unverified. Official ballot lists may establish biographical facts, such as occupation or hometown, but never a policy position. We do not substitute a party platform for a candidate's views.

Candidate pages include a contents menu with links to individual issues, funding, research sources and any published voting record. They list verified sites, social accounts linked from those sites, every page we read and links to official records. You can open the search record, including what was left out. Candidates with nothing to compare are checked again every few hours until the election. After election day you can still compare candidates on a ballot that has been held, and the page says so. From then on the comparison is for signed-in voters, whose answers are kept for their next election; races nobody researched before election day are researched only when a signed-in voter asks. When research is requested from a profile, it covers every candidate in that race equally, and no voter information is sent or kept with the request.

Information is labeled “official” when it comes directly from an election-office feed or page, and “reconstructed” when assembled from official sources. Missing information is identified rather than filled in.

Where AI fits

AI helps find and read sources, extract candidate statements, ask questions in plain language, read what you type about your concerns and write explanations. Code keeps a reading of your words only when it quotes them exactly, and only links them to questions on your ballot. Code checks quotes against stored pages and rejects explanation sentences without valid citations. Each sentence of an explanation must say which of your answers it is about and rest only on claims about that question, so a cited biography cannot stand in for a policy reason; a quality such as local roots is labeled as a quality. Measure explanations give both sides equal space and do not recommend a vote.

The interviewer is not shown candidate positions. A model links specific office questions to broad policy questions once, including which direction their scales run, and code validates those links. Code then calculates candidate positions and scores. Models never score candidates or decide your match.

Your privacy

  • Your street address is discarded. We use it to find your voting districts and keep only the district identifiers. Address and contact details never appear in a link.
  • Location stays on your device. We remember your district identifiers for about six months. Uncheck “Remember my location” to keep them for this visit only. “Change location” clears them.
  • Your concerns stay with your answers. What you type about what matters to you travels to our server only in the body of a request, never in a link, and is not logged. It is kept in this browser, and on your account only if you sign in.
  • Signing in is optional. Match answers stay in this browser until you start over. If you sign in, answers are saved to your account for future elections. You can delete your account and its answers at any time.
  • We do not store your email address. Sign-in uses an emailed code. We keep a protected fingerprint of the address to recognize your account, and use the address only to send sign-in codes.
  • A suggested state is only an estimate. Without a saved location, we may use the rough location your network reports to suggest elections. That estimate is not saved or placed in a link. The location you enter takes precedence.
  • Page views help us improve the site. Vercel Web Analytics records pages viewed, referral information, country, device and browser type. It uses no cookies, and our analytics filter excludes URLs that contain an address or contact details.

Get another perspective

The homepage includes a prompt you can copy into your own AI chat. It asks that AI to verify every page of your ballot, including amendments and selection limits, then help you discover overlooked concerns before comparing candidates. It asks for the same research standards for everyone, sourced arguments on both sides of measures and a final checklist recording choices you make. The full prompt is available to read before you copy it.

The prompt asks the other AI to keep missing evidence visible and distinguish tentative choices from decisions you confirm. If that service cannot provide personalized recommendations, it asks it to continue with neutral research and comparisons. Those instructions describe the external conversation; they do not extend the app's ballot coverage or matching features.

Copying includes no saved location, answers or account information, and sends nothing to another AI. You choose what to paste and share with that service. Its research, conclusions and handling of your information are separate from VotingVibe's. The prompt asks for fair treatment and a bias audit; it cannot guarantee an unbiased or accurate answer.

The results page also offers a detailed research download: the ballot on file, including measures and selection limits; published candidate claims and quotations; source links and dates; voting records and finance where available; and your current comparisons and recorded choices. Interview answers and priorities are included only if you check the option. Saved voter location and account details are omitted. The file is prepared for download in your browser and is not sent to another AI. It marks unavailable research and does not certify that the ballot on file is complete. Candidate research is fetched at download time and may be newer than your displayed results.

Coverage is still growing

Some areas have full ballots and polling-place information; others have only partial coverage. Local races may depend on your exact address. We show these gaps and link to official sources so you can check before you vote.