AI vs AI debate. On a real clock.
Pick two AI brains. Give them a motion. They take opposite sides and argue out a real round in proper competitive format. POIs at the right windows, extensions where the format requires, a judge ballot at the end. Free during beta. No card asked.
Six brains, fifteen matchups
One of the things that separates a real AI vs AI debate from a chatbot doing dialogue: brain choice on each side. Each model has its own house style as a debater. Set the matchup deliberately and you can flow a different round on the same motion every time.
Case-depth vs sharp rebuttal. The most-run matchup. Claude builds the framework, GPT closes hard on the back end.
Run this matchup →Rhetorical depth vs cheap-reasoner discipline. DeepSeek pressures the warrant, Claude defends with structure.
Run this matchup →Current events vs contrarian framing. Good on motions where recent sourcing matters and the round needs an edge.
Run this matchup →Character-rich prose (Nous Hermes 4) vs framework. Open Lab takes the unusual side; Claude steadies the round.
Run this matchup →Sharp vs sharper. Both lean rebuttal-heavy. Often the highest-clash round on the exhibition page.
Run this matchup →Technical reasoner vs character-driven argumentation. Quiet matchup, often the most surprising verdict.
Run this matchup →How to make two AIs debate each other
Open the exhibition
Hit /exhibition. Free to start, no card asked. Five rounds before the soft cap; signing in with Google gets you to ten.
Pick a motion
Type any debate topic, or pick one from the 150+ motion library. Real tournament motions are tagged by format and difficulty.
Pick the format
Asian Parliamentary, British Parliamentary, World Schools, Public Forum, or Lincoln-Douglas. Format steers speech timing, structure, and citation rules.
Pick two brains
One brain for PROP, one for OPP. Six brains, fifteen matchups. Different brains debate differently on the same motion.
Watch and read the ballot
Speeches generate turn by turn. An AI judge writes the reason-for-decision at the end with winner, margin, central clash, and one drill.
Which formats does the AI vs AI round support?
- Asian Parliamentary · 3-on-3, definitions matter, WSDC pipeline
- British Parliamentary (BP) · 4 teams, openings + closings, extensions
- World Schools (WSDC) · 3-on-3, POIs, reply speeches
- Public Forum (PF) · 2-on-2, monthly topics, lay-flow audience split
- Lincoln-Douglas (LD) · 1-on-1, value/criterion, trad or circuit
- Quick Clash · 3-minute casual round, good for stress-testing motions
The voice block for each format is server-side, which means the two AIs aren't doing "Harvard debate society" generic English. APDA stays impromptu with no fabricated citations. Policy-style cards are kept out of formats that don't use them. BP extensions actually extend the prior half. Try a round and flow it.
Why run AI vs AI?
Prep against a side you don't get assigned often. If you're an LD circuit kid stuck on neg every weekend, run an AI vs AI aff on the topic and flow the strongest case the AI can build. Bring the best arguments into your next round.
See how a different brain attacks your case. Build your case, set Claude as PROP running it, set GPT or Grok as OPP. Watch where the AI cross-ex finds your weakest joint. That's the joint your real opponent finds in round.
Pre-tournament motion review. Set ten brain pairings on the same motion, watch the central clashes converge or diverge. The clashes that come up across every matchup are the ones you actually need to prep for.
Pure entertainment. Claude vs Grok on "THBT billionaires should not exist" is the kind of round you watch for the rhetoric, not because you need to prep. That's also fine.
What's different about the AI argumentation here
The AI on every round 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 (admin-weighted by rating) 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. The learning loop runs every night at 04:00 UTC.
Each brain is also adversarial by default — pinned to a side, instructed to defend it. The AI does not hedge when pressed, 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. Everything is debatable. We built it that way.
AI vs AI debate FAQ
Can two AIs actually debate each other?
Yes. The exhibition mode at /exhibition pins one AI brain to PROP and another to OPP, then walks them through the actual speech order of the format you pick. They take points of information at the right windows, build extensions where the format requires them, and refuse to break their pinned side. An AI judge writes the ballot at the end.
Which AI is the best debater?
Depends on the round, not the brand. Claude tends to build cases with structural depth and writes ballots that read like a varsity judge. GPT closes hard in rebuttals. Gemini wins on current-events motions where recent sourcing matters. Grok runs sharper cross-ex. DeepSeek anchors the judge panel because its register is technical without house style. Open Lab (Nous Hermes 4) produces character-driven prose. Pick by motion shape, not by brand.
Can I run Claude vs GPT?
Yes. Any two of the six brains. Claude vs GPT is the most-run matchup; Claude vs DeepSeek and Claude vs Grok are common for stress-testing case construction against different attack styles. Run Claude vs GPT directly.
Is AI vs AI debate free?
Yes. Anonymous visitors get 5 rounds before the soft cap. Signing in with Google bumps that to 10 total. All tiers are currently $0 during the Debate AI beta; the post-beta reference pricing lives at /pricing.
How long does an AI vs AI round take to generate?
Most APDA / Quick Clash exhibitions finish in 45-90 seconds end to end, including the ballot. BP / WSDC with 4 brains and full speech timing runs 90-180 seconds. Faster than reading a longform argument in real time.
Can I watch instead of debating?
Yes, that's literally what the exhibition mode is for. /exhibition is watch-only. If you want to debate one of the brains yourself, go to /voice-debate (spoken) or /debate-ai (typed). Same engine, you're in the round instead of watching.
Is there an API to run AI vs AI debates programmatically?
Not yet. The exhibition orchestrator is web-only for now. If you need batch AI vs AI rounds for research, email the team and describe the use case.