Human or Not? Take the Quiz to Find Out 2026
Human or Not is the viral AI quiz game that has challenged over 1.5 million players worldwide to answer one deceptively simple question: are you chatting with a real person or an artificial intelligence?
Created by AI21 Labs, this free online game pairs you with a mystery partner for a two-minute conversation — and then asks you to guess who you were actually talking to. It sounds easy. It is not. Research shows that nearly one in three players gets it wrong.
What Is the Human or Not Game?
Human or Not is a free, browser-based social Turing game developed by AI21 Labs.
It works like a text-only chatroulette. You are paired with a mystery partner — either a real human player from somewhere in the world, or an AI chatbot powered by advanced large language models. You chat for two minutes on any topic you choose, and at the end you must guess: human or bot?
The answer is revealed instantly. Your result is recorded alongside millions of other players as part of one of the largest public AI experiments ever conducted.
Who Built Human or Not?
The game was created by AI21 Labs, an Israeli AI company founded in 2017.
AI21 Labs is known for developing the Jurassic series of large language models and the writing assistant Wordtune. The company built Human or Not as both a public entertainment product and a large-scale scientific experiment to measure how well modern AI can pass as human in real-world, uncontrolled conversation.
The game launched in April 2023 and became a viral hit almost immediately. By the end of its first month, it had generated over 10 million conversations from 1.5 million participants across the globe.
How Does Human or Not Work? Step-by-Step
Playing the game is straightforward. Here is exactly how each round works.
You visit humanornot.so in any web browser on desktop or mobile. No account, registration, or download is required. The game is completely free to play as many times as you like.
The system searches for a match and randomly pairs you with either a human player or an AI chatbot. You do not know which one you have been paired with at the start.
A two-minute text chat begins. You can talk about anything — weather, hobbies, random questions, philosophy, current events, or whatever comes to mind. There is no voice, no video, and no profile information shared. The interaction is purely text-based.
When the two minutes end, you are presented with a single choice: Human or Bot? You make your guess and the result is revealed immediately.
| Step | What Happens |
|---|---|
| Step 1 | Visit humanornot.so — no account needed |
| Step 2 | System randomly pairs you with a human or AI |
| Step 3 | Two-minute anonymous text chat begins |
| Step 4 | Chat ends — you guess: Human or Bot? |
| Step 5 | Result revealed instantly with accuracy stats |
The AI Models Behind the Bots
The AI opponents in Human or Not are not simple rule-based chatbots. They are powered by some of the most advanced large language models available.
The original game used AI21’s own Jurassic-2 model alongside GPT-4. The current version on humanornot.so also incorporates Claude, Cohere, and other leading LLMs. These models are specifically prompted to behave as naturally as possible — to make small talk, express opinions, use slang, and even simulate human imperfections like hesitation or casual grammar.
This is precisely what makes the game so challenging. You are not trying to spot a simple script. You are trying to detect some of the most sophisticated conversational AI systems ever built.
| AI Model Used | Developer | Known For |
|---|---|---|
| Jurassic-2 | AI21 Labs | Natural language, long-form reasoning |
| GPT-4 | OpenAI | Broad knowledge, human-like conversation |
| Claude | Anthropic | Nuanced, thoughtful responses |
| Cohere | Cohere | Enterprise-grade language understanding |
The Numbers: What the Data Actually Shows
The Human or Not experiment produced some of the most significant real-world data on human-AI interaction ever collected.
Over 1.5 million players participated in the first month alone, generating more than 10 million conversations. The scale makes it the largest public Turing-style test ever conducted at the time of its release.
The headline finding: players correctly identified their chat partner in only 68% of games overall. That sounds reasonable until you realize it is barely above flipping a coin for AI encounters.
When players were specifically paired with an AI bot, correct identification dropped to just 60%. In other words, when facing a real AI opponent, 40% of players thought they were talking to a human.
Players were actually better at spotting other humans than they were at spotting AI. Humans correctly identified their human partners roughly 73% of the time — but identified AI correctly only 60% of the time.
| Metric | Result |
|---|---|
| Total players (first month) | 1.5 million+ |
| Total conversations generated | 10 million+ |
| Overall correct guess rate | 68% |
| Correct rate when facing AI | 60% |
| Correct rate when facing human | 73% |
| Players who couldn’t tell the difference | 32% |
These numbers have enormous implications. Nearly one in three people cannot reliably distinguish a machine from a human in a short, casual conversation. As AI models become more capable year on year, that number is expected to continue falling.
Why Is the Human or Not Quiz So Hard?
The difficulty comes from how dramatically AI conversational ability has improved in recent years.
Modern large language models are trained on hundreds of billions of words of human text. They have absorbed enormous amounts of human conversation, slang, humour, hesitation, emotional expression, and cultural reference. When prompted to sound human, they are extraordinarily good at it.
Players who expect AI to be obviously robotic, stiff, or overly formal are repeatedly surprised by how casual and natural modern chatbots can be. The bots crack jokes, ask follow-up questions, express preferences, and even simulate being distracted or uncertain.
At the same time, real humans sometimes type unusually formally, give short abrupt answers, or respond in ways that seem scripted — making them look like bots. The confusion goes both ways.
Top Strategies to Win the Human or Not Quiz
Experienced players and researchers have identified the most reliable signals to look for during your two-minute chat. None of these are foolproof, but used together they give you a meaningful edge.
Test Emotional Depth and Specificity
AI models tend to give emotionally balanced, diplomatic, well-rounded answers to emotional questions.
Ask something like “what is the worst thing that ever happened to you?” or “what makes you genuinely angry?” A real human typically answers with something specific, messy, personal, and non-generic. An AI will often give a thoughtful but suspiciously well-structured response.
Ask About Very Recent Events
AI models have training data cutoff dates. Even internet-connected models sometimes struggle with highly specific recent events.
Ask about something extremely recent and local — last night’s sports result, a specific regional news story, or something that happened this week. A human can draw on lived experience. An AI may hedge, approximate, or give a slightly generic answer.
Try Unusual Wordplay or Typography Tricks
One of the most popular player strategies is to test text-level awareness.
Ask your partner to spell a word backwards, count the letters in a word, or read a sentence written in reverse. Early AI models struggled significantly with these tasks. More advanced models have improved, but unusual typographic challenges still occasionally trip up bots in ways humans handle naturally.
Notice Response Timing and Rhythm
Humans take varying amounts of time to respond depending on what they are thinking about.
AI systems tend to respond very quickly and consistently regardless of how complex the question is. If your partner answers a genuinely difficult emotional question at the same speed as a simple greeting, that is a signal worth noting.
Ask Something Morally Complex or Controversial
AI models are typically trained to avoid taking strong controversial positions.
Ask a genuinely controversial personal opinion question — something where a real person would likely have a clear, possibly heated view. If the response is balanced, diplomatic, and carefully hedged from all angles, that is often a sign of a bot trying not to offend.
Look for Over-Politeness
Research from the Human or Not experiment found that players often associated extreme politeness with AI.
Human players in casual anonymous conversations tend to be more direct, blunt, or even slightly rude. If your partner is unusually gracious, formal, or constantly courteous in a way that feels slightly unnatural for a random internet chat, that is a weak signal pointing toward AI.
| Strategy | What to Look For | Reliability |
|---|---|---|
| Emotional depth test | Generic vs. personal, specific answers | High |
| Recent events | Hedging or generic answers to current topics | Medium-High |
| Wordplay / backwards text | Errors or unusually slow processing | Medium |
| Response timing | Consistently fast regardless of complexity | Medium |
| Controversial opinion | Balanced, diplomatic non-answer | Medium-High |
| Over-politeness | Unnatural formality in casual chat | Low-Medium |
What the Game Reveals About AI in 2026
Human or Not is more than a quiz. It is a window into the state of artificial intelligence in the real world.
The game demonstrates clearly that AI has already crossed a threshold that many people did not expect to arrive so soon. In short, casual, text-based interaction, modern AI is genuinely difficult to distinguish from a human without very specific probing.
This has direct implications for customer service, online communication, social media, and virtually every digital space where text interaction happens. If trained players in a dedicated game setting can only correctly identify AI 60% of the time, ordinary users in everyday digital interactions are likely doing no better — and quite possibly worse.
The experiment has been cited in academic research, referenced in AI policy discussions, and used as an educational tool by digital literacy advocates to demonstrate the real-world capabilities of LLMs.
Human or Not vs. the Traditional Turing Test
The original Turing Test was proposed by British mathematician Alan Turing in his landmark 1950 paper “Computing Machinery and Intelligence.”
Turing proposed a simple question: can a machine exhibit intelligent behaviour indistinguishable from a human? In the classic setup, a human judge converses with both a human and a machine and tries to tell them apart. If the judge cannot reliably distinguish the machine, it is considered to have passed.
Human or Not updates this concept for the modern internet age. Instead of a formal lab experiment with trained judges and structured questions, it puts millions of ordinary people in the judge’s seat with no special training, no time to prepare, and no structured prompts.
| Feature | Classic Turing Test | Human or Not Game |
|---|---|---|
| Setting | Formal, lab-based | Casual, web-based |
| Judges | Trained evaluators | Any member of the public |
| Duration | Typically 5+ minutes | Exactly 2 minutes |
| Scale | Small (dozens of participants) | Massive (1.5M+ players) |
| AI used | Varies by experiment | GPT-4, Claude, Jurassic-2, Cohere |
| Outcome | Academic result | Instant personal feedback |
Similar Games and Alternatives in 2026
Human or Not has inspired a range of related games and tools for anyone who wants to test AI detection skills.
HumanOrNot.io is a separate platform (different from humanornot.so) that offers a similar chat-and-guess format with slightly different AI models and interface.
TuringTest.live allows players to compete directly against various LLMs in a similar format, with optional leaderboards and extended conversation times.
AI or Not by Sightengine shifts the format from chat to images, challenging users to identify whether photographs were taken by a real camera or generated by AI image tools. Most players score between 55% and 75% accuracy.
Humanize AI’s Human or AI game focuses entirely on headshot images, asking players to distinguish real human faces from AI-generated portraits — a rapidly growing challenge as tools like Midjourney and Stable Diffusion advance.
The classic ELIZA bot, originally developed in the 1960s at MIT, remains a useful comparison point. In modern Human or Not experiments, even the basic ELIZA-style bot was correctly identified as AI only 76% of the time — a sign of how low the baseline of human detection actually is.
Why Human or Not Matters Beyond the Game
The broader significance of Human or Not extends well into everyday digital life in 2026.
Online customer support now uses AI chatbots in the majority of consumer-facing interactions. Social media platforms contain growing numbers of AI-powered accounts and comment generators. Dating apps, forums, and messaging platforms all contain non-human participants that may or may not identify themselves as such.
Understanding how convincingly AI can pose as human — and learning to recognise the signals — is a practical literacy skill in the modern internet era. Human or Not gamifies that skill development in a way that is accessible, engaging, and free.
It also raises important ethical questions. If AI can consistently pass as human in casual text conversation, what obligations do companies and developers have to disclose when a user is talking to a bot? These are questions being actively debated in AI policy circles right now.
Frequently Asked Questions (FAQs)
What is the Human or Not game?
Human or Not is a free online game where you chat for two minutes with a mystery partner and guess whether you were talking to a real human or an AI chatbot. It was created by AI21 Labs.
Is Human or Not free to play?
Yes, completely free. The game runs in any web browser at humanornot.so with no account, registration, or download required.
How accurate are players at detecting AI in Human or Not?
Research from the game’s first month shows players correctly identify their partner only 68% of the time overall, and just 60% of the time when paired against an AI bot.
What AI models does Human or Not use?
The game uses advanced large language models including AI21’s Jurassic-2, OpenAI’s GPT-4, Anthropic’s Claude, and Cohere, all prompted to converse as naturally as possible.
Can you trick the AI in Human or Not?
You can probe it with emotional depth questions, very recent events, backwards text, or controversial opinions to detect AI patterns — but modern models are highly sophisticated and resist simple tricks.
Is there a time limit in Human or Not?
Yes. Each round lasts exactly two minutes. When time runs out, you are asked to make your guess immediately.
Who created the Human or Not game?
The game was created by AI21 Labs, an Israeli AI company known for the Jurassic language model series and the Wordtune writing assistant tool.
How many people have played Human or Not?
Over 1.5 million players participated in the first month of the original experiment, generating more than 10 million individual conversations.
Is Human or Not based on the Turing Test?
Yes. The game is directly inspired by Alan Turing’s 1950 concept of the Imitation Game, updated for mass public participation through a free, casual web interface.
What is the difference between humanornot.so and humanornot.io?
They are two separate platforms. Humanornot.so is the original game by AI21 Labs. Humanornot.io is a different but similar alternative platform with its own AI models and interface design.
Conclusion
The Human or Not quiz is far more than a casual internet game. It is a real-time window into how far artificial intelligence has come — and a practical test of a skill that matters in everyday digital life in 2026.
When nearly one in three players cannot distinguish a machine from a human in a two-minute chat, it forces a serious reckoning with how we interact online.
Whether you play for fun, to sharpen your AI detection instincts, or simply out of curiosity, the game delivers an experience that is genuinely surprising almost every time. The bots are better than you expect. The humans are stranger than you remember.
The best strategy is to go in without assumptions, ask questions that require emotional specificity, and pay attention to the rhythm of replies.
The more you play, the better you get — and the more you understand about the technology quietly reshaping how we communicate. Start your first round today at humanornot.so and find out where you stand.
