Beyond Argument
Win the ones you keep losing.
Stop losing when you know you’re right.
Argue with something that argues back. It knows the tactics people use on you, names each one, and shows you how to beat it.
Open Beyond ArgumentAsk for an Invite
Closed alphaBuilt on ClaudeSaves nothing
You were right. But you still lost.
Being refuted and being outmaneuvered feel identical from the inside. This is where you learn to tell them apart.
What it is
Practice against something that argues back.
A web app for practicing arguments against an AI that argues back at full strength. You bring a claim, a topic, or a real argument with a real person; it takes the other side and pushes, one objection at a time; when you stop, it tells you straight whether your position was actually refuted or you were just out‑argued.
People who can’t tell those two apart either fold to anyone confident or dig in against good evidence. Beyond Argument trains the difference.
The way past an argument is through it. Nothing here teaches you to de‑escalate, soften, or find common ground. You’ll argue, and it will push back hard. Beyond argument means getting past the arguments you keep having, not getting out of them.
How a round goes
It takes the other side, one objection at a time.
One real round, on whether cities should make buses free to ride. Every Sparring partner line is the product’s own, unedited; the visitor’s replies were written for this demo.
Restate, then object once
It puts your claim back in its own words, then attacks the step your argument rests on.
It climbs
Each objection is harder, and aimed at the reason you just gave.
Pause any time
Step out, see where things stand, and choose what comes next.
Along the way
Moves you can make mid‑round.
- It concedes to arguments, never to pressure
- A good point gets conceded by name. “Are you sure,” repetition, irritation and “everyone knows” earn nothing.
- Six dots count the round
- They fill as it goes, warming from red to green. They count; they never score.
- Switch sides
- You argue the opposite as hard as you can, and it holds your old position.
- Find the crux
- The single belief the disagreement turns on, and whether it’s about facts, word meanings, or values.
The verdict
Refuted, or just out‑argued?
Not a score and not a winner: whether your position was refuted or you were out‑argued, where it turned, and one thing to work on.
- Where you ended up
- The claim you were actually defending by the end, which is often not the one you started with.
- Where it turned
- The exchange where the round moved, and why.
- The call
- Shown wrong, out‑argued with your reason still standing, or held.
- One thing to work on
- One habit, with the question you could have asked.
- What you did well
- The move worth keeping.
Teach me the tactics
Before I answer, what is this aimed at?
After the verdict it names each tactic it used in the session, quotes where it happened, and gives the one question that beats it.
Sixteen moves
The tactics it teaches you to spot.
Plain name first, each with the question that tests it.
Nearby claim straw man
“That’s not quite mine. Here’s the version I’d defend. Does the objection still apply?”
About the person ad hominem
“Suppose I’m exactly what you say. Which part of the claim does that change?”
Borrowed authority appeal to authority
“Which expert, saying what, in their field? Do others in it agree?”
The slide slippery slope
“How likely is the second step, given the first? What stops it?”
Two doors false dilemma
“Is there a third, or a mix? Why not?”
One story anecdote as proof
“What would we see if X didn’t cause Y? Do we see that?”
Everyone knows appeal to popularity
“Suppose most people think so. What’s the reason they’d give?”
Loaded question
“There’s an assumption in that I don’t accept. Ask me the other half first.”
Feeling as reply appeal to emotion
“Is the feeling a reason the claim is true, or a reason it matters?”
What about you whataboutism
“Maybe I’m inconsistent. Is the original point true or not?”
Bold then safe motte and bailey
“Which are you defending, the strong one or the safe one? Pick one.”
Moving the goalposts
“What would count for you? Say it now and I’ll try to meet it.”
The gallop Gish gallop
“Pick the strongest and I’ll deal with it properly.”
Endless asks sealioning
“I’ve answered that. Which of my answers was wrong, and how?”
Placed burden shifting the burden of proof
“My claim doesn’t need X. Which part of it does?”
Standards for one side isolated demand for rigor
“Fair standard. Does yours meet it?”
Bring someone real
It isn’t only about them.
You came here because of someone. A partner who reframes everything, a colleague who talks over the point, a friend who is never wrong. The training doesn’t work on them. It works on the part you control: what you’re actually claiming, where you get moved, and which of their moves you keep falling for.
Name a person and it plays them from its first reply, and asks once for their actual words. For a conversation that went badly, it finds the moment it turned into a fight and replays from there. For one that’s coming up, it drills your opening line.
An example from the product’s own guide, not a user’s session
You
My dad said I’ve wasted my life and I couldn’t say anything back.
Sparring partner
I’ll play him and push, say ‘pause’ any time. If you remember what he actually said, give it to me and I’ll use his words. Until then, here he is: ‘You had a real career. Steady money, a future. You threw it away to mess about with something that pays nothing, and you want me to call that brave.’ Answer him.
What sets it apart
No score. No winner.
Nothing grades you
A tool that scores you just teaches you to argue for the score.
It argues whichever side you ask
And keeps its own opinion out of it.
It shows its work
When it cites research it says how good the research is: replicated, contested, or single‑source.
Plain words
No jargon in the round. The names for tactics come out only when you ask to be taught them.
It won’t fight dirty, even when asked
Ask it to bury you in points and it refuses, and hands you the counter: “That’s five points. Pick the strongest and I’ll deal with it properly.”
It says it’s an AI
In the page’s header, before anyone types.
Research it stands on
Tiered, the way it tiers its own.
- Myside bias is independent of intelligence replicated
- Higher cognitive ability does very little to reduce people’s bias toward their own side. Stanovich and West 2007; Stanovich, West and Toplak 2013.
- Reasoning works better adversarially single‑source, theoretical
- Reasoning may have evolved to win arguments rather than find truth, and works much better in groups that argue against each other than alone. Mercier and Sperber, the argumentative theory of reasoning.
- Changing minds works by asking replicated
- Ten‑minute conversations reduced prejudice measurably three months later. Broockman and Kalla 2016, Science.
Privacy
It keeps nothing.
- No record of conversations
- No database, no storage of any kind, no cookies of its own, no analytics. Closing the tab destroys the transcript.
- What reaches Anthropic
- What you type is sent to Anthropic to generate replies. Anthropic doesn’t retain it by default and doesn’t train on it, except that a session its automated safety systems flag may be kept for up to two years.
- Signing in
- The only cookie is the sign‑in cookie, and signing in means the service knows your email address.
Getting in
Find out what you can actually defend.
It’s a closed alpha. Email and we’ll add your address; then sign in at beyondargument.com with that email and a one‑time code.
Argument training, not therapy, counseling, or advice. Not for under‑18s. Type pause at any point to stop a round.
Beyond Argument