Since we are testers we tend to spot pattern and issues even when we do not want to 
.
So i was wondering how do you spot a text that is generated by an LLM/AI.
Please do not say I use a scanner.
For me the tells are usually these 3
What about you?
@restertest AI generally uses too many adverbs and adjectives to make its point.
I actually do these 2 myself xD
Damn I might be a robot! 
I knew it @kristof .
That explains it how you are so good at security.
You talk to the computer in robot languge.
Got you
It appears very polished
what I have noticed is that in the ChatGPT response, the semicolon is inserted between lines, so wherever I noticed a semicolon between lengthy lines, I understood that it is copy pasted from ChatGPT, because usually, we use a comma or full stop in real.
Very complex words are visible which we usually donāt use in day-to-day life.
Instead of āareā, the apostrophe re is visible. like weāre, youāre,
Two main things stand out, I donāt know 100% how to describe it, but itās super obvious when I see it. The paragraphs are broken up almost like fleshed out dot points? Like each one was a brief thought that has been expanded upon. And the last one is always a clear conclusion paragraph - much of the time even starting with the phrase āIn conclusion, xyzā. It reminds me of high school essay writing, very formulaic.
I think itās because people write like they talk and like LLM systems write to polished and doesnāt feel like somebody talking.
Why?
Some posts/texts/messages have this artificial vibe and sometimes I donāt even think about why exactly and I donāt really care if they are AI-generated or not.
Another point is that many of the characteristics mentioned above are just features of people for whom English isnāt their native language but they have quite a good level of studying it using books and other learning materials, passing exams rather than using it naturally in conversations with native speakers. People from some regions have quite a sophisticated language but again they donāt use it naturally so sometimes I can say that it sounds like a person from a particular country (wonāt mention the country because it might be stereotypic).
Additionally, using some tools to correct mistakes and improve readability like Grammarly may lead to adding this AI scent to the text. Is it okay for you?
Iām writing this because there might be lots of different cases (AI might be used in various ways even to develop unique ideas in certain ways which wonāt make the text completely AI-generated but it may look like so) and what it may seem at the first glance doesnāt mean a lot, even if itās completely AI-generated. Better concentrate on the content, ideas, and usefulness, and if you need to determine for sure if a text was AI generated (e.g. maybe youāre a teacher) then use particular tools and methods not just your intuition.
PS: Was this text AI-generated? Or AI-enhanced? Does it sound natural to you?
As one of those rare people who tries to keep the semicolon alive, I think you might be right.
I agree with a lot of the ātellsā listed here, but disagree with others. I also see people hold up some various grammar feature that they never use (e.g. em dashes) as ātheā way to tell a post is AI-generated. I tend to avoid relying on a single ātellā and use them as heuristics. Some of mine:
ā
)Apart from overused phrases and certain stylistic choices, one thing I would stress out is:
AI generated content is shallow. Thereās no insight, no knowledge behind it, it does not speak of real experiences.
Itās not necessarily wrong. Very often it is more or less correct, insofar as any writing may be correct. But thereās this veil of correctness, and thereās nothing behind it. Itās just the empty shell.
Itās also not about being vague. There are many reasons to be vague or to hide some details. But when you interact with a work that is purposefully vague, you can see the hints there was something more, but was omitted. While AI generated thing does not leave such hints, because there was never anything more.
It gives very similar feeling to SEO presell texts. If you know anything about the subject that is covered, you quickly realize that author has only theoretical knowledge of what they are talking about. They arenāt wrong and they might use the phrases or concepts correctly, but itās all completely abstract for them.
Unfortunately, there are way too many articles written by real humans that show these exact properties.
You both raise very strong points that I think lock down what bugs me the most about LLM responses - the ambivalence, and the emptiness.
A more in-depth analysis could be done by trying to engage the author in a meaningful discussion. If they change their style or donāt respond to any jibes or jokes, thereās your tell 
We lean on GPTZero to help inform us if text is written by AI. We do this as part of our article submission review process.
Amazingly, many Iāve asked just double down and their comments are also often clearly AI-generated. Itās almost as if thinking just isnāt very popular.
I think there is a a big difference between a text written by an AI VS a text written by a content creator from scratch and use an AI just to correct the syntax (but the content, structure and style of the narrative are still the creators).
In this case the AI is just a tool or an assistance) like asking a collaborator to have a look.
To me the creativity in an article itās the essence, the vibe, the meaningful content without unnecessary embellishments, that so far a AI cannot nail down (at least not yet). 
@rosie , does that tool able to distinguish when something is AI written/generated VS AI/reviewed?
I completly agree with you.
As a non native speaker, I tend to write my thoughts in an academic way which seems unnatural.
I also use AI to enhance my expressions and/or my grammar if Iām not sure how to say things.
Also u can check out Hugging Faceās Detect AI