ShouldaSaid is a free AI debate game. You argue a topic — or redo a real fight — against an AI opponent across 3 rounds. Your logic, vocabulary, and structure are scored. No grammar policing, no signup required.
It's built for people who replay arguments in their head and want a place to actually win them — or practice arguing in a second language without the pressure of a real conversation.
Debate practice — pick a rotating daily topic, AI takes the other side, 3 rounds, scored.
Redo a fight — describe a real situation, AI plays the other person, you finally say what you meant.
Vent — no scoring, no judgment. Say everything. The AI responds in character.
English, German, Arabic, and Spanish. Scoring is fair across all four — the AI understands slang, typos, and informal phrasing. It grades the argument, not the grammar.
ShouldaSaid is made by Klarschritt, an independent software studio based in Berlin, Germany. We build small, focused products that do one thing well.
Website: www.klarschritt.de
Address: Irmtraud-Morgner-Str. 11, 10318 Berlin, Germany
ShouldaSaid started with a specific frustration: most language learning and communication tools are passive. You memorize vocabulary, you complete grammar exercises, you translate sentences. None of that prepares you to actually use a language under pressure — when someone disagrees with you, when you need to explain yourself in a meeting, when you're in the middle of a real conversation and your brain goes blank.
The core idea was that debate practice hits several skills simultaneously that isolated exercises miss: vocabulary in active context, logical structure, real-time response to pushback, and the kind of mild pressure that builds genuine fluency rather than just recognition. The goal was something that felt more like a sparring partner than a textbook.
The three-factor scoring model — argument length, vocabulary variety, and logical connectors — was developed iteratively and chosen because each factor captures something real about argument quality:
Argument length rewards development over assertion. A one-sentence claim scores lower than a claim with explanation and supporting reasoning — because "just trust me" is not an argument.
Vocabulary variety rewards range over repetition. Using the same three words throughout a debate scores lower than drawing on a wider lexical range. This incentivizes richer expression without penalizing imperfect grammar.
Logical connectors — words like "because," "however," "therefore," "as a result" — reward structured thinking. The connector detection uses fuzzy matching (Levenshtein distance) so typos don't unfairly penalize a player who clearly intended a connector word.
Insults and personal attacks deduct points. The system is designed to reward arguments, not aggression.
The four supported languages — English, German, Arabic, Spanish — required different implementation work beyond translation. Arabic uses right-to-left layout throughout, including the share card generated on results. Each language has its own connector dictionary with the logical transition words natural to that language's argumentative style. The AI opponent is prompted in the selected language and responds with culturally appropriate counterarguments. German, Spanish, and Arabic connector sets were validated against native-speaker usage patterns before launch.
Topics rotate daily and are reviewed before going live. We avoid topics that are gratuitously divisive without educational value, that require specialist knowledge most users won't have, or that risk causing real psychological harm. The AI is prompted to argue a position — not to attack the person. The goal is rigorous, fair debate practice.
The educational articles on this site — How to Win Any Argument and Why We Rehearse Arguments in Our Heads — are written with the same standard: original analysis grounded in psychology and rhetorical theory, not recycled listicle content. The argument technique guides draw on formal debate methodology, negotiation research, and cognitive psychology literature on performance under pressure.
All debate topics are reviewed for appropriateness. A profanity filter runs on every argument. The AI opponent is instructed to argue the topic — not attack the player. If you see something that shouldn't be there, let us know.
You can play fully anonymously. We don't require an account. If you log in with your email to save progress, that email is used only for the magic link — nothing else. Full details in our Privacy Policy.
More from ShouldaSaid