AI Judge · Scores out of 100

It listens. It judges.

Set your phone on the table and hit listen. It follows the argument and returns the result, the reason, a score out of 100, and specific feedback for each person.

  1. 1 Hit listen. Debate the round.
  2. 2 Hit stop when it's over.
  3. 3 Read the decision.

This tool uses the AI brain you pick and names the exact model on the decision. Ranked rounds use a published panel of three brains: two matching votes decide it, and a split records no winner. How the judge is kept honest Open to critiques: try to break it

Two private judged rounds per account are free. After that, Individual ($10/year) includes private judging. This allowance is shared with private live rooms.

Listen to the round
Put your phone on the table and debate. Nothing to fill in first.
or
Reading the round…
Finding the key disagreements and weighing both sides.
You vs the AI judge
Your call
AI judge

Decision · Winner

The short version

Why they won

Debate a real person Run a live round instead Argue a rematch

Think the judge got it wrong? Try to break the judge, and what breaks gets published →

For your first argument and the next one

The decision explains what each side argued, what decided the round, and what to work on next. It tracks claims, answers, unanswered points, and the comparisons that mattered. Each person gets clear, specific feedback to take into the next round.

Two candidates at podiums on a televised presidential debate stage
A result tells you who won. A useful decision shows you why and what to change next.

Explains what decided it

See the key clashes, which arguments survived, and why one side won. The decision uses plain language and names the answers or missing comparisons that changed the result.

Gives both people feedback

Each person gets a score out of 100, a strength to keep, a specific fix, and a habit to carry into the next round.

One casual 1v1

One person supports the claim and one opposes it. This recorder judges a casual one-on-one round.

How it works

One. Hit listen and set the phone down where it can hear both people. It transcribes the round as it goes. Two. Hit stop and add names if you want them on the ballot. Three. You get a winner, a score out of 100 for each person, a reason for the decision, and concrete feedback. If you would rather not record, you can paste a transcript instead.

What gets scored

What the judge is listening for

The criteria are written down before you speak, they are the same for both sides, and you can read them yourself. Six things get a number. Together they produce one argument score out of 100. The verdict itself comes from the arguments, resolved in a fixed order of force.

Clarity

1 to 10

Structure, signposting, and whether the speech could be followed the first time.

Reasoning

1 to 10

The reasons behind each claim, and whether the steps connect. A claim with nothing behind it is an assertion, not an argument.

Responsiveness

1 to 10

Answers to what the other side actually said, not to a friendlier version of it.

Weighing

1 to 10

Impact comparison, and whether you told the judge which issue the round turns on.

Strategy

1 to 10

Choices about which arguments to press, answer, narrow, or leave behind.

Persuasion

1 to 10

Whether the case moved a reasonable listener. Concrete stakes, a world you can picture.

The headline score is a weighted mean of the six axes, normalized to 100. Reasoning carries the most weight, followed by responsiveness and weighing, then clarity, strategy, and persuasion. Neither score decides the winner.

Interruptions and pressure

Getting cut off is not the score. What you did next is.

The judge tracks whether a point survived the response. A later speaker has to beat the answer that was actually given, not repeat the original claim with more confidence. A point you never came back to counts as dropped, and dropped points decide rounds.

In a live voice round, a speech that got cut off is marked as interrupted in the transcript the judge reads. So holding the thread under pressure shows up in the decision as responsiveness instead of vanishing into the crosstalk.

Out of bounds

It does not score how you sound.

Delivery, fluency, accent, speaking speed, volume, polish, and confidence are not scored at all. That fence is deliberate, it is published with the rest of the rubric, and it is enforced in the code rather than trusted to the prompt.

A judge that quietly rewards polish is a judge that marks a second-language debater down for their second language. Speak the way you speak. The decision is about the argument.

AI debate judge FAQ

Is there an AI that can judge a debate round?

Yes. Hit listen and it hears a casual one-on-one round through your device's microphone, then returns the result, a score out of 100 for both speakers, a clear reason for decision, and specific feedback. You can also paste a transcript.

What kind of round does it judge?

This recorder uses a casual 1v1 structure. One person supports the question, one person opposes it, and the stronger argument wins.

How accurate is the AI judge?

It runs the same adjudication engine as our live rounds, with explicit weighing across magnitude, probability, timeframe, and reversibility, plus a speaker score calibrated to use the full range. Treat it as a sharp second opinion. It catches drops and weighing gaps that a casual listener can miss.

Which AI brains are used?

A transcript submitted on this page is read by the one brain selected in Round options, and the decision names its provider and exact model. Ranked async rounds and server-written live ballots use the season's council: Anthropic Claude, OpenAI GPT, and Google Gemini. Each gets the same transcript and rubric, then votes independently. The live roster shows the exact pinned model IDs and whether every seat is available.

Where should I put the phone?

Somewhere it can hear both sides. A table between the two speakers works; so does a laptop at the front of the room. It handles normal debate delivery, including fast speaking, but it cannot judge what it cannot hear, so keep it off the far side of a large room and away from an air vent. It shows you the words as it picks them up, so you can tell within a few seconds whether it is hearing the round.

Do I have to fill anything in?

No. The motion and who is on which side can be worked out from the round itself. Use the optional setup if you want the motion quoted word for word or need to add neutral background.

Does it judge how I sound?

No. Delivery, fluency, accent, speaking speed, volume, and confidence are not scored at all, and that limit is published with the rest of the criteria on the integrity page. What does get scored is clarity, meaning whether the speech could be followed the first time, which is about structure and signposting rather than polish. Persuasion is scored on its own axis, sits last in the weighing order, and can only break a tie the arguments left level.

What happens if I get interrupted or talked over?

Being cut off costs you nothing by itself. The judge scores what happened to the argument: whether the point survived the response, whether you beat the answer that was actually given rather than repeating your original claim, and whether anything you dropped ever came back. In a live voice round a turn that got cut off is marked as interrupted in the transcript the judge reads, so holding the thread under pressure shows up as responsiveness. The same applies to any interruption or back-and-forth.

What if I think the judge is wrong?

Two routes. On a recorded round either debater can appeal within 72 hours, and a person decides it, never the model that made the call. Beyond one round, the judge itself is open to critiques: the strongest objections to AI judging are published with our answers, and if you can build a round where the panel rewards the worse argument, send it in. The strongest critiques get run against the live panel and the results are published whether they hold or not.

Does it replace a person?

No. It gives both sides a consistent second read and explains what decided the round. It is useful when nobody else was there to listen.

Is it free?

Sign in for two free private judged rounds per account, shared with private live rooms. After that, Individual ($10/year), BYOK, Voice, and Team include private judging. Compare plans.

How casual 1v1 works The published rubric, and how the judge is kept honest Open to critiques: try to break the judge How accurate is the judge? The benchmark Judge a Zoom or Twitch room Voice debate AI Typed debate AI Argue with an AI Things to argue about Learn to make a stronger argument About Debatable