How to Actually Use AI in Your Writing Workflow

Most arguments about AI and writing are really arguments about whether you should touch the tool at all. That is a waste of a good morning. The tool exists, it sits on your desk, and pretending it does not is like a batsman refusing to look at the scoreboard because he dislikes the total. The useful question is narrower and far more practical: where in your real process does the machine help, and where does it quietly wreck the thing you are trying to build.

I have written 21 books, more than 2,000 articles, and six screenplays that sit registered and waiting. I use these tools most weeks. I also throw out the majority of what they hand me. Here is the workflow I actually run, stage by stage, so you can take the parts that fit your desk and leave the rest.

Sort the work into two piles before you open anything

Before I touch a tool, I split the task in two. One pile is mechanical: formatting, summarising, spellchecking, generating ten flat variations of a beat so I can feel which direction is alive, catching the continuity slip I am too close to see. The other pile is the actual writing: the felt moment, the line only this one character would say, the scene that has to cost something. The machine is welcome in the first pile. It is banned from the second. Get that boundary wrong and you will produce pages that read smooth and lie dead on the table, which is the exact failure I warned about in the larger argument about AI and screenwriters.

Where the machine earns its place

Used honestly, the tool is a fast, tireless assistant, not a co-author. I use it to break a logline six ways at once so I can react instead of stare at a blank line. I use it to pressure-test structure, to ask what act two is missing, to build a quick character questionnaire I then answer myself. Outlets like No Film School and Writer's Digest keep circling the same word for this: assistant. Not the writer. The intern who never sleeps, never complains, and whose work you always check line by line before you trust it.

Think of it like the engineering I left in 2015. A calculator does not design the bridge. It clears the arithmetic so the engineer can spend her attention on the load, the risk, the joint that actually fails if she gets it wrong. The tool clears the arithmetic of writing. It does not decide what the story is about, and it never will, because it does not know what it would feel like to be the person on the page.

The tool can carry the bricks. It cannot decide where the house should hurt.Wr. Sarkhedi

Where it will quietly ruin you

Here is the trap, and it is a soft one, which is why it catches good writers. The machine is most dangerous exactly when it is most helpful. You ask for a line, it gives you a decent line, and because the decent line is right there, warm and finished, you stop reaching for the true one. Decent is the enemy. Decent is what a million other people will also generate from the same prompt tomorrow. The scene that makes a stranger cry is never the first decent option. It is the specific, slightly wrong, lived thing that only you could have known, because only you were in that hospital corridor at 3 a.m.

So my rule is blunt. I never let the tool write the emotional beat. I let it clear the ground around the beat. The beat itself, the moment the whole film is actually for, stays in my hands, because that is the one thing a pattern machine has never had. That is not sentiment. It is the working core of why empathy cannot be faked.

A loop you can run tomorrow

Stripped to the bones, this is the cycle I use on every project:

• You decide the want. Who wants what, and what breaks if they fail. The machine never starts this. You do, from something that matters to you.
• Let the tool widen your options. Ask for many rough versions of a beat, a title, a structure. Treat them as a mirror, not a manuscript. You are feeling for the live one, not adopting the smooth one.
• Write the real scenes by hand. Every moment that has to land emotionally, you write yourself, out of something you have actually felt.
• Send it back for the mechanical pass. Continuity, repeated words, a bloated paragraph, a scene heading you forgot. Honest work for a machine.
• You keep authorship. You decide what stays. Every choice on the page is yours to defend to a director's face.

Let the machine do the typing you would not miss. Never the feeling you could not fake.

Speed was never the point

People sell these tools on speed, and speed is real, but speed is not why you use them well. You use the machine's speed to spend it, to buy back the hours you would have lost on the parts that were never the point, so that every hour of your own attention lands on the part that is. A faster route to a hollow script is not a win. It is just a hollow script that arrived early.

When I close a session, I ask one question. Could anyone with the same prompt have produced this page? If the answer is yes, I have not written anything. I have assembled something, and assembly is exactly what is about to become free and infinite. Learn the craft until the mechanics are second nature, and the tool stops being a crutch and becomes what it should have been all along: a way to reach the hard, human part faster. If you are trying to break in with no connections, this discipline matters even more, and I wrote about that road here. The workflow is not about doing less. It is about making sure the only thing left in your hands is the thing that was always yours.

What a working day with this loop actually looks like

Picture an invented but representative scene, the kind I have written some version of more times than I can count: a father waiting outside a hospital ward. I open the file with nothing but the want written at the top, he needs to hear that his daughter will walk again, and he does not believe he deserves to hear it. That sentence is mine, and no tool touched it. Then I open a chat window and ask for ten different ways a nurse might deliver bad news badly, structurally, so I can see the shape of the wrong version before I write the right one. That is the machine widening my options, and I read all ten in under two minutes, keeping none of them, but now I know the shapes I want to avoid. Then I close the tool completely and write the scene by hand, in my own broken, human draft, because the father's actual line, the one that will land, has to come from somewhere I have actually stood, waiting on a hard floor for news that might break me. When the draft is done, I hand it back to a tool for the pass that bores me most, checking the slug lines, catching the repeated word, flagging the moment where I gave the nurse a line she already used in an earlier scene. That is the mechanical pile, cleared honestly, and it buys me the better part of an hour I would otherwise have spent proofreading instead of writing the next scene.

The two-question test I run before I trust a page

I run this test badly on tired days, and I know it, which is why I built it as a checklist and not a feeling. On the days I skip it, I can tell within a week, because the pages that came out of a rushed session read fine on the first pass and then curdle on the third, the way food looks fresh under the wrong light. The fix is never dramatic. I go back, find the two or three lines that were only decent, and rewrite them slowly, by hand, asking what this particular character, in this particular room, would actually say if nobody was listening for a clever line. It rarely takes long. What it costs is patience, the same patience I wrote about needing for the discipline of a daily writing practice, because a workflow only holds up on the days you do not feel like using it properly.

Why I still insist the hard part stays human

There is a deeper reason I hold this line so stubbornly, and it is not only about craft quality, though that would be reason enough. It is that writing the hard part by hand is one of the last disciplines left that forces me to actually feel the thing before I hand it to anyone else. If I let a tool write the father's line at the hospital, I skip the part where I sit with what it might feel like to fear the very news you are praying for. That sitting is not wasted time. It is, as far as I can tell, one of the only reliable ways left in a busy life to practice empathy on purpose, on a schedule, whether or not you feel like it that day. The workflow is not only protecting the script. It is protecting the one habit that made me capable of writing anything worth reading in the first place. If you want the sharper edge of that argument, on where the tool's fluency actually breaks down in dialogue, I wrote it out in why AI dialogue always sounds slightly off, and if you are earlier in your career and wondering whether any of this threatens your entry point, I addressed that directly in what AI actually changes for entry-level writers.

Questions writers ask about using AI honestly