AI Debate: 12 Practice Prompts and What Rules Allow

In brief

ArguFight is the free AI debate tool with a judge: its core AI debate is free and an AI judge scores each round 0 to 100, as of November 2026. Any chatbot works too if you tell it to hold one side. The NSDA says generative AI should not be cited as a source.

How do you debate an AI?

Paste this into Claude, ChatGPT, Gemini or Whizi and fill the brackets. It turns a chatbot from an agreeable helper into an opponent that holds its ground.

Two matching podiums facing each other on a black stage, one holding a terracotta Claude starburst and the other a brushed silver OpenAI knot, with a small rainbow glass Gemini star on a plinth between them as the judge, in a charcoal studio under white rim light.

You are my debate opponent. Format: [Lincoln-Douglas, Public Forum, policy or Model UN]. Resolution: [exact motion]. I take [affirmative or negative]. You take the other side for the whole session and don't concede unless I win a point on the merits. Give one speech at a time, no more than [word count] words, and wait for my reply. Cite only sources you could name and date, and mark any claim you can't source as unsourced. After round [number], judge us both against [rubric, such as clash, evidence and clarity] and name the one argument I dropped.

Each bracket closes a gap. The word limit stops the model reading you an essay. The sourcing line matters most, because a model will invent a statistic to win a point if you let it.

What does a good AI debate look like? Three worked motions

Each example below was written for this page to show what to ask and what a strong answer contains. None is a tested output from a named model.

  1. Lincoln-Douglas, values. Prompt: "Be my negative for the resolution that civil disobedience is justified in a democracy. Give a 6 minute constructive with one value and one criterion, then stop." A good answer names a value such as the rule of law, gives a criterion that measures it, and has two contentions that each tie back to that criterion. If it only lists reasons with no framework, send it back.
  1. Public Forum, evidence. Prompt: "Be the con side on the motion that schools should replace homework with in class practice. Give a 4 minute constructive with two contentions, a source for each, and one weighing line." A good answer ends with a weighing line, such as why its impact is larger or likelier than yours. A pile of facts with no comparison isn't a case.
  1. Model UN, a delegate stance. Prompt: "Act as the delegate of [country] in committee on [topic]. Give a 90 second opening speech in that country's stated policy, then answer two points of information from me." A good answer stays in the country's known position and doesn't invent treaty clauses. Open the real country statement before you rely on any line of it.

In all three, the useful moment is your reply. Answer the strongest point first, then say aloud why yours matters more.

Can you use AI in a debate round?

Not during the round, and not as a source. The NSDA Competition Rules Board voted 7 to 0 on 12 September 2023 that in debate events "generative AI should not be cited as a source," while it "may be used to guide students to articles, ideas, and sources." The ruling adds that "the original source of any quoted or paraphrased evidence must be available if requested." Read the board's minutes and then your tournament's own rules, because a league or a tournament can set stricter ones.

The NCFCA is stricter about the round itself. Coach Christy Shipe's January 2026 post says "Accessing AI for any purpose during a debate round would violate NCFCA Debate Rule A4." She says an AI works as a brainstorming tool, and its list is "a starting point for research." She adds that students shouldn't use it to generate evidence. She says she isn't speaking for the league, so check its rulebook.

The NSDA's policy debate page adds that internet access is permitted in that event unless the tournament says otherwise. That's about research tools, not permission to quote a chatbot. Practice with AI before the round, and bring real sources to it.

Which AI debate tool is best, and which is free?

ArguFight is the one with a stated free core and a scored verdict. The table lists what each tool says on its own page as of 22 November 2026. Where a page was silent, the cell says so.

ToolFree useSign upPaid priceWhat it does
ArguFight"Free to start", on web and mobile; core AI debate "always free"Account, no credit cardOptional extras such as fact checking use coins; coin price not statedAI opponent, AI judge scores 0 to 100, coach feedback, human matches and ELO rankings
SuperDebate"Start debating free" button; whether the AI coach costs anything is not statedSign up link on the pageNot statedAI coach Aurelius with warm up rounds, drills and a scorecard; live events from $20
FlowHuntFree trial, scope not statedTrial needs no accountNot statedGeneral AI agent platform with a debate chatbot; the page states no limits or price
Champion PressFree classroom activity, dated 25 October 2023None statedBooks sold separatelyTeacher led activity using ChatGPT arguments
ChatGPTFree plan listed at $0Not statedGo $8, Plus $20 a monthOpen ended opponent from your prompt
ClaudeFree, "limited usage" every 5 hours and every week; search the webNot statedPro $20 a month, or $17 billed annuallyLong, careful speeches from your prompt
Whizi10 messages a day on Whizi AI, Auto and Claude Haiku, no cardAccount neededPro $29.99 a monthTwo models argue one motion side by side on Pro and Powerhouse

Whizi loses to ArguFight on the one thing a debater wants most, a built in judge and a ranked ladder of human opponents. It also loses to any tool that needs no account for a first round. A chatbot with your own prompt beats a template tool on format, because you set the speech order and the rubric yourself.

The pages are not equal. SuperDebate's home page lists the coach's drills and scorecard but no AI price or limit. FlowHunt's debate chatbot page offers a free trial with no account but states no limit or price. ArguFight's debate AI page says the core AI debate is always free, with optional extras on coins. Anthropic's pricing page lists Free at $0 with "limited usage" and Pro at $17 a month billed annually.

How do you watch two AIs debate each other?

Give each model one side of the same motion, in separate chats, and paste each speech into the other. That works in any two chat apps, and it takes about a minute a speech.

On Whizi's Pro and Powerhouse plans, Compare puts two models on one prompt in two columns, each with its own conversation, so follow ups stay independent. Web search is on by default in a chat, so each side can look up evidence. Use this one.

Motion: [resolution]. Give a 250 word constructive for the [affirmative or negative] with two contentions and a source for each. Then I will paste in the other side's speech, and you rebut it in 150 words.

Send it with the affirmative to one model and the negative to the other, then paste each rebuttal across. Read where they disagree on a fact. That's where one of them is guessing, and how to fact check AI answers shows the two minute test.

What are the best AI debate prompts for students?

Twelve prompts cover the jobs students ask for, grouped by use. Fill the brackets and paste.

Topics and cases

  1. Topic list: "Give me 15 debate motions on [theme] for [grade] students. For each, say which side is harder to argue and why, in one line."
  1. Case outline: "Outline an affirmative case for [resolution] in [format]: framework, two contentions, an impact for each. Mark every claim that needs a real source, and don't invent one."
  1. Both sides: "Give the strongest argument for and against [motion] in three sentences each. Then say which side has more to prove."

Opponents and drills

  1. Opponent: "Argue the [side] of [motion] against me. One argument at a time, ending with a question I must answer. Don't concede early."
  1. Cross examination: "Play the other side in cross examination for [format]. Ask me one question at a time, press on my last answer, and after five questions tell me which answer was weakest."
  1. Rebuttal drill: "Here is an argument: [paste]. Give me the three strongest responses, ranked, then ask me to write the best one in 60 words."
  1. Fallacy hunt: "Read this speech: [paste]. Name any fallacies by type, quote the sentence for each, and say how an opponent would use it."

Judging and review

  1. Judge my round: "Here is my speech: [paste]. Score it out of 10 for clash, evidence and clarity using this rubric: [rubric]. Name my best moment and my biggest gap."
  1. Weighing: "Compare these two impacts: [A] and [B]. Write the weighing line a judge would accept, using magnitude, probability and timeframe."
  1. Flow notes: "Summarise this round as a flow: each argument, each response, each drop. Here are the speeches: [paste]."

Model UN and evidence

  1. Country stance: "What is [country]'s stated position on [topic]? Use only its official statements, quote a line with its date, and say which parts you couldn't source."
  1. Evidence check: "Here is a card I plan to read: [paste]. List what to verify at the original source: author, date, whether the quote is in context, and whether my tag claims more than the text does."

Prompt 12 is the safe use of AI on evidence. You still read the original, as the NSDA ruling requires.

What mistakes ruin an AI debate practice session?

Four mistakes cause most wasted sessions, and each has a quick fix.

  1. The AI agrees with you by round two. The fix: add "don't concede unless I win the point on the merits" and restate it every few rounds.
  2. Invented evidence. The fix: ask for sources you could name and date, then open them. The NCFCA post says to double check that the citations are correct.
  3. A wall of text. The fix: set a word limit that matches the speaking time, and ask for one speech at a time.
  4. Typing instead of speaking. The fix: say each reply aloud before you type it, because the round is spoken.

A fifth is subtler. A model that judges your round is grading against its own taste. Hand it a rubric, and ask a second model to grade the same speech.

Where does Whizi fit for AI debate?

Whizi is one app for debate practice across Claude, GPT and Gemini, instead of a pile of single purpose tools. It suits you if you want two models arguing one motion or a second opinion on your judge.

The free plan gives you 10 messages a day on Whizi AI, Auto and Claude Haiku, with no card. Pro carries Claude, GPT and Gemini in one app, and its first week is $1.99. Side by side Compare is a Pro and Powerhouse feature, and it's how you run the two model debate above. Web search is on by default in a chat.

You can upload a PDF or Word file, such as a case you wrote or a reading, and ask for a flow or a critique. Scanned PDFs get a vision transcription pass. See Whizi pricing for what each plan holds. A dedicated tool like ArguFight is still better if you want a ranked ladder and a built in judge.

What can an AI debate partner not do?

It can't be the source in your case, and it can't tell you what your league allows. A model predicts likely text, and a likely statistic isn't a true one. Why AI gets things wrong explains the mechanism.

It also can't replace a coach who has watched you speak, or a real opponent who surprises you. For essay style feedback on a written case, the AI essay grader page covers what a rubric catches. Check your league's AI rule and your tournament's rule before you rely on any tool in a round.

Workflow checklist
  • Name the format, the exact resolution and your side in every opponent prompt
  • Tell the AI not to concede unless you win the point
  • Set a word limit that matches the speaking time
  • Open the original source behind every card before you read it
  • Check your league and tournament rules on AI before a round
Common questions

Frequently asked questions

Is there a free AI to debate with?

Yes. ArguFight says it is free to start with no credit card, and ChatGPT and Claude both list free plans, as of November 2026. A free chatbot works as an opponent if you give it the motion, your side and a no concession rule.

Can AI generate debate evidence?

It can point you to ideas and sources, but it should not be your source. The NSDA ruling says generative AI should not be cited as a source and that the original source must be available if requested. Open the original and read it in context.

Is it cheating to use AI to prepare for a debate?

Not by itself. The NSDA ruling allows AI to guide students to articles, ideas and sources, and the NCFCA coach calls it a brainstorming tool. Using it during a round breaks NCFCA Rule A4, so check your own league and tournament rules.

Can AI judge a debate round?

It can score a round against a rubric you give it, and ArguFight uses an AI judge scoring 0 to 100. Treat the score as practice feedback, not a prediction of a real judge, and ask a second model to grade the same speech.