Debate an AI · Free in beta · Voice or typed

Debate an AI. On a real clock.

Pick a motion. Take a side. The AI argues back, takes points of information, interrupts when the warrant is weak, and writes a judge ballot when the round ends. Format-accurate across ten competitive debate formats. Voice or typed. Free while in beta. No card asked.

Voice · Spoken round

Debate an AI out loud

Real-time spoken round over WebRTC. The AI listens, interrupts, takes POIs, and writes the ballot when the clock runs out. Sub-300ms latency on a good connection. Sounds like a debater, not a voice assistant.

Format-aware across APDA, BP, WSDC, Asian Parli, PF, LD, Policy, Congress, MUN, Quick Clash.
Start voice round
Typed · Async-friendly

Debate an AI in writing

Typed round if you want to drill a specific speech, prep a case, or work without a mic. Same format-aware AI, same judge ballot at the end. Pick which brain runs the round: Claude, GPT, Gemini, Grok, DeepSeek, or Open Lab.

Six AI brains, selectable per round. BYOK Claude available.
Open typed mode

What this AI is actually good at

An AI you can debate against is not the same thing as a general assistant you ask to play a debate game. The AI on every round here is wrapped in a format-aware system prompt with per-format speech timing, citation rules, register, judging criteria, and known failure modes. Then exemplar past rounds and a nightly distillation pass over top-rated rounds get prepended on top, so the AI on motion X today is meaningfully different from the AI on motion X last month.

How to debate an AI in 60 seconds

1

Pick a motion

Use a real tournament motion from your format, hit shuffle for a random one, or type your own. The AI sees the same prompt you do.

2

Pick a side

Government or opposition, proposition or negative, depending on the format. You argue your side, the AI argues against you. Switch sides and run the same motion again to drill both.

3

Argue

Voice or typed. Take points of information, give them, push back on the warrant. Round ends and the AI judge writes the reason-for-decision with speaker points and one drill for next round.

Six AI brains, selectable per round

One of the things that separates an AI you can actually debate from a single-model chatbot: brain choice. Each model has its own house style as a debater. Default is Claude for case construction depth; DeepSeek anchors the judging panel so the verdict has no single-model bias.

Claude
Case depth, framework rigor, judge ballots
GPT
Sharp rebuttals, broad knowledge
Gemini
Current-events motions, recent sourcing
Grok
Aggressive cross-ex, contrarian framing
DeepSeek
Cheap reasoner, technical, judging-mode default
Open Lab
Character-rich prose via Nous Hermes 4

Why this isn't the same as asking a general assistant to debate you

Most general-purpose AI assistants are tuned for cooperative dialogue. Pressed on a position, they hedge or concede. They will play a "debate game" if asked, but they collapse every format into the same generic English: a Public Forum summary and a British Parliamentary closing extension come out the same way. An AI built to debate is different in three concrete ways.

Format-pinned register. Policy gets tag-cite-body cards. LD gets value and criterion. BP gets extensions that actually extend the prior half. APDA stays impromptu with no fabricated citations. The format awareness is in the system prompt, not the user prompt.

Adversarial by default. The AI is pinned to a side and instructed to defend it. It does not hedge when you press, does not concede when the warrant holds, does not "see both sides" mid-round. It collapses to its strongest voting issues in the rebuttal and writes the ballot in its own favor when you give it the chance.

Judge ballot in the format's actual register. Each format has its own speaker scale, weighing axes, and adjudication norms. The judge mode walks them in order. Multi-brain consensus fans the same round to Claude, GPT, and DeepSeek in parallel so the verdict does not lean on one model's house style. See DebateAI vs ChatGPT and Best AI for debate practice for honest side-by-sides.

Debate an AI FAQ

Can I really debate an AI?

Yes. /voice-debate opens a real-time spoken round. /debate-ai opens a typed round. Both put you on a clock, give you a real motion, and run the format's actual speech order. The AI argues back, takes points of information, interrupts when relevant, and writes a judge ballot at the end. Free while in beta. No card asked.

Is there an AI you can debate for free?

This one. Anonymous visitors get 5 rounds before the soft cap. Signing in with Google bumps that to 10 total. Every tier is currently $0 while DebateAI.com is in beta; the post-beta plans on /pricing are reference only.

What does debating an AI actually look like?

You see a motion ("THBT social media platforms should be liable for the content their algorithms amplify"). You pick your side. The AI opens with a real constructive in your format's structure. You respond. The AI takes a point of information mid-speech, you take one back. Round runs the format's full speech order: 4 to 8 speeches depending on the format. AI judge writes the reason-for-decision with speaker points when the round ends.

Which debate format does the AI know?

Ten of them. APDA, British Parliamentary, Asian Parliamentary, World Schools (WSDC), Public Forum, Lincoln-Douglas, Policy, Student Congress, Model UN, and a Quick Clash casual mode. Each has its own speech timing, citation rules, register, and judging criteria baked in.

Can the AI take points of information?

Yes. In formats that have POIs (BP, WSDC, Asian Parli, APDA), the AI offers and accepts them at the correct windows. It refuses cleanly when it should, answers in 10-15 seconds, then returns to its own speech. POI mechanics are format-specific and tuned per format.

Does the AI interrupt me?

In voice rounds yes, when set to. Server-side voice activity detection means the AI can cut in if you stall or leave a warrant hanging. Off by default for new users; toggle on once you want the pressure.

How is debating an AI different from debating a human?

The AI is available right now, runs at the level you set, never no-shows, and writes the ballot when the round ends. A human is a real human. Both are worth doing. If you want a human, /spar matches you with one on video; /debate-chat pairs you in seconds.

What is the best AI for debate practice?

Depends on the job. Research and case drafting: a general assistant like Claude or ChatGPT. Live format-aware sparring with a real clock and judge ballot: this one. Honest category guide here.

Can I debate the AI in a language other than English?

Yes. The language picker on the landing page sets the AI's output language across 14 options including Hindi, Spanish, French, Mandarin, Arabic, Portuguese, and Bahasa Indonesia. The AI replies in your language; the TTS synthesizes voice in the matching language too.

Who built this AI to debate?

A national APDA champion at UChicago, with the voice-guideline bank that drives every AI opponent authored from inside the formats. The AI's voice block knows the Sofia 2026 WUDC manual changes, India's 2025 WSDC win, the NDT 2026 Arctic-topic meta, the LD trad-vs-circuit split, and the PF 1st-vs-2nd speaker asymmetry. More on the story.