The Ethics of AI-Assisted Writing, Plainly

Most writing about the ethics of AI is either a sermon or a sales pitch. The sermon says the tool is theft and you are a fraud for touching it. The pitch says everything is fine, use it for everything, credit no one. Both are lazy, and both are written by people who have not actually sat with a deadline, a blinking cursor, and a genuine question about what they are allowed to do with the help sitting on their desk. The truth is a set of plain, livable lines that a working writer can actually hold while the deadline breathes down the neck. Let me lay them out the way I would to a younger writer over chai, without panic and without permission-granting, because I think most of the noise around this subject comes from people who have never had to actually decide, on a Tuesday, at 11 p.m., whether the sentence in front of them is theirs.

I want to be honest about where I am writing from. I have used these tools. I use them still, for research, for catching a structural hole I am too close to see, for testing whether a line of dialogue sounds like a human or a pamphlet. I am not writing this from outside the practice, wagging a finger at it. I am writing it from inside the practice, having drawn lines for myself that I have had to defend more than once, including to myself, at hours when it would have been easier not to.

The first line: authorship is not negotiable

Start here, because everything else hangs off it. If your name is on the work, you are responsible for every word, whether you typed it or a machine suggested it. That is not a new rule. It is the oldest rule in publishing, only now the ghost in the room is software instead of a hired hand. The moment you paste a line you did not read closely enough to defend, you have not saved time. You have signed something you did not read. The Authors Guild's best-practice guidance circles the same principle: the tool can propose, but authorship and accountability stay with the human whose name is on the cover.

This is why I never treat the machine as a co-writer. A co-writer shares the blame. The machine cannot. It has no stake, no reputation, no 3 a.m. dread about whether the scene is honest. All of that stays with you, so the authorship stays with you too.

Let me make this concrete with a scene I am inventing to illustrate the point, not reporting as fact. Imagine a writer, call her Riya, working on a family drama. She is stuck on a scene where a daughter confronts her father about a debt he hid from the family for a decade. She runs the scene through a tool to see a version of the confrontation, likes the shape of it, and pastes three lines almost whole into her draft because the rhythm works. Nothing illegal happened. But if a reader later asked her why the father says exactly that, in exactly that order, and she cannot answer because she never truly interrogated the choice, something has gone wrong that has nothing to do with the software and everything to do with ownership. The tool supplied a shape. She was supposed to supply the reason. If she cannot supply the reason, the scene is not hers yet, no matter whose name sits at the top of the page.

That is the whole test, really. Not where the words came from, but whether you can stand behind every choice in them as a decision you made and would make again. A published page is a series of claims: this happened, this is how the character would respond, this is the truth of this moment. You cannot make a claim you have not actually inspected.

The second line: honesty about the felt part

Here is where I draw a harder line than most. It is one thing to let a machine format your pages or catch a plot hole. It is another to let it manufacture the emotional truth and then present that truth as your own lived feeling. The first is using a tool. The second is a small lie told to the reader, and readers, strange sensitive animals that we are, can usually smell it even when they cannot name it.

The reason is not superstition. It is the same wall I described in the larger argument about what a machine can and cannot do for this craft. A machine can describe grief in flawless grammar because it has seen the word used correctly ten million times. It has never felt the floor drop. When you pass its assembled grief off as yours, you are asking the audience to trust a feeling that no one felt. That is the one thing empathy cannot survive being faked, and faking it is the ethical center of this whole question, not the legal footnotes.

Picture two versions of the same invented writer working on the same invented scene, a son sitting at his mother's hospital bed the night before her surgery. In the first version, the writer has never sat at a hospital bed at night. He asks a tool to write the scene, likes the cadence, and ships it. It reads fine. It is also faintly generic, because it is built from a pattern of ten thousand hospital scenes rather than from one specific, remembered silence. In the second version, the writer has sat in that exact chair, at that exact hour, and remembers the specific detail that the vending machine down the hall made a sound like a small animal every time it dispensed. He might still use a tool to tighten a line of dialogue. But the scene is built on something the machine cannot supply, which is a fact only a body that was actually there could know. The difference is not visible in every sentence. It shows up in the one detail that only a person who lived it would think to include.

The tool can help you say it. It cannot be the one who felt it. Do not let it pretend it was.Wr. Sarkhedi

The third line: know the ground you are standing on

Ethics is not only conscience. It is also the agreements the craft has fought for. When screenwriters organised in 2023, one of the guardrails they won was simple and worth remembering: AI cannot be credited as a writer, and it cannot be used to quietly strip a human of credit or pay. That was not a rejection of the tool. It was a fence around the human. The principle is portable even if you never join a guild. Use the machine to help you work. Do not use it to erase the people whose labour and credit the work depends on.

If you are a new writer with no connections, this matters more, not less. The people warning that AI is severing the entry-level bridge are pointing at a real harm, one I take personally, because the entry-level bridge is the only one I ever had. I wrote for years with no connections before anyone in this industry knew my name, and the jobs that let me eat while I learned, script coverage, junior polish work, uncredited pass-throughs, are exactly the jobs a tool can now do faster and cheaper than a beginner. An ethics that only protects the writer already inside the room is not much of an ethics. If you are senior enough to have a say in how a production uses these tools, that is precisely the moment to ask who is being quietly removed from the credit line, and to say something before it becomes the norm rather than the exception.

Where I have actually drawn the line, in practice

Theory is easy. The harder part is the specific moment where a rule either holds or bends. So let me be specific about a few of my own working rules, not as commandments for you but as the actual shape of a practice.

• Research, always fine. Asking a tool to summarise how a particular legal process works, or what a particular profession's daily routine looks like, is research. I would do the same with a book or an interview.
• Structural sounding board, fine with scrutiny. Asking whether a scene's logic holds, or whether an act break lands, is closer to what an editor does. I take the note. I do not take the rewrite.
• Line generation for a feeling I have not earned, not fine. If I have not sat with the specific grief, rage, or joy of a scene, I do not ask a tool to manufacture it for me and then sign my name under it as though I had.
• Anything that could displace a junior writer's paid work, not fine. If a task exists specifically because it is normally the first rung someone climbs into this profession, automating it away without a second thought is not efficiency. It is closing a door I walked through myself.

Common ways this goes wrong, even for careful writers

I have watched capable, well-meaning writers slip on this in a handful of predictable ways, and I have caught myself drifting toward a couple of them too.

• The slow erosion of the felt-part rule. It rarely happens in one leap. A writer starts by asking a tool to help phrase a hard emotional beat "just this once," and six months later the hardest, most personal scenes in the draft are the ones being outsourced first, because they are the ones that are hardest to write and the tool makes the pain of writing them disappear. The pain was doing something. It was the signal that the scene mattered enough to be worth the difficulty.
• Treating disclosure as optional because no one asked. The absence of a question is not the same as honesty. If you would answer differently depending on whether you were asked, you already know the answer is uncomfortable.
• Confusing legality with ethics. A workflow can be entirely within the letter of a contract and still be a quiet betrayal of the people around you, particularly juniors whose paid tasks quietly vanish.
• Assuming the reader cannot tell. Readers and audiences are more perceptive than writers give them credit for. The gap between an assembled feeling and a lived one rarely announces itself, but it accumulates, page after page, until the work reads as competent and hollow at once.

Why this is simpler than it looks

People make AI ethics complicated because the technology is new. But strip the novelty away and the questions are ancient. Whose work is this. Are you telling the truth. Are you taking credit or pay that belongs to someone else. Writers have answered these for centuries; the machine just changed the props, not the play. Guides from places like Writer's Digest keep landing in the same practical place, and so do I. There is also, underneath all of this, an older idea I keep coming back to, closer to the logic of karma than the logic of a terms-of-service agreement: the consequence of an action lives inside the action, not in whether anyone catches you. A shortcut taken quietly does not vanish because no one saw it. It sits inside the finished page, and eventually a careful reader finds it, the way you can always find the seam in a badly patched garment even after it has been ironed flat.

My whole position fits in one sentence. Use the tool to do the work you were always going to do, honestly, with your name meaning what it has always meant. The day the tool starts doing the feeling for you and you let the audience believe you felt it, you have not broken a new rule of the AI age. You have broken the oldest rule there is, which is do not lie to the reader. If you are weighing whether and how to say any of this out loud to your own readers or collaborators, I have written more specifically about when and how to disclose AI use, and if you are earlier in the argument, wondering whether a machine can write a script at all, that question is worth settling first at can AI actually write a screenplay.