The recent outrage around the article “How to Do Good Research on X” has made me think about whether using AI to write articles or conduct research is right or wrong. With the advent of AI, both readers and writers are facing new problems. Let’s look at them individually.
how to be good at research
The reader’s concern is whether they are reading AI slop or something truly informative and educational. The writer’s concern is whether they are writing with intent and purpose, or just shipping AI slop.
Why we publish
We publish to satisfy our own curiosity, and so that others can find our work useful and build on it, or perhaps simply to entertain or inform. Writing begins with your own questions, your own observations, your own need to make sense of something. AI is a tool to help you express and refine that; it should never become the source of it. The arguments, the thoughts, the curiosity, those must remain yours. Which brings us to the question of how much AI use in that process is actually permissible.
How much AI use is permissible?
People use AI to different extents in their workflows. Some use it to polish their language, some to run experiments, some to generate hypotheses, and some use AI entirely, letting it plan experiments, run them, and write the paper. I think all of these are permitted. What truly matters is whether, when you publish your final piece of work, you are adding value, and by that I mean genuinely satisfying a reasonable reader: giving them something they wanted to learn, information they were seeking, or entertainment they came for, or whether you are simply prompting and sitting back while AI does everything.
I believe the research question you are tackling originates from your own curiosity, and you are pursuing it. In that pursuit, if you use AI to run experiments, write for you, or polish your language, all of that is fine. What is not fine is shipping something that adds no value, writing just for the sake of producing output. That is slop.
A note on transparency
Despite everything above, I think being transparent is better. AI conferences already have options to declare AI use in papers. Something similar in articles would be good too, a note at the end informing the reader that the piece was polished with AI assistance.
A caution for genuine writers
Genuine writers must be careful when using AI, because the biases it introduces are more subtle than they first appear. At the surface level, AI can shift the tone of your writing, turning a sentence you were still contemplating into one that reads as settled conviction, making your prose sound stronger or more authoritative than you actually intended. There is originality in retaining your own style; it is uniquely yours, and that is worth protecting.
But the deeper risk comes when you let AI become a true co-pilot of your thinking. At that point, it can quietly shape not just how you say things but what you say, and those biases are harder to catch and more consequential. It is not unlike asking a friend to write something for you and then putting your name on it. Be careful about distinguishing what is yours and what is not.
At the end of the day, you need to be the judge. If AI is writing for you, validate it. If a point AI makes genuinely adds value, that is fine, but do not let it sway you into believing something simply because it sounds right. AI can be very convincing. The arguments you publish should be ones you have actually examined and stood behind, not ones that just felt persuasive in the moment.
To reflect on my own journey
I used to use AI to rewrite my work with simple prompts, and it would change the entire language, tone, and texture of my writing. Now I have matured in my approach, my prompts ask only for sentence-level language improvement while preserving the paragraph structure and core ideas entirely. This way, the soul of the text stays intact and AI’s biases don’t creep into the substance. I will share the exact prompt Is given below
The prompt I use to polish, not rewrite
Write in complete sentences using clear, natural English. When the given text is wordy, reduce the length without making it too dense or hard to read. Do not add any information that is not already present; your job is strictly to rewrite what is there. Keep the original formatting exactly as you find it, maintaining bullet points as bullet points and paragraphs as paragraphs. Ensure the writing style stays personal, matching the tone of the provided text to reflect how someone would take notes in their own personal knowledge base. Avoid making it sound like corporate documentation or a company policy, as it needs to feel personal. But at the same time it should not be too preachy or dialogue-like or typical blog like. Maintain the breveity while maintaining the soul.
Intent is everything
At the end of it, everything comes down to intent.
If the intentions behind your work are genuine, your writing is worth it. If not, you are harming the broader ecosystem of knowledge.
And if you are a genuine writer using AI, stay aware of the biases it introduces, make sure the ideas, arguments, and curiosity driving your work remain unmistakably your own.
A few ways AI can genuinely help
To close, it is worth being concrete about where AI actually earns its place in a writing workflow.
- Tightening. Good writing has always required multiple rounds of revision, cutting redundancy, sharpening sentences, making the prose easier to move through. AI can do this well, and it is a purely linguistic service. Your ideas stay untouched; the reader just gets to them faster.
- Identifying gaps. You can ask AI to review your work and surface weaknesses, missing arguments, logical jumps, unanswered objections. Think of it as a mock peer review. The gaps AI finds are yours to address; the rebuttals and improvements come from you. AI finds the holes, you fill them.
- Polishing. Not everyone writing valuable things is a native speaker of the language they are writing in. AI is genuinely useful here, smoothing out sentences without changing what they mean. This is one of the cleanest uses of AI in writing, because the thought is entirely yours and the assistance is entirely surface-level.
- Ideation. You can hand AI your article and ask whether there is anything that could strengthen it, a relevant quote, a framing device, an angle you missed. Think of it like asking a thoughtful friend for their read. They might suggest something genuinely useful. But this is also where the line gets thin. There is a real difference between AI giving you an idea that you then evaluate, own, and decide to use, and AI quietly becoming the one generating your arguments for you. Stay on the right side of that line.