How can I tell if I am talking to a bot?

Look at behaviour more than wording: posting rhythm, generic replies, ignoring follow-ups, and many accounts saying the same thing. Ask specific local or personal questions. None of these is certain.

Can AI pass as human in a chat?

In short text exchanges, often yes. Modern language models write fluent, casual replies, and grammar is no longer a reliable sign.

What is a "bot or human" test?

Any check that tries to separate automated accounts from people, from CAPTCHAs to quizzes. The Human or AI test at hourofhumans.com/test shows how hard it is to tell by text alone.

Is there a way to prove I am human without ID?

Yes. Hour of Humans counts people anonymously by having them show up in person for 30 minutes; strangers confirm each other and no name or face scan is used.

Are you human?

Bot or human? How to tell, and why it is getting harder

For years, spotting a bot was easy: broken grammar, the same message pasted everywhere, an account created yesterday. Language models changed that. Here is what still works, and what does not.

Signs it is probably a bot

The account posts at all hours, at a steady rhythm, across unrelated topics.
Replies are polite, balanced and generic, and could be pasted under almost any post.
It never answers a follow-up, or answers a different question than the one you asked.
Many accounts say nearly the same thing within minutes of each other.
The profile is new or was empty for years, then suddenly very active.

Signs that no longer work

Bad grammar and spelling used to give bots away; now the bots write cleaner than most of us. Long dashes, lists and a confident tone are not proof either: plenty of people write that way, and AI can be told not to. In our Human or AI test, we even give both sides the same punctuation so you can only judge by what is said.

Questions that still help

Ask about something local and recent: "which café near the station closed last month?"
Ask for a personal detail that cannot be checked and see if the story stays consistent over several replies.
Ask it to do something slightly odd, like answer only in lowercase with one spelling mistake.
Change the subject abruptly and see whether it notices.

None of these is proof. A careful operator can pass all of them, and a tired human can fail them.

So how do you know someone is human?

Online, you mostly cannot be sure any more, which is the honest answer. Offline is different: a bot cannot sit across a table from you. That is why Hour of Humans counts people by having them show up: a short, anonymous check-in in a public place, where a few strangers confirm each other. If you want to know whether you can still tell people from AI by text, try the test first.