How NOT to let AI smother your voice
I believe that AI smothers points of view and voices that we ought to treasure. When used incorrectly, it reduces creativity and little-heard perspectives to a mush of insipid drivel. That drivel is often objectively correct, but it has all the flair and appeal of yesterday’s chips.
Writers with novel perspectives and those for whom English is an additional language often suffer from the homogenising effects of AI text generators.
What’s more, this effect can be strong enough to stifle or change the impact of their writing. So, I advise anyone who’s working with an editor to avoid using AI for anything more than research.
AI smothers marginalised voices
This point resonated with me recently, when I received a manuscript – a chapter for an edited reference work – from an author I’ve worked with before. This author is highly educated and an acknowledged expert in their field. They are from South Asia, as is their first language, but I know from past projects that their written and spoken English is great; easily good enough for academic discussion and writing.
In fact, I think this person’s use of English adds insights, cultural nuance and perspectives that enhance their writing.
So I felt sad when I read their latest work. It was too perfect (if grammatically correct, smoothly reassuring blandness is your idea of perfect). The author’s distinctive voice had vanished. Its replacement was a reading experience like motorway driving: smooth, unchallenging and disturbingly likely to lull the reader into autopilot.
Perfection can problematic
As it turned out, this author had used an AI text generator ‘to improve my English’. But to my reading, it had achieved the complete opposite. In important – but perhaps unquantifiable – ways, an objectively ‘better’ use of English actually reduced the quality of this author’s work.
So, how did this AI-generated problem affect my editing of that piece? To cut a long story short, I asked questions. Lots of questions. The author responded with lots of narrative ‘top of my head’ responses that had not been sterilised by AI, and I worked these in, which gave the copy more vivacity and originality. Also, I had worked with the author before, so while I didn’t try to mimic their voice, I was better able to edit out the more egregious examples of AI homogenisation and replace them. Between us, using highly collaborative editing, the author and I honed the copy into a publisher-compliant article that was much less bland and unpolluted by AI slop.
But, importantly, I think the outcome would have been even better if that copy had never been near AI.
The editor’s role
This is because a good editor can maintain a writer’s voice while imposing house style, correctness and clarity as required. My job is to retain the cultural nuances and the benefits of different perspectives (which are often encapsulated in language use, as this extended abstract from the 2026 CHI Conference in Human Factors in Computing Systems explains), while making sure the content is engaging and appropriate. That’s fine. That’s what I do.
What is almost impossible for an editor to do, certainly without input from the author, is add that distinctive voice later on in the editing process. And even where that does happen, I suggest that in almost all cases the text would have been more effective had the author written those insights and nuances, then let the editor do their thing.
How can writers avoid AI smother?
AI text generation is useful for writers. It’s great for research and scoping, as long as you double-check what it tells you. It’s good for assessing subject landscapes.
But I think that writers – if they want to say anything insightful, nuanced or original, or to produce copy that takes a novel perspective and/or is actively engaging – should avoid using AI text generators for anything more than research.
Specifically, I strongly advise against generating a first or even a rough draft through AI.
Once you have an AI-generated draft, you must work your own voice and views into a conglomeration of previously-published ideas. And that is much harder than simply writing what you think, then working with an editor to refine it. In that latter situation, at least you focus on your own ideas and perspective from the outset.
In theory, you could train the AI on your own voice draft accordingly. But that too is fraught with danger. . So, will you prompt AI in one language and then translate the output? That introduces further issues of loss or replication of nuance. What if English is your first language, but not the American English that so much AI has been trained on? How do you train an AI text generator to write from a specifically Australian or Indian perspective, for example?
AI cannot be original; people are
AI cannot think originally (although it can mimic the appearance of doing that). So if you publish anything that AI generated, you are merely restating what has been said elsewhere. If everyone does that, AI ends up feeding on AI-generated copy that is already bland. And the indications are that we’ll generate a desolate world of increasingly familiar and literally uninspiring blandness, as this article from The Conversation explains.
This is a truly dreadful prospect. And it immediately prompts a question. Namely, what will the tech bros and elites be doing while the rest of us are wading through dross?
I’ll leave that one with you.
I’ll close with this thought. If you’re writing something, and you’re working with an editor, you do not need an AI text generator. Write down your ideas and work with the editor to refine or express them to best effect. That way, you’ll communicate effectively and represent others who see things like you do. You’ll also add to the sum of human knowledge, and help to save humanity a terrible fate.
Comments are closed