The main way SE checks to see if someone is botting is have a GM send them a tell, under the impression if you see a GM tell you would reply to it thinking wtf. All GM names in FFXIV have Rep as the first name, therefore its easy to write a program that replies to a tell from someone called Rep xxx (xxx is a wildcard), the GM can also ask random questions, because AI is normally very limited, for example if a GM says "what do you think of this bear?" AI could reply "it is very nice" or seeing it as a question with what do you think in reply "I neither agree nor disagree" whereas a human would reply with a question like "what bear?"

The problem comes when people use highly advanced AI, its becoming better and always getting harder for people to recognise the differance with an AI reply and a human reply (I'm talking about advanced AI, not these robot chat websites you see.)

If SE takes bans too lightly legitimate players will get banned, they need to throughly investigate, which is why it seems to take a long time to do anything. Take me I spent most of yesterday mining in Nanawa, which some people could see as a potential bot, which is why SE needs to investigate.