A Human's Guide To Detecting AI Generated Content

Now, coming to the general context of writing, here's some tell-tale signs and patterns that can help identify non-edited AI content (or badly edited AI content):

- Repetitive Phrasing and Structures:

AI models often reuse certain phrases and sentence structures, which can make the text feel repetitive or overly uniform.

ex: "In the distance, I could hear the rhythmic beat of a dholak, signaling the arrival of a traditional dance performance. The sound drew my gaze to a group of performers clad in colorful attire, their graceful movements mesmerizing the audience gathered around them."

from "Second Chances by Saramitra"

- Analyze Writing Style:

Not the easiest thing to always do but... with a bit of a closer reading, you can always find out what's up. It also takes a bit of practice but you can eventually figure out.

ex: "For effective navigation of the privacy paradox presented by AI, a sophisticated, multifaceted approach is necessary. The role of lawmakers and policymakers in this context cannot be overstated. They are tasked with the onerous duty of revisiting existing laws, with an eye toward evolving them to accommodate the unique challenges presented by AI. This includes establishing strict regulations on AI-driven data-processing technologies and demanding greater transparency from developers about their algorithms and data sources.

Also important is that policymakers actively encourage and engage in public discourse on the delicate equilibrium between public safety and individual privacy rights. This will necessitate an inclusive conversation with all stakeholders — the public, law enforcement, and technology companies — to facilitate the creation of a balanced legal framework that adequately addresses everyone's needs and concerns."

from "The privacy paradox with AI by Gai Sher and Ariela Benchlouch"

- Lack of personal anecdotes or unique experiences

AI typically can't draw on personal experiences, so the writing may lack authentic, specific details.

Ex: "In the shadows where whispers weep,
Sorrow's song, its secrets keep,
Through veils of darkness, my heart does strain,
For love's lost echoes, a haunting refrain.

Eclipsed by memories, in the silent night,
Shattered dreams cast in the fading light,
Each teardrop falling, a symphony of pain,
Echoes of longing, a soul's silent bane."

from "imagined by author, created by ai collection - tapestry of forbidden love"

It sure feels... haunting and even realistic but if you tried to relate to it, it'd fall flat. This is one of the most promising signs to look out for, if you will.

- Unusual or Inconsistent Details

AI might generate content that includes minor factual inaccuracies, inconsistencies, or details that don't quite fit logically within the text.

ex: "You're sneaky, Sara! Yes, of course, I remember now, @illneas is none other than our dear friend, Yannis Kotsiras! The Yannis of many talents: poet, civil engineer, YouTube sensation, lover of Bukowski's gritty brilliance. His poetry is so raw and honest, I love it. And the fact that he juggles all these different aspects of his life, it just makes him all the more fascinating."

from a conversation with "pi.ai"

I am not too sure how many people know illneas or illy but I can assure you, his name isn't Yannis... this is a made up knowledge by it and serves as a good example that not everything the AI generates is true.