Hello! And sorry to hijack the post, but I just want to add a little thing.
As someone who works with AI daily, there ARE telltale signs that indicate people using AI in their replies, and no, as the previous reply said, it's not em-dashes and semicolons. Additionally, every AI (Claude, ChatGPT, Gemini, Perplexity, C.AI, etc.) writes differently, so pointing out that basic punctuation is a tell is an extremely flawed approach to AI detection.
Just remember: AI is not capable of creating. It gives the illusion of creation, but it is all based on an algorithm of probability. Large language models use probability to construct sentences based on the prompt it is given. So, it usually just replies to whatever actions have been thrown by the previous partner.
Do you remember that one boring ass roleplayer who forced you to come up with the plot and how to advance it? Everyone knows at least one. AI works similarly.
For example, what is the probability that the next word in this sentence will be X? If the probability is high, it will use it. And depending on the prompt, it might go to second best. And if you ask it to find something else, it will be the third-best possibility.
The AI will likely say blue, but if you ask it to wax poetic, it'll give it a shitty comparison about the ocean because the next probability about something being blue would be the ocean or the sea.
The BIGGEST tells you can find in AI-written replies is:
They're not likely to further the plot much, unless the prompter went all the fuck out to type a long prompt about how their character personality is, what kind of behavior they tend to use, and atmospheric details. AI roleplayers are lazy, and most likely, their prompting will be as well.
The writing is almost too perfect. No grammar mistakes, no missing coma, but also feels flat. Because the AI writes almost perfectly, reading AI feels like reading a god damn grocery list, or a furniture assembling guide. There is no emotional depth, people call it robotic. TRUST YOUR GUT. If you read something and it feels flat, then it's not written by a human. Even the shittiest roleplayers have a hint of humanity in their writing (except maybe the real bad ones who write one liners replies like 'he smiled at X and nodded his head' but they don't count. They're not roleplayers, they're looking for fanfictions.
Hallucinations. This one might be harder to spot in a creative writing context but AI is still very very prone to hallucination. If your AI roleplay partner doesn't scan / edit the reply provided by his robot buddy, you might detect something incredibly out of character, or some god modding. AI will do that. If you feel like your partner's character is changing on a dime, it's normal. Unless you use a paid version of AI (especially ChatGPT), the conversation memory is usually very weak, and hallucinations tend to happen.
Em-dashes and semi-colons are not a sign of AI.... until they're overused. And by that I don't mean using it like 5 times in a 500 words reply, I mean, you get them every other sentences. Same things with some words. If you see the same word repeated more than 3-4 times in a few sentences, it might be a sign of AI use. AI loves repeating high-probability words. This one, I would ask to be a bit lenient though, as it can also be attributed to writing style - but again, Trust Your Gut. If you feel a word has been overused in the past 2-3 replies of your partners, along other telltale signs, it might be an AI roleplayer.
Contradictions : Similarly to the hallucinations and characters doing OOC things, contradictions are a sign of AI. Imagine setting up a scene at the beach on a hot day in a reply, but then when your character talks, they forgot where they are and suddenly they're putting on a jacket because they're cold.
But even with all those signs, you can't 100% tell someone is using AI, but you can decide for yourself, based on YOUR GUT and some of these, whether you want to pursue a RP relationship with that roleplayer. Don't go calling out anyone, it's useless. Just stop replying.
A couple of signs alone might not be enough, but your GUT is enough. Stop wasting time on roleplays you won't enjoy anyway. This is a hobby, do it for the love of it, not because you have to.
Personally, if I have even a hint of a doubt about someone using AI, I will stop replying. But that's because I work with AI, and I am very confident in determining if someone has used it. And no, I don't use AI detectors, they're also prediction machines and they're bullshit most of the time. I have written entire essays by hand with an AI telling me it was over 50% AI, and then I've used AI to write work stuff and the AI detected less than 15% AI.
One last time: TRUST YOUR GUT. If it feels flat, if there's a lot of repetition, if the character is OOC or there are contradictions, if there are almost no grammatical mistakes-
IF IT SUCKS, HIT DA BRICKS.
