diff --git a/docs/testing.md b/docs/testing.md
index 16b7eaa0..d817cfa0 100644
--- a/docs/testing.md
+++ b/docs/testing.md
@@ -154,14 +154,15 @@ npm run verify
The gate runs, in order:
1. documentation checks
-2. lint
-3. type checking
-4. unit tests
-5. collaboration service tests
-6. native Edge Function tests
-7. production build
-8. production service-worker offline smoke
-9. the desktop Chromium `@smoke` set (`npm run verify:full` adds the full desktop and mobile suite)
+2. the guide jargon check (`npm run seo:jargon`): no generated guide in `public/` other than `/lincoln-douglas-debate` and `/policy-debate` may use framework, contention, warrant or impact calculus more than once in its visible text or its JSON-LD ([`scripts/check_jargon.py`](../scripts/check_jargon.py))
+3. lint
+4. type checking
+5. unit tests
+6. collaboration service tests
+7. native Edge Function tests
+8. production build
+9. production service-worker offline smoke
+10. the desktop Chromium `@smoke` set (`npm run verify:full` adds the full desktop and mobile suite)
It does not start local Supabase, run pgTAP, run WebKit, or use protected staging.
diff --git a/package.json b/package.json
index e30b4bd9..778cae64 100644
--- a/package.json
+++ b/package.json
@@ -23,6 +23,7 @@
"preview": "vite preview",
"smoke:jev": "node scripts/smoke-jev.mjs",
"seo:indexnow": "python3 scripts/indexnow_submit.py",
+ "seo:jargon": "python3 scripts/check_jargon.py --max 1",
"test": "vitest run",
"test:collaboration": "vitest run --config services/collaboration/vitest.config.ts",
"start:collaboration": "tsx services/collaboration/src/index.ts",
@@ -35,7 +36,7 @@
"test:sw": "node scripts/test-service-worker-offline.mjs",
"test:e2e:staging": "playwright test --project=staging",
"test:agentic": "bash .claude/skills/verify-debatelab/bin/verify.sh e2e run",
- "verify": "npm run docs:check && npm run lint && npm run typecheck && npm test && npm run test:collaboration && npm run test:edge && npm run build && npm run test:sw && P=$(bash .superset/e2e-port.sh) && CI=1 PORT=$P PLAYWRIGHT_BASE_URL=http://127.0.0.1:$P npm run test:e2e:smoke",
+ "verify": "npm run docs:check && npm run seo:jargon && npm run lint && npm run typecheck && npm test && npm run test:collaboration && npm run test:edge && npm run build && npm run test:sw && P=$(bash .superset/e2e-port.sh) && CI=1 PORT=$P PLAYWRIGHT_BASE_URL=http://127.0.0.1:$P npm run test:e2e:smoke",
"verify:full": "npm run verify && P=$(bash .superset/e2e-port.sh) && CI=1 PORT=$P PLAYWRIGHT_BASE_URL=http://127.0.0.1:$P npm run test:e2e"
},
"dependencies": {
diff --git a/public/ai-debate-coach/index.html b/public/ai-debate-coach/index.html
index 64da0dca..a97e92e2 100644
--- a/public/ai-debate-coach/index.html
+++ b/public/ai-debate-coach/index.html
@@ -43,7 +43,7 @@
Can an AI debate coach actually help you prepare?
Yes. DebateOwl's Ask coach is an AI debate coach that answers specific prep questions about what your side has to prove, which argument decides the round and how to structure a speech, and gives feedback on practice speeches; it works best when you name the format, side and speaker role. Vague questions get generic answers. Treat any sources it cites as leads to check. An AI coach has not watched you speak and does not know your judges or your partner, so take its conclusions to your human coach or team.
What is an AI debate coach?
A chat tool that answers a debater's prep questions about strategy, arguments, answers to the other side and speech structure, and gives feedback on practice speeches, for the debater to test and adapt.
Strategy help
Ask which argument decides the round, what your side has to prove and where to spend your speaking time, with answers that use your debate's motion and sources.
Speech structure
Get an outline for a constructive, rebuttal, summary or reply speech: the order of points and how long to give each one.
Practice feedback
Paste a speech or a flow and ask what was unclear, under-warranted or dropped, then pick one fix to drill.
-
What an AI debate coach is for
A good human coach rarely gives you an answer first. They ask what the round is about, which side you are on, what you have to prove, and what the judge will be comparing at the end. The AI coach in DebateOwl works best the same way: as a partner you think out loud with, not a machine that writes speeches.
Ask is a coaching chat that streams its answer as it writes, and it already knows your Current Debate: the motion and the Library sources you have marked active. That context is the main difference from a general chatbot. You do not have to paste the motion and your research into every question, and the advice stays tied to one debate instead of drifting between topics.
How to ask a coaching question that gets a useful answer
Vague questions get generic answers. Compare these two:
Weak: How do I win this debate?
Strong: I am second opposition in a WSDC round on this motion. Our first speaker argued the policy is unworkable. What should I add so I am not repeating her, and which proposition argument should I spend most of my rebuttal on?
The strong version names the format, the speaker role, what has already been said and the decision you actually have to make. A reliable template is: format, side and speaker, what is already on the table, what you are stuck on, and what kind of answer you want (a list, an outline or a single recommendation).
Five coaching conversations worth having before a tournament
Burden check: ask what each side has to prove for the motion to stand or fall, and where the line between them sits. Many rounds are lost on a burden nobody named.
Clash map: ask for the three places where the cases will meet. Plan your speeches around those, not around your own contentions.
Weighing rehearsal: give the coach your impact and the likely opposing impact and ask for the comparison in two sentences. Repeat until you can say it without notes.
Speech audit: paste a practice speech and ask which claim had no warrant, which point ran too long, and what a judge would remember from it.
Opponent prediction: ask how a strong team on the other side would frame the debate, then ask what your first response to that framing should be.
Worked example: coaching a weak rebuttal
Motion: This house believes social media companies should be legally liable for misinformation on their platforms
Student's rebuttal: Their argument is wrong because free speech is important and this would destroy it.
Coach's diagnosis: The response asserts a value but never connects it to the proposition's mechanism. It does not say how liability leads to less speech, so the judge has nothing to weigh.
Coach's question back: What would a platform rationally do if it could be sued for every false post?
Rebuilt rebuttal: Liability makes every borderline post a legal risk. A platform facing that risk removes anything it cannot verify quickly, which includes breaking news, whistleblowers and minority viewpoints. So the proposition does not just remove misinformation; it removes true speech that is hard to check, and the people who lose most are those with no other platform.
Follow-up drill: Deliver the rebuilt response in forty-five seconds, then ask the coach for the strongest proposition answer to it and prepare a single-sentence reply.
Notice that the coach did not write the rebuttal outright. It asked the question that exposed the missing link, which is the habit you want to build for rounds where no coach is available.
Coaching by format
World Schools (WSDC)
Ask the coach to check your team split: what the first and second speakers each own, and whether the third speaker can rebut without introducing a new argument. For the reply speech, ask for a structure built around the two or three questions the debate turned on, told from your side, without new material.
British Parliamentary (BP)
Role fulfilment decides BP rooms. Ask the coach whether your member speech is a true extension or a restatement of your opening bench, and ask what a whip speech should do with the five other speeches already in the room: group them into clashes, not summaries of each speaker.
Public Forum (PF)
Use coaching to plan the summary before the round. Ask which two issues you can realistically extend from constructive through final focus, and what the second rebuttal must frontline so those issues survive. Also ask for crossfire questions that set up your summary rather than score points.
Lincoln-Douglas (LD)
LD coaching is often framework coaching. Ask the coach whether your criterion actually measures your value, how your contentions link back to the criterion, and when it is safer to concede the opponent's framework and show that you win under it anyway.
Getting answers with sources
When a coaching answer depends on facts, turn on Citations in Ask. The coach then answers with sources you can open. Treat those as leads: read the original, confirm it says what the answer claims, and only then use it in a speech. Adding your own trusted articles to the Library has the same effect from the other direction, because active sources that fit the question are used in its answers for that debate.
What an AI coach cannot do
It has not watched you speak, it does not know the judges at your tournament, and it cannot tell you how your partner works under pressure. Bring the conclusions from AI coaching sessions to your human coach or team, and let them correct the advice that does not fit the people in the room.
+
What an AI debate coach is for
A good human coach rarely gives you an answer first. They ask what the round is about, which side you are on, what you have to prove, and what the judge will be comparing at the end. The AI coach in DebateOwl works best the same way: as a partner you think out loud with, not a machine that writes speeches.
Ask is a coaching chat that streams its answer as it writes, and it already knows your Current Debate: the motion and the Library sources you have marked active. That context is the main difference from a general chatbot. You do not have to paste the motion and your research into every question, and the advice stays tied to one debate instead of drifting between topics.
How to ask a coaching question that gets a useful answer
Vague questions get generic answers. Compare these two:
Weak: How do I win this debate?
Strong: I am second opposition in a WSDC round on this motion. Our first speaker argued the policy is unworkable. What should I add so I am not repeating her, and which proposition argument should I spend most of my rebuttal on?
The strong version names the format, the speaker role, what has already been said and the decision you actually have to make. A reliable template is: format, side and speaker, what is already on the table, what you are stuck on, and what kind of answer you want (a list, an outline or a single recommendation).
Five coaching conversations worth having before a tournament
Burden check: ask what each side has to prove for the motion to stand or fall, and where the line between them sits. Many rounds are lost on a burden nobody named.
Clash map: ask for the three places where the cases will meet. Plan your speeches around those, not around your own arguments.
Weighing rehearsal: give the coach your impact and the likely opposing impact and ask for the comparison in two sentences. Repeat until you can say it without notes.
Speech audit: paste a practice speech and ask which point never said why it is true, which point ran too long, and what a judge would remember from it.
Opponent prediction: ask how a strong team on the other side would frame the debate, then ask what your first response to that framing should be.
Worked example: coaching a weak rebuttal
Motion: This house believes social media companies should be legally liable for misinformation on their platforms
Student's rebuttal: Their argument is wrong because free speech is important and this would destroy it.
Coach's diagnosis: The response asserts a value but never connects it to the proposition's mechanism. It does not say how liability leads to less speech, so the judge has nothing to weigh.
Coach's question back: What would a platform rationally do if it could be sued for every false post?
Rebuilt rebuttal: Liability makes every borderline post a legal risk. A platform facing that risk removes anything it cannot verify quickly, which includes breaking news, whistleblowers and minority viewpoints. So the proposition does not just remove misinformation; it removes true speech that is hard to check, and the people who lose most are those with no other platform.
Follow-up drill: Deliver the rebuilt response in forty-five seconds, then ask the coach for the strongest proposition answer to it and prepare a single-sentence reply.
Notice that the coach did not write the rebuttal outright. It asked the question that exposed the missing link, which is the habit you want to build for rounds where no coach is available.
Coaching by format
World Schools (WSDC)
Ask the coach to check your team split: what the first and second speakers each own, and whether the third speaker can rebut without introducing a new argument. For the reply speech, ask for a structure built around the two or three questions the debate turned on, told from your side, without new material.
British Parliamentary (BP)
Role fulfilment decides BP rooms. Ask the coach whether your member speech is a true extension or a restatement of your opening bench, and ask what a whip speech should do with the five other speeches already in the room: group them into clashes, not summaries of each speaker.
Public Forum (PF)
Use coaching to plan the summary before the round. Ask which two issues you can realistically extend from constructive through final focus, and what the second rebuttal must frontline so those issues survive. Also ask for crossfire questions that set up your summary rather than score points.
Lincoln-Douglas (LD)
In LD, much of the coaching is about how the judge should decide the round: your value and criterion. Ask the coach whether your criterion actually measures your value, how each argument ties back to the criterion, and when it is safer to accept the opponent's value and criterion (their framework) and show that you win under them anyway.
Getting answers with sources
When a coaching answer depends on facts, turn on Citations in Ask. The coach then answers with sources you can open. Treat those as leads: read the original, confirm it says what the answer claims, and only then use it in a speech. Adding your own trusted articles to the Library has the same effect from the other direction, because active sources that fit the question are used in its answers for that debate.
What an AI coach cannot do
It has not watched you speak, it does not know the judges at your tournament, and it cannot tell you how your partner works under pressure. Bring the conclusions from AI coaching sessions to your human coach or team, and let them correct the advice that does not fit the people in the room.
Frequently asked questions
What can I ask the AI debate coach?
Anything you would ask a coach between rounds: what your side has to prove, which argument decides the round, speech structure, how to answer a specific argument, or why a practice speech did not land. Specific questions with format, side and speaker role get the best answers.
Does the coach know my case?
It knows your Current Debate: the motion, and the Library sources you have marked active. Paste a speech or a case section into the question when you want feedback on that exact text.
Can the AI coach cite sources?
Yes. Turn on Citations in Ask and the answer includes sources you can open. Check each one against the original before using it in a round.
Will the coach write my speeches for me?
It can draft an outline or a passage, but you will speak better from structure you understand. Use it to find what is missing, then write the speech in your own words.
How is this different from a general AI chatbot?
The coach works inside one debate at a time, so it keeps your motion, debate settings and sources in context, and it is set up for debate questions rather than essays.
Is there a free AI debate coach?
Not a permanent one in DebateOwl. The Ask coach comes with the app's 3-day free trial, no credit card needed, and after that with Pro at $5/month + tax. The tools you can use free without an account, the debate timer, motion generator, flow sheet and speech time calculator, do not include coaching.
Can an AI judge my debate speech?
Partly. Sparring rounds end with a judged verdict that names the reason for the decision, and the Ask coach can tell you what was unclear, under-warranted or dropped in a speech you paste in. It has not watched you speak and does not know the judges at your tournament, so treat its verdict as practice feedback, not a prediction of a real ballot.
Can practising debate against an AI actually make you better?
Yes, if it shortens the loop of giving a speech, finding what failed, fixing one thing and giving it again. A useful session needs a motion, a side, real speech times, an opponent that responds to what you said and a verdict that names why you won or lost. DebateOwl's Sparring runs School 4-6-6-3, Lincoln-Douglas and Public Forum rounds this way. AI practice adds repetitions between team practices but does not replace them: judge adaptation, delivery under pressure and strategic judgement still develop fastest with a coach and real rounds.
Practice against AI opponents
Run a full Sparring round in School 4-6-6-3, Lincoln-Douglas or Public Forum, type or speak each speech, and finish with a judged verdict.
Turn feedback into drills
Take the weakest point from a verdict or a Crossfire debrief, rebuild it with Attack or Ask, and run that speech again.
Keep private prep out of search
Cases, sources, rounds and history stay inside your Current Debate behind sign-in. Only these public guides are indexed.
-
What a useful AI practice session looks like
Most debaters improve through repetitions with feedback: give a speech, find out what failed, fix one thing, give it again. AI practice is valuable when it shortens that loop, not when it hands you a finished speech to read.
A chat window can produce arguments on request, but it does not hold a round together. It forgets which side you are on, does not time you, and rarely tells you which of your points the judge would have dropped. Practice needs a motion, a side, a format with real speech times, an opponent that responds to what you actually said, and a verdict that names the reason you lost or won.
DebateOwl is built around that loop. You start a Current Debate for one motion, and every case, source, round and debrief you create stays filed under it. That means the AI opponent in your third practice round argues the same motion you wrote your case for, and your History shows how the debate developed across sessions.
A 45-minute practice session, step by step
This plan fits a weeknight. It assumes you have a motion but no case yet.
Start a Current Debate with the motion written exactly as it will be announced (2 minutes). Add any articles or notes you already trust to the Library so later answers are grounded in them.
Run Brainstorm on the motion and pick the two or three angles you can explain in one sentence each (5 minutes). Discard angles that need evidence you do not have.
Generate a case for your side (10 minutes). Read its "If they say…" answers before your own arguments: they tell you what you will have to answer.
Run a Sparring round in your format (15 minutes). Give every speech in full, at time, without pausing to edit. Voice mode lets you speak instead of type.
Read the verdict and write down the single reason it gives for the decision (3 minutes).
Rebuild only that part (5 minutes): ask Attack to stress-test the rewritten argument, or regenerate the one case section that failed.
Give the rebuilt speech again against the timer (5 minutes). Compare it with the first attempt.
The discipline is in step six. Fixing one problem per session sticks; rewriting the whole case after every round mostly produces new problems.
Worked example: one practice round
Motion: This house would ban smartphones in secondary schools
Side and format: Proposition, run as a School 4-6-6-3 Sparring round.
First argument: Phones fragment attention in class, and attention is the input every other kind of learning depends on.
AI opponent's response: Schools already confiscate phones during lessons, so a ban changes only break times, where the harm you described does not happen.
What the verdict flagged: Proposition never explained what a ban adds over existing classroom rules, so the opposition's mitigation stood unanswered.
The drill: Ask Attack to find the gap in the argument, then rewrite it around breaks: social exclusion, cyberbullying during the school day, and the cost of teachers enforcing rules lesson by lesson.
Second attempt: The R1 speech now opens with the comparison: classroom rules depend on every teacher enforcing them every hour, while a ban sets one rule at the school gate. The opponent's mitigation becomes a point in your favour.
The example shows the pattern to look for. The first version was not wrong; it was incomplete in a way only a responsive opponent exposes. That is the gap AI practice fills between team practices.
Format-specific practice tips
World Schools (WSDC)
WSDC rounds reward a consistent team line, so practise as a team even when you use AI. Agree the line in one sentence, then have each speaker run their own section against the AI and check that the line survives. Points of Information matter in every substantive speech: use Crossfire with the AI asking the questions to rehearse taking a POI, answering in under fifteen seconds, and returning to your structure. Sparring does not run a WSDC round, so time the eight-minute substantives and four-minute reply yourself.
British Parliamentary (BP)
In BP, two teams on each bench compete with each other as well as with the other side. Practise both halves. For opening, generate a case with the Current Debate's format set to British Parliamentary and drill a clear definition and model in the first two minutes. For closing, take an opening case you have already generated and practise finding an extension it does not contain, then deliver it in seven minutes without repeating opening material.
Public Forum (PF)
Sparring runs Public Forum with four-minute constructives and rebuttals, three-minute summaries and two-minute final focuses. Use the round to practise the hardest skill in PF: keeping summary and final focus consistent. Before each summary, write the two issues you will collapse on, and check after the round whether your final focus extended exactly those issues and nothing new.
Lincoln-Douglas (LD)
The Lincoln-Douglas round follows the standard times: six-minute affirmative constructive, seven-minute negative constructive, four-minute first affirmative rebuttal, six-minute negative rebuttal and three-minute second affirmative rebuttal. The 1AR is where most affirmatives lose. Run it repeatedly against the same negative constructive and practise grouping responses and cross-applying your framework, so four minutes covers seven.
How to tell whether practice is working
Keep a short log for each session on paper or in a document. Track the same four things every time:
Arguments you dropped: anything the opponent said that you never answered. This number should fall first.
Time use: whether you finished each speech within ten seconds of the limit without rushing the last point.
Weighing: whether your final speech compared impacts explicitly, or just repeated them.
Verdict reasons: whether the judge's reason for the decision changes from session to session, or keeps naming the same weakness.
If the same verdict reason appears three sessions in a row, stop running full rounds and drill that one skill in isolation until it changes.
AI vs traditional prep
AI and traditional preparation are good at different parts of the work. AI is fast at structure; reading, coaching and real rounds are where judgement comes from. Most strong debaters use both and give each the tasks it does well.
Task
AI does it well
Traditional prep does it better
First case outline
Drafts an opening, arguments with evidence and "If they say…" answers in minutes
Choosing which outline fits your team and your circuit
Rebuttal blocks
Lists likely arguments and first-draft answers
Testing answers against opponents you have actually met
Research
Suggests evidence leads and search directions
Reading the original source, checking dates and context
Practice rounds
A responsive opponent at any hour, at real speech times
Unpredictable opponents, panels and live crossfire
Feedback
A consistent verdict that names a reason
A coach who knows your habits and the judges you will face
A simple split works for most teams: use AI to get from a blank page to a first draft and for extra practice volume, then spend coaching time and team practice on strategy, delivery and the arguments the AI draft got wrong. Never read AI-suggested evidence in round until you have opened the original source yourself.
Where AI practice stops being enough
An AI opponent is consistent and always available, which is exactly what makes it a good training partner and a limited judge. It will not reproduce the judge who dislikes speed, the panel that splits, or the opponent who asks a question you never anticipated. Use AI rounds for volume between real practices, and save team practices and coached rounds for adaptation, delivery and the feedback only an experienced person can give.
+
What a useful AI practice session looks like
Most debaters improve through repetitions with feedback: give a speech, find out what failed, fix one thing, give it again. AI practice is valuable when it shortens that loop, not when it hands you a finished speech to read.
A chat window can produce arguments on request, but it does not hold a round together. It forgets which side you are on, does not time you, and rarely tells you which of your points the judge would have dropped. Practice needs a motion, a side, a format with real speech times, an opponent that responds to what you actually said, and a verdict that names the reason you lost or won.
DebateOwl is built around that loop. You start a Current Debate for one motion, and every case, source, round and debrief you create stays filed under it. That means the AI opponent in your third practice round argues the same motion you wrote your case for, and your History shows how the debate developed across sessions.
A 45-minute practice session, step by step
This plan fits a weeknight. It assumes you have a motion but no case yet.
Start a Current Debate with the motion written exactly as it will be announced (2 minutes). Add any articles or notes you already trust to the Library so later answers are grounded in them.
Run Brainstorm on the motion and pick the two or three angles you can explain in one sentence each (5 minutes). Discard angles that need evidence you do not have.
Generate a case for your side (10 minutes). Read its "If they say…" answers before your own arguments: they tell you what you will have to answer.
Run a Sparring round in your format (15 minutes). Give every speech in full, at time, without pausing to edit. Voice mode lets you speak instead of type.
Read the verdict and write down the single reason it gives for the decision (3 minutes).
Rebuild only that part (5 minutes): ask Attack to stress-test the rewritten argument, or regenerate the one case section that failed.
Give the rebuilt speech again against the timer (5 minutes). Compare it with the first attempt.
The discipline is in step six. Fixing one problem per session sticks; rewriting the whole case after every round mostly produces new problems.
Worked example: one practice round
Motion: This house would ban smartphones in secondary schools
Side and format: Proposition, run as a School 4-6-6-3 Sparring round.
First argument: Phones fragment attention in class, and attention is the input every other kind of learning depends on.
AI opponent's response: Schools already confiscate phones during lessons, so a ban changes only break times, where the harm you described does not happen.
What the verdict flagged: Proposition never explained what a ban adds over existing classroom rules, so the opposition's mitigation stood unanswered.
The drill: Ask Attack to find the gap in the argument, then rewrite it around breaks: social exclusion, cyberbullying during the school day, and the cost of teachers enforcing rules lesson by lesson.
Second attempt: The R1 speech now opens with the comparison: classroom rules depend on every teacher enforcing them every hour, while a ban sets one rule at the school gate. The opponent's mitigation becomes a point in your favour.
The example shows the pattern to look for. The first version was not wrong; it was incomplete in a way only a responsive opponent exposes. That is the gap AI practice fills between team practices.
Format-specific practice tips
World Schools (WSDC)
WSDC rounds reward a consistent team line, so practise as a team even when you use AI. Agree the line in one sentence, then have each speaker run their own section against the AI and check that the line survives. Points of Information matter in every substantive speech: use Crossfire with the AI asking the questions to rehearse taking a POI, answering in under fifteen seconds, and returning to your structure. Sparring does not run a WSDC round, so time the eight-minute substantives and four-minute reply yourself.
British Parliamentary (BP)
In BP, two teams on each bench compete with each other as well as with the other side. Practise both halves. For opening, generate a case with the Current Debate's format set to British Parliamentary and drill a clear definition and model in the first two minutes. For closing, take an opening case you have already generated and practise finding an extension it does not contain, then deliver it in seven minutes without repeating opening material.
Public Forum (PF)
Sparring runs Public Forum with four-minute constructives and rebuttals, three-minute summaries and two-minute final focuses. Use the round to practise the hardest skill in PF: keeping summary and final focus consistent. Before each summary, write the two issues you will collapse on, and check after the round whether your final focus extended exactly those issues and nothing new.
Lincoln-Douglas (LD)
The Lincoln-Douglas round follows the standard times: six-minute affirmative constructive, seven-minute negative constructive, four-minute first affirmative rebuttal, six-minute negative rebuttal and three-minute second affirmative rebuttal. The 1AR is where most affirmatives lose. Run it repeatedly against the same negative constructive and group similar responses and use the same test for how the judge should decide (your framework) across several arguments. That helps four minutes cover seven.
How to tell whether practice is working
Keep a short log for each session on paper or in a document. Track the same four things every time:
Arguments you dropped: anything the opponent said that you never answered. This number should fall first.
Time use: whether you finished each speech within ten seconds of the limit without rushing the last point.
Weighing: whether your final speech compared impacts explicitly, or just repeated them.
Verdict reasons: whether the judge's reason for the decision changes from session to session, or keeps naming the same weakness.
If the same verdict reason appears three sessions in a row, stop running full rounds and drill that one skill in isolation until it changes.
AI vs traditional prep
AI and traditional preparation are good at different parts of the work. AI is fast at structure; reading, coaching and real rounds are where judgement comes from. Most strong debaters use both and give each the tasks it does well.
Task
AI does it well
Traditional prep does it better
First case outline
Drafts an opening, arguments with evidence and "If they say…" answers in minutes
Choosing which outline fits your team and your circuit
Rebuttal blocks
Lists likely arguments and first-draft answers
Testing answers against opponents you have actually met
Research
Suggests evidence leads and search directions
Reading the original source, checking dates and context
Practice rounds
A responsive opponent at any hour, at real speech times
Unpredictable opponents, panels and live crossfire
Feedback
A consistent verdict that names a reason
A coach who knows your habits and the judges you will face
A simple split works for most teams: use AI to get from a blank page to a first draft and for extra practice volume, then spend coaching time and team practice on strategy, delivery and the arguments the AI draft got wrong. Never read AI-suggested evidence in round until you have opened the original source yourself.
Where AI practice stops being enough
An AI opponent is consistent and always available, which is exactly what makes it a good training partner and a limited judge. It will not reproduce the judge who dislikes speed, the panel that splits, or the opponent who asks a question you never anticipated. Use AI rounds for volume between real practices, and save team practices and coached rounds for adaptation, delivery and the feedback only an experienced person can give.
Frequently asked questions
Which formats can I spar against the AI in?
Sparring runs School 4-6-6-3, Lincoln-Douglas and Public Forum rounds with their speech order and times. For WSDC, British Parliamentary and other formats you can still generate cases, get coaching and drill rebuttals, and time speeches yourself.
Can I give my practice speeches out loud?
Yes. Sparring and Crossfire can run as spoken rounds in the browser, so you can practise delivery as well as content.
How often should I practise against an AI opponent?
Two or three focused sessions a week usually beats one long one. Each session should end with one specific fix that you drill before the next round.
Does AI practice replace a coach or real rounds?
No. It adds repetitions between team practices. Judge adaptation, delivery under pressure and strategic judgement still develop fastest with people.
Is AI debate prep better than traditional prep?
Neither is better on its own. AI is faster at outlines, block drafts and evidence leads; reading sources, coaching and real rounds build the judgement AI drafts lack.
Are my practice rounds private?
Yes. Rounds, verdicts and debriefs are saved to your account inside the Current Debate they belong to, behind sign-in. Search engines only see these public guides.
Is an AI rebuttal generator worth using for debate prep?
It is worth it for preparation when the output is structured for a round: the exact words being attacked, the broken link, a response you can say, a cross-ex question and the missing weighing. A general chatbot tends to write a counter-essay instead, which is rarely usable. Run it on opposing cases and on your own case. It cannot hear how your opponent framed an argument or know what a judge will find persuasive, and most leagues restrict AI during rounds. Rewrite responses in your own words and check any evidence it cites.
Weaknesses ranked by damage
Attack quotes the exact words it is targeting, names the broken link and orders the weaknesses from critical to minor.
Responses you can say
Each weakness comes with one to three sentences written to be spoken in round, not a description of what you could argue.
Cross-ex questions and weighing
Get questions tied to each weakness, a check for fallacies and an analysis of the weighing the argument is missing.
-
What a good rebuttal generator should give you
Generic AI tools answer the request to rebut an argument by writing a counter-essay: a few paragraphs on why the other side is wrong. That is rarely usable. In a round you need to know which part of the argument to attack, what the flaw is called, and what to say in the next forty seconds.
Attack in DebateOwl is built around that need. Paste an argument or a full case and it returns a structured critique. Every weakness has to do three jobs: quote the exact words it attacks, explain why the link fails and why that matters for the ballot, and give a response you can say word for word. If the argument has fewer real weaknesses, it returns fewer rather than inventing problems.
What one Attack analysis contains
An overall assessment: a score out of 100, a short summary and what the argument does well, so you know which parts not to waste time on.
Weaknesses, usually three to six, each marked critical, major or minor and labelled as evidential, logical or structural.
For each weakness: the quoted words, why the link is weak (missing warrant, no evidence, no link, non-unique, turned or outweighed), and a response to say in round.
Fallacies detected, with where they appear and how to call them out without sounding like a textbook.
Cross-examination questions, each tied to one weakness and written so that a short honest answer exposes the gap.
A weighing analysis: the weighing the argument relies on, the comparisons it never makes, and how to out-weigh it.
Attack also runs web research on the argument, so responses can point to counter-evidence. Treat any evidence it mentions as a lead and check the original source before you use it.
Settings that change the output
Attack intensity
Mild gives constructive criticism, useful when you are checking your own case. Moderate balances strengths and weaknesses. Ruthless looks hardest for anything a strong opponent could exploit, which is what you want when preparing against an opposing case.
Judge paradigm
Logic-Heavy prioritises broken reasoning and missing warrants. Values-Focused looks at framework and principle, which suits Lincoln-Douglas and principled motions. Policy/Impact concentrates on links, probability and the size of impacts, which suits policy-style motions and Public Forum.
Fallacies and weighing
Both checks are on by default. Turn fallacy detection off when you only want substantive responses; keep weighing analysis on almost always, because unweighed rebuttal is the most common reason good responses do not win rounds.
Worked example: one argument through Attack
Motion: This house would tax sugary drinks
Argument pasted: A sugar tax reduces consumption, because higher prices make people buy less. Lower consumption means less obesity, which saves lives and reduces the burden on public health systems.
Critical weakness, logical: Quoted: lower consumption means less obesity. The argument assumes people who buy fewer sugary drinks do not replace those calories with other foods. Without that warrant, the impact does not follow from the mechanism.
Response to say: Even if people buy fewer fizzy drinks, they can get the same sugar from sweets, juice and snacks that are not taxed. Cutting one source of calories is not the same as cutting calories.
Major weakness, evidential: Quoted: saves lives. The argument gives no estimate of how much consumption falls or how that translates into health outcomes, so the judge cannot weigh the size of the benefit.
Cross-ex question: If someone stops buying a sugary drink but buys a chocolate bar instead, has the tax reduced their sugar intake?
Weighing gap: The argument never compares its health benefit with the cost to low-income households, who spend a larger share of their income on these drinks. That comparison is your weighing.
The useful part is not that the analysis disagrees with the argument. It identifies the one assumption the argument cannot survive without, gives you a sentence to attack it, and sets up a cross-ex question that makes the opponent concede it before your speech.
From generated output to a rebuttal you own
Start from the right input. Use a claim from the "If they say…" answers in a case you generated, an argument you heard in a previous round, or your own case when you want to find its weak points first.
Read the critical weaknesses and pick one. In a short speech, one well-explained weakness beats four rushed ones.
Rewrite the suggested response in your own words. Keep the structure (quote, flaw, response, why it matters) and cut anything you would not naturally say.
Check any evidence it relies on against the original source.
Time the response. Most single responses should fit in forty-five seconds.
Import the cross-examination questions into Crossfire and practise asking them against the AI, which answers from the opposing side.
Keep the analysis in your Current Debate so the block sits next to the case it answers.
Using generated rebuttals by format
Public Forum (PF)
Use Policy/Impact for most PF resolutions. Prepare the response and the cross-ex question together, because PF crossfire is where you get the opponent to commit to the assumption you attack in rebuttal.
Lincoln-Douglas (LD)
Run Values-Focused on the negative case to find where its contentions fail to link to its own criterion. That gap often lets you win under their framework as well as yours.
World Schools and British Parliamentary
Motions are often released shortly before the round, so use Attack in practice on likely arguments for motion areas you expect, and build a bank of responses to the reasoning that recurs across motions, such as slippery slopes, unproven behaviour change and ignored alternatives.
What it cannot do
Attack responds to the text you give it. It cannot hear how your opponent actually framed an argument, and it cannot tell you what a particular judge will find persuasive. Most leagues also do not allow internet or AI tools during rounds, so it is a preparation tool: build and practise the blocks before the tournament, then adapt them by ear in the round.
+
What a good rebuttal generator should give you
Generic AI tools answer the request to rebut an argument by writing a counter-essay: a few paragraphs on why the other side is wrong. That is rarely usable. In a round you need to know which part of the argument to attack, what the flaw is called, and what to say in the next forty seconds.
Attack in DebateOwl is built around that need. Paste an argument or a full case and it returns a structured critique. Every weakness has to do three jobs: quote the exact words it attacks, explain why the link fails and why that matters for the ballot, and give a response you can say word for word. If the argument has fewer real weaknesses, it returns fewer rather than inventing problems.
What one Attack analysis contains
An overall assessment: a score out of 100, a short summary and what the argument does well, so you know which parts not to waste time on.
Weaknesses, usually three to six, each marked critical, major or minor and labelled as evidential, logical or structural.
For each weakness: the quoted words, why the link is weak (no reason it is true, no evidence, no link, not unique to their side, turned or outweighed), and a response to say in round.
Fallacies detected, with where they appear and how to call them out without sounding like a textbook.
Cross-examination questions, each tied to one weakness and written so that a short honest answer exposes the gap.
A weighing analysis: the weighing the argument relies on, the comparisons it never makes, and how to out-weigh it.
Attack also runs web research on the argument, so responses can point to counter-evidence. Treat any evidence it mentions as a lead and check the original source before you use it.
Settings that change the output
Attack intensity
Mild gives constructive criticism, useful when you are checking your own case. Moderate balances strengths and weaknesses. Ruthless looks hardest for anything a strong opponent could exploit, which is what you want when preparing against an opposing case.
Judge paradigm
Logic-Heavy prioritises broken reasoning and points that never say why they are true. Values-Focused looks at principles and how the judge should decide, which suits Lincoln-Douglas and principled motions. Policy/Impact concentrates on links, probability and how big the harms and benefits are, which suits policy-style motions and Public Forum.
Fallacies and weighing
Both checks are on by default. Turn fallacy detection off when you only want substantive responses; keep weighing analysis on almost always, because unweighed rebuttal is the most common reason good responses do not win rounds.
Worked example: one argument through Attack
Motion: This house would tax sugary drinks
Argument pasted: A sugar tax reduces consumption, because higher prices make people buy less. Lower consumption means less obesity, which saves lives and reduces the burden on public health systems.
Critical weakness, logical: Quoted: lower consumption means less obesity. The argument assumes people who buy fewer sugary drinks do not replace those calories with other foods. Without that step, the health benefit does not follow from the mechanism.
Response to say: Even if people buy fewer fizzy drinks, they can get the same sugar from sweets, juice and snacks that are not taxed. Cutting one source of calories is not the same as cutting calories.
Major weakness, evidential: Quoted: saves lives. The argument gives no estimate of how much consumption falls or how that translates into health outcomes, so the judge cannot weigh the size of the benefit.
Cross-ex question: If someone stops buying a sugary drink but buys a chocolate bar instead, has the tax reduced their sugar intake?
Weighing gap: The argument never compares its health benefit with the cost to low-income households, who spend a larger share of their income on these drinks. That comparison is your weighing.
The useful part is not that the analysis disagrees with the argument. It identifies the one assumption the argument cannot survive without, gives you a sentence to attack it, and sets up a cross-ex question that makes the opponent concede it before your speech.
From generated output to a rebuttal you own
Start from the right input. Use a claim from the "If they say…" answers in a case you generated, an argument you heard in a previous round, or your own case when you want to find its weak points first.
Read the critical weaknesses and pick one. In a short speech, one well-explained weakness beats four rushed ones.
Rewrite the suggested response in your own words. Keep the structure (quote, flaw, response, why it matters) and cut anything you would not naturally say.
Check any evidence it relies on against the original source.
Time the response. Most single responses should fit in forty-five seconds.
Import the cross-examination questions into Crossfire and practise asking them against the AI, which answers from the opposing side.
Keep the analysis in your Current Debate so the block sits next to the case it answers.
Using generated rebuttals by format
Public Forum (PF)
Use Policy/Impact for most PF resolutions. Prepare the response and the cross-ex question together, because PF crossfire is where you get the opponent to commit to the assumption you attack in rebuttal.
Lincoln-Douglas (LD)
Run Values-Focused on the negative case to find where its arguments fail to tie back to its own criterion. That gap often lets you win under their value and criterion as well as yours.
World Schools and British Parliamentary
Motions are often released shortly before the round, so use Attack in practice on likely arguments for motion areas you expect, and build a bank of responses to the reasoning that recurs across motions, such as slippery slopes, unproven behaviour change and ignored alternatives.
What it cannot do
Attack responds to the text you give it. It cannot hear how your opponent actually framed an argument, and it cannot tell you what a particular judge will find persuasive. Most leagues also do not allow internet or AI tools during rounds, so it is a preparation tool: build and practise the blocks before the tournament, then adapt them by ear in the round.
Frequently asked questions
What does the AI rebuttal generator produce?
A ranked list of weaknesses, each quoting the attacked words with an explanation and a response to say in round, plus detected fallacies, cross-examination questions and a weighing analysis.
Can I use it on my own case?
Yes, and you should. Set the intensity to Mild or Moderate to find your weak points before an opponent does, then fix them or prepare an answer.
Is it allowed to use an AI rebuttal generator in a debate round?
Most leagues restrict or ban internet and AI use during rounds. Check your league's rules and use the tool to prepare blocks beforehand.
Are the rebuttals backed by evidence?
Attack runs web research and can point to counter-evidence. Treat anything it cites as a lead and check the original source before using it.
How is this different from asking a chatbot to rebut an argument?
The output is structured for a round: each response targets quoted words, names the broken link and comes with a cross-ex question and weighing, instead of a general essay against the other side.
It depends on the job. For brainstorming or learning a topic, a general chatbot such as ChatGPT is flexible. For motion-by-motion prep plus timed practice rounds that end in a verdict, a dedicated platform such as DebateOwl fits. As of October 2026, debatelab.ai, a separate product, says it is free and runs World Schools and Asian Parliamentary practice, and WinDebate adds round replay and progress tracking from $19.99/month. Give each tool the same motion and tasks before you decide.
Feature comparison
What each kind of tool does well: general chatbots, research assistants, dedicated debate platforms, and timers or flowing tools.
Format coverage
Which formats a tool actually models, from Public Forum and Lincoln-Douglas to WSDC and British Parliamentary, and where you still prep by hand.
Workflow differences
Whether cases, sources, practice rounds and feedback stay connected to one motion, or end up scattered across chats and documents.
-
The four kinds of AI debate tool
Debaters use AI tools for different jobs, and the categories overlap less than the marketing suggests. Most prep setups combine at least two of these.
General AI chatbots
Assistants such as ChatGPT, Claude and Gemini are flexible and good at brainstorming, explaining a concept or tightening a paragraph. Their weakness for debate is structure: they do not hold a motion, side and format across sessions unless you rebuild the context each time, and they will produce confident claims without sources unless you ask.
Research and citation tools
AI search and research assistants are strongest at finding sources quickly. They help with evidence but not with turning evidence into arguments, rebuttals or speeches.
Dedicated debate platforms
These are built around debate concepts: cases with contentions, rebuttals, opponents and verdicts. They vary a great deal in which formats they model and whether they let you practise or only generate text.
Practice utilities
Timers, flowing templates and argument maps. Often simple, sometimes not AI-based at all, and still essential for practice that matches tournament conditions.
Eight criteria for comparing AI debate tools
Structure: does the output separate claim, warrant and impact, or is it a wall of text?
Format awareness: does it know your speech order, times and roles, or treat every debate as an essay?
Sources: can you see where a claim comes from, and can you add your own material for it to use?
Opposition: does it predict and argue the other side, or only help with yours?
Practice: can you deliver a speech and get a response to it, not just read generated text?
Feedback: does it tell you why an argument failed, with something specific to fix?
Continuity: does work stay organised by motion, or do you lose it in a long chat history?
Limits and privacy: what is the daily or monthly allowance, and who can see your prep?
Weight the criteria by what you need. A novice benefits most from structure and feedback; an experienced competitor often cares most about sources, opposition and practice.
A one-hour test you can run on any tool
Reviews and feature lists only go so far. The fairest comparison is to give every tool the same motion and the same tasks:
Pick a motion in your format that you know well, so you can judge the quality of the answers.
Ask each tool for a case on one side. Score it on structure and on whether the warrants actually prove the claims.
Ask each tool for the strongest argument against that case. Score whether it found the real weak point or a generic objection.
Paste one of your own past speeches and ask what it would change. Score how specific the feedback is.
Ask for evidence on one factual claim. Open every source and score how many check out.
Note how much context you had to repeat between steps.
Fifteen minutes per tool is enough to see the difference. Keep the scores; they are more useful than any ranking list, including this one.
Worked example: scoring answers to one motion
Motion: This house would make voting compulsory
Task: Find the weakest link in the proposition argument that compulsory voting produces a more representative government.
Strong answer: Points out that turnout and representation are different: compelled voters who know little about the candidates may vote randomly or for the first name on the ballot, so higher turnout may not mean preferences are better represented.
Weak answer: Says compulsory voting violates freedom. That is an opposition argument, not an attack on the proposition's link, and it does not engage with representation at all.
How to score it: Give the point only if the tool attacked the warrant (why more voters means better representation) rather than changing the subject.
Running the same task through several tools shows quickly which ones understand debate structure and which ones produce plausible but generic text.
Which tools fit which format
World Schools (WSDC)
WSDC prep is team prep under time pressure, often for impromptu motions. Prioritise fast brainstorming and motion analysis, plus a way to practise Points of Information. Few tools model the full WSDC round, so expect to time speeches yourself.
British Parliamentary (BP)
BP rewards role knowledge and extensions. Look for tools that let you set the format and your team position, and that can show you what the opening half is likely to say so closing teams can plan something new.
Public Forum (PF)
PF depends on evidence and on keeping summary and final focus consistent. Strong source handling matters more here than in most formats, along with practice rounds that use real PF speech times.
Lincoln-Douglas (LD)
LD needs tools that handle value and criterion, not just policy arguments. Test whether a tool can defend a framework and link contentions to it, and whether it can practise the time-pressured 1AR.
Where DebateOwl fits
DebateOwl is a dedicated debate platform that keeps every case, source, round and debrief inside one Current Debate. It covers case generation with "If they say…" answers to the other side, Library sources that ground its answers, Attack for finding weak links, Brainstorm, Factoids for checking claims, an Ask coaching chat with optional citations, Crossfire and full Sparring rounds with a verdict, a speech timer and an argument flowchart.
Its limits are worth stating plainly. Sparring currently runs School 4-6-6-3, Lincoln-Douglas and Public Forum rounds, so WSDC and BP practice uses the prep and coaching tools rather than a full simulated round. Like every AI tool, its evidence leads need checking against the original source. Run the one-hour test above and decide on your own results.
+
The four kinds of AI debate tool
Debaters use AI tools for different jobs, and the categories overlap less than the marketing suggests. Most prep setups combine at least two of these.
General AI chatbots
Assistants such as ChatGPT, Claude and Gemini are flexible and good at brainstorming, explaining a concept or tightening a paragraph. Their weakness for debate is structure: they do not hold a motion, side and format across sessions unless you rebuild the context each time, and they will produce confident claims without sources unless you ask.
Research and citation tools
AI search and research assistants are strongest at finding sources quickly. They help with evidence but not with turning evidence into arguments, rebuttals or speeches.
Dedicated debate platforms
These are built around debate concepts: cases with arguments, rebuttals, opponents and verdicts. They vary a great deal in which formats they model and whether they let you practise or only generate text.
Practice utilities
Timers, flowing templates and argument maps. Often simple, sometimes not AI-based at all, and still essential for practice that matches tournament conditions.
Eight criteria for comparing AI debate tools
Structure: does each argument say what is true, why, and why it matters, or is it a wall of text?
Format awareness: does it know your speech order, times and roles, or treat every debate as an essay?
Sources: can you see where a claim comes from, and can you add your own material for it to use?
Opposition: does it predict and argue the other side, or only help with yours?
Practice: can you deliver a speech and get a response to it, not just read generated text?
Feedback: does it tell you why an argument failed, with something specific to fix?
Continuity: does work stay organised by motion, or do you lose it in a long chat history?
Limits and privacy: what is the daily or monthly allowance, and who can see your prep?
Weight the criteria by what you need. A novice benefits most from structure and feedback; an experienced competitor often cares most about sources, opposition and practice.
A one-hour test you can run on any tool
Reviews and feature lists only go so far. The fairest comparison is to give every tool the same motion and the same tasks:
Pick a motion in your format that you know well, so you can judge the quality of the answers.
Ask each tool for a case on one side. Score it on structure and on whether the reasons it gives actually prove its points.
Ask each tool for the strongest argument against that case. Score whether it found the real weak point or a generic objection.
Paste one of your own past speeches and ask what it would change. Score how specific the feedback is.
Ask for evidence on one factual claim. Open every source and score how many check out.
Note how much context you had to repeat between steps.
Fifteen minutes per tool is enough to see the difference. Keep the scores; they are more useful than any ranking list, including this one.
Worked example: scoring answers to one motion
Motion: This house would make voting compulsory
Task: Find the weakest link in the proposition argument that compulsory voting produces a more representative government.
Strong answer: Points out that turnout and representation are different: compelled voters who know little about the candidates may vote randomly or for the first name on the ballot, so higher turnout may not mean preferences are better represented.
Weak answer: Says compulsory voting violates freedom. That is an opposition argument, not an attack on the proposition's link, and it does not engage with representation at all.
How to score it: Give the point only if the tool attacked the reason behind the claim (why more voters means better representation) rather than changing the subject.
Running the same task through several tools shows quickly which ones understand debate structure and which ones produce plausible but generic text.
Which tools fit which format
World Schools (WSDC)
WSDC prep is team prep under time pressure, often for impromptu motions. Prioritise fast brainstorming and motion analysis, plus a way to practise Points of Information. Few tools model the full WSDC round, so expect to time speeches yourself.
British Parliamentary (BP)
BP rewards role knowledge and extensions. Look for tools that let you set the format and your team position, and that can show you what the opening half is likely to say so closing teams can plan something new.
Public Forum (PF)
PF depends on evidence and on keeping summary and final focus consistent. Strong source handling matters more here than in most formats, along with practice rounds that use real PF speech times.
Lincoln-Douglas (LD)
LD needs tools that handle value and criterion, not just policy arguments. Test whether a tool can defend a value and criterion (the framework the judge decides by) and tie each argument back to it, and whether it can practise the time-pressured 1AR.
Where DebateOwl fits
DebateOwl is a dedicated debate platform that keeps every case, source, round and debrief inside one Current Debate. It covers case generation with "If they say…" answers to the other side, Library sources that ground its answers, Attack for finding weak links, Brainstorm, Factoids for checking claims, an Ask coaching chat with optional citations, Crossfire and full Sparring rounds with a verdict, a speech timer and an argument flowchart.
Its limits are worth stating plainly. Sparring currently runs School 4-6-6-3, Lincoln-Douglas and Public Forum rounds, so WSDC and BP practice uses the prep and coaching tools rather than a full simulated round. Like every AI tool, its evidence leads need checking against the original source. Run the one-hour test above and decide on your own results.
Frequently asked questions
What is the best AI tool for debate?
It depends on the job. General chatbots are flexible for brainstorming, research tools find sources, and dedicated debate platforms add structure, opponents and practice. Most debaters do best with a dedicated platform plus a way to verify sources.
Is a general chatbot good enough for debate prep?
For brainstorming and explaining concepts, often yes. It is weaker at keeping a motion, side and format in context, predicting the opposing case, timing speeches and giving round-style feedback.
Are free AI debate tools enough?
Free tiers are a good way to run the comparison test. Check the daily or monthly limits against how many practice sessions you plan before a tournament.
Can I use AI tools during a debate round?
Usually not. Most leagues restrict or ban internet access and electronic help during rounds. Check your league's rules, and use AI tools for preparation.
What are the alternatives to DebateOwl?
General chatbots, AI research assistants, other dedicated debate platforms, and simple practice utilities such as timers and flowing templates. Run the one-hour test above on DebateOwl and on each alternative you are considering.
How should I compare AI debate tools fairly?
Give every tool the same motion and the same tasks, such as a case, the strongest counter, feedback on your speech and evidence for one claim, then score each answer against the same criteria.
How do you find the key clash in a debate and collapse to it?
List every open disagreement on your flow, then ask of each one whether winning it alone wins you the round and whether losing it loses you the round. The clash that passes both tests, and where your warrant is stronger, is the one to go for. In your last speeches, say plainly that you are going for it, respond to the opponent's attacks on it and weigh it against their best point. Drop the rest, unless an argument has been turned against you, which you still have to answer.
+ Strategy guide
Clash and Collapse: Find the Key Argument and Go All In
Map every disagreement, pick the one the judge has to resolve, and spend your last speeches there.
How do you find the key clash in a debate and collapse to it?
List every open disagreement on your flow, then ask of each one whether winning it alone wins you the round and whether losing it loses you the round. The clash that passes both tests, and where your reasoning is stronger, is the one to go for. In your last speeches, say plainly that you are going for it, respond to the opponent's attacks on it and weigh it against their best point. Drop the rest, unless an argument has been turned against you, which you still have to answer.
Map the clashes
List every point where the two sides disagree, and sort out direct clash from arguments that only pass each other.
Choose the deciding clash
Four questions that tell you which disagreement wins the round if you win it, and which you can afford to lose.
Collapse cleanly
Drop the rest on purpose, say so, and use the time you saved on weighing.
-
What counts as a clash
A clash is a point where both sides make claims that cannot both be true, or that pull the judge in opposite directions. It is not the same as two arguments on the same topic. If the proposition says a policy raises test scores and the opposition says it hurts student wellbeing, those arguments can both be true; until someone compares them, they pass each other and the judge has to do the comparing.
Clashes usually come in three kinds:
Factual or causal: will the policy actually cause the outcome? This is where most rebuttal happens.
Comparative: both outcomes are real, so which matters more? This is weighing, and it is where most close rounds are decided.
Framework: what standard should the judge use? In Lincoln-Douglas this is the value and criterion; in other formats it is the burden each side has accepted.
How to find the clash that decides the round
Midway through the round, write each open disagreement on one line of your flow. Then ask four questions of each line:
If we win only this clash, do we win the round?
If we lose this clash, can we still win?
Do we have the stronger warrant here, or only more material?
How much speaking time does it take to win it properly?
The deciding clash is usually the one where the first answer is yes and the second is no. If two clashes both pass that test, pick the one where your warrant is stronger. A clash you can win in thirty seconds of explanation is worth more than one that needs three minutes of evidence.
Worked example: from five clashes to one
Practice motion: This House would replace homework with a longer school day
Clashes on the flow after the first speeches: Learning outcomes, equity for students without a quiet place to study, family time, teacher workload and cost to schools.
Proposition's choice: Equity. If extra practice happens at school, every student gets the same conditions; the opposition's answer, that schools could open study rooms, concedes that home conditions differ.
What proposition drops: Teacher workload and cost, which they were losing, and family time, which the opposition never weighed.
How they say it: "Three clashes are left. We will win on equity, and we will show why it outweighs the cost argument even if the opposition is right about the budget."
Opposition's best reply: Make learning outcomes the deciding clash instead, arguing that longer days add fatigue rather than learning, so equal conditions for less learning is no gain.
Both teams are now arguing about which clash matters, and that comparison is what a judge writes on the ballot. The team that names the deciding clash first usually sets the terms for the rest of the round.
Collapsing: why fewer arguments win
Collapsing means choosing the few arguments you are winning and spending your remaining time on them, instead of extending everything at the same depth. Late speeches are short. Every argument you keep costs time you could spend explaining its warrant and weighing it against the opponent's best point.
To collapse cleanly, say what you are going for, answer what the opponent said about it, and weigh. If you drop an argument the opponent has turned into a reason to vote against you, you still have to answer the turn, or concede the defence that takes the turn away; otherwise the dropped argument becomes their offense.
Where each format collapses
Format
Where the collapse happens
Time
Public Forum
Summary, then final focus extends only what summary did
3 minutes, then 2
Lincoln-Douglas
Negative rebuttal and second affirmative rebuttal
6 and 3 minutes
Policy
Second negative and second affirmative rebuttals
5 minutes each
British Parliamentary
Whip speeches, which organise the debate into clashes without new arguments
7 minutes
World Schools
Third speeches organise the clashes; reply speeches give the biased summary
8 and 4 minutes
These times come from the formats the free debate timer runs. Practise the collapse against the exact slot, because a two-minute final focus fits one issue well and two at most.
Common mistakes
Extending every argument in the last speech because dropping one feels like losing it.
Collapsing to the argument you prepared best instead of the one you are winning in this round.
Calling a clash "the key issue" without explaining why the judge cannot decide the round without it.
Dropping an argument that the opponent turned, which hands them a free reason to vote.
Switching the deciding clash between summary and final focus, or between the block and the 2NR.
A 12-minute clash drill
Three minutes: take a flow from a recent round or a Sparring transcript and write every open disagreement on its own line.
Two minutes: answer the four questions above for each line and circle one clash.
Four minutes: give a two-minute final speech on that clash alone, timed, then a second version on the clash you did not circle.
Three minutes: write down which version was easier to weigh and why. That reason is your collapse test for the next round.
In DebateOwl, Flowchart lays out a case's claims and links so the disagreements are easier to see, and Case Attacker shows which of your arguments has the weakest link before you decide what to keep.
+
What counts as a clash
A clash is a point where both sides make claims that cannot both be true, or that pull the judge in opposite directions. It is not the same as two arguments on the same topic. If the proposition says a policy raises test scores and the opposition says it hurts student wellbeing, those arguments can both be true; until someone compares them, they pass each other and the judge has to do the comparing.
Clashes usually come in three kinds:
Factual or causal: will the policy actually cause the outcome? This is where most rebuttal happens.
Comparative: both outcomes are real, so which matters more? This is weighing, and it is where most close rounds are decided.
How the judge should decide (the framework): what test should the judge use? In Lincoln-Douglas this is the goal and the test for reaching it; in other formats it is what each side has agreed to prove.
How to find the clash that decides the round
Midway through the round, write each open disagreement on one line of your flow. Then ask four questions of each line:
If we win only this clash, do we win the round?
If we lose this clash, can we still win?
Do we have the stronger reason here, or only more material?
How much speaking time does it take to win it properly?
The deciding clash is usually the one where the first answer is yes and the second is no. If two clashes both pass that test, pick the one where your reasoning is stronger. A clash you can win in thirty seconds of explanation is worth more than one that needs three minutes of evidence.
Worked example: from five clashes to one
Practice motion: This House would replace homework with a longer school day
Clashes on the flow after the first speeches: Learning outcomes, equity for students without a quiet place to study, family time, teacher workload and cost to schools.
Proposition's choice: Equity. If extra practice happens at school, every student gets the same conditions; the opposition's answer, that schools could open study rooms, concedes that home conditions differ.
What proposition drops: Teacher workload and cost, which they were losing, and family time, which the opposition never weighed.
How they say it: "Three clashes are left. We will win on equity, and we will show why it outweighs the cost argument even if the opposition is right about the budget."
Opposition's best reply: Make learning outcomes the deciding clash instead, arguing that longer days add fatigue rather than learning, so equal conditions for less learning is no gain.
Both teams are now arguing about which clash matters, and that comparison is what a judge writes on the ballot. The team that names the deciding clash first usually sets the terms for the rest of the round.
Collapsing: why fewer arguments win
Collapsing means choosing the few arguments you are winning and spending your remaining time on them, instead of extending everything at the same depth. Late speeches are short. Every argument you keep costs time you could spend explaining why it is true and weighing it against the opponent's best point.
To collapse cleanly, say what you are going for, answer what the opponent said about it, and weigh. If you drop an argument the opponent has turned into a reason to vote against you, you still have to answer the turn, or concede the defence that takes the turn away; otherwise the dropped argument becomes their offense.
Where each format collapses
Format
Where the collapse happens
Time
Public Forum
Summary, then final focus extends only what summary did
3 minutes, then 2
Lincoln-Douglas
Negative rebuttal and second affirmative rebuttal
6 and 3 minutes
Policy
Second negative and second affirmative rebuttals
5 minutes each
British Parliamentary
Whip speeches, which organise the debate into clashes without new arguments
7 minutes
World Schools
Third speeches organise the clashes; reply speeches give the biased summary
8 and 4 minutes
These times come from the formats the free debate timer runs. Practise the collapse against the exact slot, because a two-minute final focus fits one issue well and two at most.
Common mistakes
Extending every argument in the last speech because dropping one feels like losing it.
Collapsing to the argument you prepared best instead of the one you are winning in this round.
Calling a clash "the key issue" without explaining why the judge cannot decide the round without it.
Dropping an argument that the opponent turned, which hands them a free reason to vote.
Switching the deciding clash between summary and final focus, or between the block and the 2NR.
A 12-minute clash drill
Three minutes: take a flow from a recent round or a Sparring transcript and write every open disagreement on its own line.
Two minutes: answer the four questions above for each line and circle one clash.
Four minutes: give a two-minute final speech on that clash alone, timed, then a second version on the clash you did not circle.
Three minutes: write down which version was easier to weigh and why. That reason is your collapse test for the next round.
In DebateOwl, Flowchart lays out a case's claims and links so the disagreements are easier to see, and Case Attacker shows which of your arguments has the weakest link before you decide what to keep.
Frequently asked questions
What is the difference between clash and weighing?
Clash is the disagreement itself. Weighing is the comparison that tells the judge which side's winning arguments matter more when both sides win something.
How many arguments should I collapse to?
Usually one or two. Last speeches are short, and one argument explained and weighed in depth beats three extended in a sentence each.
Is it risky to drop arguments?
Only when the dropped argument has been turned against you. Answer the turn or concede the defence that removes it, and the rest can go.
When should I decide what to collapse to?
Plan likely collapses before the round, then decide after the opponent's rebuttal, when you can see which clashes you are actually winning.
AI can find leads, not reliable evidence on its own. It is good at suggesting where to look and which searches to run, but a summary can lose a qualification, mix up two studies or point to a page that says something different. Open every original, check the author, date, publication, method and the text around the quote, and record the full citation. If the evidence is weak, do not read it. Narrow your claim or make the argument analytically, because a clear warrant often beats a card that falls apart under one question.
+ Research workflow
Find Better Debate Evidence Without Losing the Argument
Use AI to find research leads, then check each source and explain how it supports your point.
AI can find leads, not reliable evidence on its own. It is good at suggesting where to look and which searches to run, but a summary can lose a qualification, mix up two studies or point to a page that says something different. Open every original, check the author, date, publication, method and the text around the quote, and record the full citation. If the evidence is weak, do not read it. Narrow your claim or make the argument analytically, because clear reasoning often beats a quote that falls apart under one question.
-
Evidence leads
Research directions for a claim: the kinds of sources that could support it and searches you can run to confirm it.
Citation checks
A checklist for each source before it goes on a card: author, date, method and whether the quote says what you claim.
Impact links
Evidence tied to the warrant and the clash it supports, not added to a speech as decoration.
-
Evidence is not the argument
Evidence wins rounds only when it supports a clear warrant. A source should answer a burden in the round, not decorate a speech. Before you look for a card, write the claim it needs to prove in one sentence. If you cannot, more research will not help.
The most useful evidence sits on a specific clash: solvency, harms, trade-offs, comparative examples and the weighing.
AI research is a starting point
AI tools are good at suggesting where to look: which kinds of reports, journals or data sources might support a link. They are not a replacement for reading the source. A summary can lose a qualification, mix up two studies, or point to a page that says something different.
Treat every AI result as a lead. Open the original, read enough of it to understand the context, and only then decide whether it belongs in your case.
Finding evidence with DebateOwl
Give Factoids a single statement from your case. It researches that statement and returns supporting factoids with citations, context and suggestions for how to use them.
Generate a case. DebateOwl researches the motion before it writes, so each piece of evidence names its source and year, and the case ends with a source list: the organisation, the year, the claim it supports and the link.
Save the sources you trust to the Library. Active Library sources that fit a request are used by the other prep tools for that debate, so later answers draw on material you have checked.
Ask for counter-evidence too. Attack runs research on an argument and can point to sources that cut against it, which tells you what the other side may read.
Verify before you read it aloud
A bad card is worse than no card. If your evidence is challenged in cross-ex and does not say what you claimed, you lose the point and some of the judge's trust in the rest of your case. Check each source before it goes on a card:
Author: who wrote it, and are they qualified to make this claim?
Date: is it recent enough for the claim, and has anything changed since?
Publication: a peer-reviewed journal, a government report, a newspaper or a blog all carry different weight.
Method: for a study, what was the sample, how was it measured, and does it show cause or only correlation?
Context: read the paragraphs around the quote. Make sure the author is not describing a view they go on to reject.
Citation: record the full citation, so you can produce the source if the other side or the judge asks.
What to do when the evidence is weak
Do not read it. Look for a better source, narrow the claim so the evidence you have does prove it, or make the argument analytically: explain the reasoning without leaning on a card. A clear warrant with no card often beats a weak card that falls apart under one question.
Worked example
Motion: This house would introduce a four-day school week
Claim: A four-day week improves teacher retention.
Lead found: A summary says districts that moved to four days saw fewer teachers leave.
Check: The original study covers a small number of rural districts, and the authors say they cannot separate the schedule from pay changes made at the same time.
Decision: Narrow the claim to rural districts, say the evidence suggests but does not prove the effect, and support it with an analytical warrant about recruitment.
The weaker claim is easier to defend, and the honesty about the study removes the easiest cross-ex question the other side could ask.
-
Frequently asked questions
Can AI find evidence for debate?
It can find leads: sources and searches that may support a claim. You still need to open each source, read it in context and record the citation before using it.
How do I verify debate evidence?
Check the author, date, publication and method, read the text around the quote, and confirm that it says what you claim. Record the full citation.
What should I do if my evidence is weak?
Do not read it. Find a better source, narrow your claim, or make the argument analytically without the card.
Where does DebateOwl keep my sources?
In the Library for your Current Debate. Sources you mark active are used by the other tools for that debate.
Is a card better than an analytical argument?
Only if it is accurate and relevant. A strong warrant without a card often beats a weak card that does not survive a question.
+
Evidence leads
Research directions for a claim: the kinds of sources that could support it and searches you can run to confirm it.
Citation checks
A checklist for each source before it goes on a card: author, date, method and whether the quote says what you claim.
Impact links
Evidence tied to the reason behind the argument and the disagreement it helps resolve, not added to a speech as decoration.
+
Evidence is not the argument
Evidence helps only when it supports a clear reason why your claim is true (debaters call this the warrant). A source should answer a burden in the round, not decorate a speech. Before you look for a card, write the claim it needs to prove in one sentence. If you cannot, more research will not help.
The most useful evidence sits on a specific clash: solvency, harms, trade-offs, comparative examples and the weighing.
AI research is a starting point
AI tools are good at suggesting where to look: which kinds of reports, journals or data sources might support a link. They are not a replacement for reading the source. A summary can lose a qualification, mix up two studies, or point to a page that says something different.
Treat every AI result as a lead. Open the original, read enough of it to understand the context, and only then decide whether it belongs in your case.
Finding evidence with DebateOwl
Give Factoids a single statement from your case. It researches that statement and returns supporting factoids with citations, context and suggestions for how to use them.
Generate a case. DebateOwl researches the motion before it writes, so each piece of evidence names its source and year, and the case ends with a source list: the organisation, the year, the claim it supports and the link.
Save the sources you trust to the Library. Active Library sources that fit a request are used by the other prep tools for that debate, so later answers draw on material you have checked.
Ask for counter-evidence too. Attack runs research on an argument and can point to sources that cut against it, which tells you what the other side may read.
Verify before you read it aloud
A bad card is worse than no card. If your evidence is challenged in cross-ex and does not say what you claimed, you lose the point and some of the judge's trust in the rest of your case. Check each source before it goes on a card:
Author: who wrote it, and are they qualified to make this claim?
Date: is it recent enough for the claim, and has anything changed since?
Publication: a peer-reviewed journal, a government report, a newspaper or a blog all carry different weight.
Method: for a study, what was the sample, how was it measured, and does it show cause or only correlation?
Context: read the paragraphs around the quote. Make sure the author is not describing a view they go on to reject.
Citation: record the full citation, so you can produce the source if the other side or the judge asks.
What to do when the evidence is weak
Do not read it. Look for a better source, narrow the claim so the evidence you have does prove it, or make the argument analytically: explain the reasoning without leaning on a card. A clear explanation with no quoted evidence often beats a weak card that falls apart under one question.
Worked example
Motion: This house would introduce a four-day school week
Claim: A four-day week improves teacher retention.
Lead found: A summary says districts that moved to four days saw fewer teachers leave.
Check: The original study covers a small number of rural districts, and the authors say they cannot separate the schedule from pay changes made at the same time.
Decision: Narrow the claim to rural districts, say the evidence suggests but does not prove the effect, and support it with a step-by-step explanation of how recruitment would improve.
The weaker claim is easier to defend, and the honesty about the study removes the easiest cross-ex question the other side could ask.
+
Frequently asked questions
Can AI find evidence for debate?
It can find leads: sources and searches that may support a claim. You still need to open each source, read it in context and record the citation before using it.
How do I verify debate evidence?
Check the author, date, publication and method, read the text around the quote, and confirm that it says what you claim. Record the full citation.
What should I do if my evidence is weak?
Do not read it. Find a better source, narrow your claim, or make the argument analytically without the card.
Where does DebateOwl keep my sources?
In the Library for your Current Debate. Sources you mark active are used by the other tools for that debate.
Is a card better than an analytical argument?
Only if it is accurate and relevant. Strong reasoning without a quote often beats a weak card that does not survive a question.