AI makes editors read the writer as well as the text

‘AI-written or -polished text can be corrosive to readers because all such texts, no matter how diverse their contents, read the same way.’ | Photo Credit: Getty Images/iStockphoto I’m a commissioning editor. Every day, I receive more submissions of articles written by AI than the last. The same goes for articles polished by AI. More than going against The Hindu’s policy, this habit has the potential to corrode the human practice of writing.Writing is often hard. You consider the relationship between your intentions and the meaning of words laid down on a page. You think through using the many degrees of freedom a sentence or paragraph affords to convey precisely what you mean, down to the feeling, while also holding readers’ attention and doing justice to yourself without becoming self-absorbed. Writing should trouble the mind. AI-made or -polished writing offers to take this trouble away — a temptation to be resisted.AI-written or -polished text can be corrosive to readers as well because all such texts, no matter how diverse their contents, read the same way. Editing it to be more human, to have an undulating rhythm rather than that sterilised glaze, is also an exasperating experience.I used to maintain a list of the hallmarks of AI-written or -polished text, but as they evolved, I turned to two ‘detector’ tools: GPTZero and Pangram. I trust them because of their favourable coverage in the scientific press, their developers’ attitude towards the tools’ role in the publishing enterprise, and their lower false-positive rates. A false positive is when a tool flags a piece as being AI-written when in reality it is not. It can be pernicious because it causes me to decline a submission I may have accepted, potentially disappointing an author. Such a result may also cause editors to develop an unfavourable opinion of such contributors.Sometimes, these tools jointly return very high or very low scores, allowing me to decide quickly about accepting or declining submissions. The problem arises when one or both tools return a middling score, e.g. “50% chance of being written by AI” or “50% of the text is written by AI”. In these instances, I need to negotiate a way forward. If the author is a rookie, I may insist on seeing the timestamped version history of the document containing their text, and ensure the words didn’t suddenly appear all at once. (If I have doubts at the time of commissioning itself, I may insist the author quote some experts firsthand and submit their reporting notes together with the finished article.) If I have a mature relationship with an author, I directly communicate the stakes and possible courses of action.But when an author doesn’t tick these convenient boxes, I need to draft a reply that balances several sensibilities. First, I must not offend them. In fact, I have no intention of villainising them: people use AI models with text for various reasons, only a small subset of which are malicious. Second, I need to elicit from the author an honest response about whether, and if so how, they used AI — especially a reply I can be sure is unencumbered by insecurities or second-guessing. Third, I need to keep the author’s trust while maintaining that I cannot accept AI-written or -polished text. Fourth, I must justify, if required, my decision to repose my trust in those tools rather than in my judgment and/or in our relationship alone.Now, if GPTZero and Pangram independently falsely flag 5 out of every 100 articles as being AI-written or -polished, they will have a combined false positive rate of just 1 in 4 million. However, the chance of one tool flagging an article right and the other flagging it wrong is still 1 in 1,000, which is much higher.But I cannot make decisions with mathematics alone as my guide. AI has caused me to think more closely about the author submitting an article and their reasons to use — or not use — AI. Similarly, it would help if authors think more closely about editors, who are caught between their duty to their readers and employers on one hand and to facilitate the publishing process on the other.mukunth.v@thehindu.co.in Published - September 18, 2026 06:15 am IST