Understand the Human or Not chat guessing game
Human or Not? is a short guessing game: chat anonymously for two minutes, then decide whether your partner is a person or an AI bot. The point is to enjoy the exchange and make a guess, not to prove who someone is. The current operator warns that chats can be shared publicly and may be used to train AI models, so keep personal details out of the conversation. Here’s how a round works, what its result can tell you, and how to play with sensible privacy boundaries.
What happens during a round?
The current Human or Not? FAQ describes the game as a two-minute chat with either a real player or an AI bot. The other participant is trying to identify you too. When the timer ends, you make your guess about whether your partner was human or AI.
The FAQ says human partners are real players from around the world. It currently lists large language models including GPT-5, Claude, and Google Gemini Flash as powering bots. Those are operator-provided details about the current game; they do not describe every AI model or every round, and the FAQ does not promise that each model appears in every session.
A useful way to approach a round is to treat it as a tiny conversation with an unknown partner. Ask something light that gives the other participant room to respond, follow up on what they say, and notice how the conversation develops. You can make a guess based on the whole exchange, but there is no need to turn it into an interrogation or to disclose personal experiences to test the other person.
What does your final guess mean?
Your result tells you whether your guess matched the game’s hidden assignment for that round. One correct or incorrect guess does not establish a reliable way to identify people or AI in other settings. A short text conversation gives you limited evidence: a person may write in an unusually polished or generic way, while an AI may produce a natural-sounding reply. A single conversational cue can point in either direction.
The operator FAQ suggests asking unconventional questions and notes that bots may struggle with recent events or deeply personal experiences. Treat that as a game suggestion, not a dependable detection test. A response that seems vague, fluent, mistaken, or specific is not proof of identity. The partner could have misunderstood, chosen not to share, or simply communicated differently than you expected.
What the original research found
The game’s name also belongs to a historical research project. In a paper submitted to arXiv on May 31, 2023, Daniel Jannai and coauthors describe an earlier online experiment. Over one month, more than 1.5 million users took part in anonymous two-minute chats with a human or an AI language model prompted to behave like a human. Players then guessed which they had encountered.
In that study, participants guessed correctly in 68% of games overall, and in 60% of games where they faced an AI. These are results from that particular experiment and its setup, not a measurement of today’s operator game or an accuracy score you should expect in your own sessions. The original paper on arXiv presents the findings as an experiment and calls for further extensions and refinements. Its historical results help explain the game’s research connection; they do not certify a general-purpose AI detector.
How to keep a round private and comfortable
The current FAQ gives a clear privacy warning: information on the site is public, chats can be shared online by other users or published on leaderboards, and chat data may be used to train AI models. It also points visitors to the site’s privacy, terms, and cookie policies. Because sharing is possible, assume anything you type could be seen outside the immediate conversation.
Before starting, use these simple boundaries:
Be aware that identifying details can include combinations—such as a school, workplace, neighborhood, and schedule—not just a name. - Keep contact details, passwords, financial information, private images, and account or security codes out of the chat. - Avoid sharing someone else’s personal information, too. - If a conversation turns uncomfortable, stop engaging rather than feeling obliged to complete the round. - Check the operator’s current privacy and terms pages if you want more detail about data handling.
These precautions follow from the operator’s own public-sharing and data-use disclosures. They also keep the game focused on what it offers: a brief, playful challenge rather than a private channel.
A simple way to make your guess
At the end of the two minutes, pause and ask yourself what actually informed your choice. Did your partner respond to the thread of the conversation? Did they handle an unexpected but harmless question? Did the exchange feel coherent overall? These are observations, not identity tests. They can help you make a considered game guess without relying on stereotypes about how people or machines “should” write.
If you’re unsure, that is a reasonable outcome. The game asks for a choice, but uncertainty is part of a short encounter with limited information. Make the guess, see the reveal, and treat the result as feedback about that one round—not as a verdict about the person, a test of your intelligence, or a rule to apply to strangers elsewhere.
Is Human or Not? a real Turing test?
The current FAQ describes Human or Not? as a social Turing test game and gives a brief explanation of the Turing Test: assessing whether a machine can exhibit intelligent behavior indistinguishable from a human. In practice, a two-minute game is a playful, constrained experience inspired by that idea. It is not a comprehensive assessment of intelligence, consciousness, or identity.
The helpful takeaway is straightforward: chat, enjoy the challenge, make a tentative guess, and protect your privacy. The original 2023 experiment is interesting historical context, while the operator FAQ describes the present-day game and its current disclosures. Neither turns a two-minute conversation into a reliable way to know who—or what—is on the other side.
Sources
Sources and scope
The cited sources support the current product description and bounded claims. Interface details, plans, and availability can change; check the live official pages for a purchase or installation decision. Editorial examples are illustrative and do not promise a particular result.