It is not often that you read an article and feel the writer understands you. The internet is awash with content that get published by the minute. But once in a while, an opinion piece comes along that has the power to move you. I recently read one such opinion piece, written by a civil servant under probation, and published in a leading Indian news daily.
The article is on the type of friendship that thrives in idle times spent in each other’s company in one’s teenage years. The article triggers intense nostalgia and a longing for the good old days of youth. No wonder the article went viral, and many readers in chat groups wow-wowed about the author’s maturity, depth, and craft. But soon, a controversy emerged when some readers alleged that the article was AI-generated.
It was hard to believe that such a touching piece could be synthetic. So I checked for myself on Grammarly, a software that detects plagiarism as well as AI content. Grammarly’s verdict was that 56% of the text has patterns that resemble AI text. But does it mean that the article was AI-generated? Or was the author’s grammar too perfect? Or is it the case that the author’s original work was AI-aided?
The patterns in AI writing
AI is trained on human writing and should (theoretically speaking) resemble it. However, because it’s trained on terabytes of data, it tends to learn to perform at its best. AI detection tools look for predictability of language. AI-generated texts tend to be predictable and measure low on perplexity. Further, human writing has high burstiness. Human writers mix short and long sentences and arbitrarily make simple and complex constructions when we write. AI bots follow a consistent rhythm, and so have low burstiness.
But someone who knows this can give an advanced AI model prompts to maintain ‘high perplexity and burstiness’. This happens to be a trick to bypass AI detection tools, and some students may indeed be using such prompts to have the chatbot write their term thesis.
At the same time, a good human writer could be predictable and consistent, especially when writing short articles. This is where the problem arises. AI detection tools analyse pattern, and are not always correct. This same article was considered 100% AI-written by some tools.
We don’t know any longer who has written what. Readers usually construct a persona of the author from their writing. Beyond just a construction of words, good writing is the voice of a human being who impresses us, readers.
AI writing feels quite fake. Ambiguity over whether a composition that impresses us, holds power over us, is human or AI work can lead to cynicism. Such ambiguity fills the reader with self-doubt, and the reader would hesitate before picking up a book next time in a bookstore.
Award-winning AI
A separate controversy that has emerged over the last week relates to a short story that won a Commonwealth Foundation prize for best writing. ‘The Serpent in the Grove’ was written by 61-year-old Jamir Nazir from Trinidad & Tobago and published in Granta, one of the world’s most influential literary magazines.
The Observer reports that some AI detectors have deemed the story to be AI-generated. Critics have pointed out motifs in the short story that are typical of AI literature. Yet, a team of human judges considered it the best entry for the prize. It does not matter anymore if the prize is withdrawn and bestowed on another short story (which we will never know was written by a human). What matters is that human judges were impressed by this short story.
It is still not conclusively proved whether Mr Nazir wrote the story, but this might be the new normal in writing. Some writers have observed that AI-assisted writing does not take away the human writer’s originality. Just like the computer only assisted the writer’s originality, and so did Microsoft Word’s grammatical prompts, AI only assists and amplifies a writer’s originality.
The piece written by the trainee civil servant reverberated with lived memory, yearning, and melancholy. I could read in it the author’s digital loneliness and a sense of exhaustion with the routines of adult life. Definitely, AI did not come up with these emotions.
Smarak Swain is an IRS officer and author of “The Scammer’s Trail” (Bloomsbury, 2025). He is learning prompt engineering on AI to write his next fiction. Views are personal.
Disclaimer: The views expressed are the author’s own and do not reflect the official policy or position of Financial Express.