This is the last post in a series on AI and teaching.
I purposely left the topic of teaching writing for last, because while working on this series of videos and posts, I very consciously looked for ways that AI could help me save time and effort.
In this last segment, my hope was to be able to share a few tips based on my own experience writing and making videos about this topic.
First the good news: there were ways that AI helped me save time and effort.
But also… this series on AI and teaching has taken more time and more work on my part than almost any other topic I’ve ever covered.
That tension, it turns out, makes a pretty good jumping-off point for talking about how to teach writing in the age of AI.
The Time Alignment Problem
If you’re in charge of teaching students to write, you are running into a whole series of issues right now. But they all fall under the umbrella of something I’ll call time alignment.
We are intuitively used to the idea that it takes longer to write something than it does to read it.
This has been true pretty much since the beginning of writing: When you read an email, you’re assuming that the person spent a little longer typing their thoughts into words than you are spending reading it. When you read a book, it’s going to take you a couple of weeks to get to the final page, but the writer likely spent years on it.
The value in a piece of writing comes from this unspoken contract between the writer and the reader, that the writer has done the majority of the mental processing. And as soon as you sense that this is not the case, the relationship between writer and reader changes.
This is one of the reasons AI-based cheating is so frustrating to teachers.
The idea that you are looking at an essay you thought a student created. . . and then, little by little, you start to realize you are marking up something that was created by a computer. . . in seconds. . . is just a gut punch.
This experience is demoralizing and frustrating.
It sets you up for a confrontation with the student.
But it also reflects the fact that we are suddenly in a new world, both inside and outside the classroom, where our understanding of writing has been flipped on its head.
We now have a way of producing words faster than any human can mentally process them.
And often those words sound smart. They can be coherent. They can even be useful. But those words are not writing.
Writing is not a process of producing words. Writing is a process of refining your thoughts so that you can hand them off to someone else.
Stephen King, his well-known book On Writing, makes the case that writing is a form of mental telepathy.
You’re transferring the thoughts inside your head, across time and space, to another person somewhere out in the world.
So the question is: if that’s your goal, is that goal still necessary and possible in the age of AI? And is there some room for AI to help with it?
I would say the answer to all of those questions is yes—as long as you keep your eye on that goal.
What AI Can and Can’t Do for Writers
One of the ways AI can help us communicate better is by bringing things into our skill set that would have previously been impossible—or at least required help from an expert.
In a writing context, this might include some aspects of research or fact-checking, help with visuals, or communicating in a new type of media.
In my own case, I never even considered creating videos until a few years ago.
That’s when I learned about my video editing program, which uses AI transcription to create an instant transcript of any video.
Instead of looking at a series of squiggly wave forms, I was suddenly able to edit videos directly from a transcript. And suddenly, video editing felt very much like the writing skills I’d for decades.
That’s an example from the more successful end of how AI has helped my communication. Now, here’s an example from the less successful end.
While working on this series of videos about teaching and AI, I used AI to try to generate a list of ideas for videos I might want to make about the topic.
What I ended up with was a list that looked good at first glance.
It was coherent, grammatical, and authoritative. Most AI-generated content is.
But when I tried to use it, I found myself stuck in an unanticipated time-suck.
Out of 20 seemingly ideas, there were (maybe) one or two things that (possibly) sparked something I could have (potentially) used.
As I attempted to elaborate on the most promising items on the list, it became clear that even these were not as useful as they seemed.
If I had pursued any of those topics, I would have been sinking my own human work time into building out something that was basically just work slop.
AI wouldn’t have saved me time; it would have cost me a tremendous amount of time just to arrive at an end product I wasn’t particularly proud of.
These two examples sit on different ends of the spectrum.
On one hand, AI can feel like it gives you superpowers. On the other hand, it can take the ball and run with it in a way that doesn’t reflect what’s in anybody’s head, let alone yours.
And then there’s real work in walking back the thought process to where AI got away from you, where it turned your written work into something that isn’t really writing.
The Three-Part Writing Process
One way to approach writing in the age of AI is to walk students through from thought to finished product and ask: What are the parts of this that you cannot skip if you want to make sure you’re really writing?
There are a lot of different ways that teachers break down the writing process, and all of them can work. But all the steps of all those versions can generally be grouped into three categories:
- Knowing what you want to say
- Saying what you want to say
- Making sure you said what you want to say
These categories don’t necessarily go in order. You jump around between them as needed.
Because, again, writing is a process of refining your thoughts.
Step 1: Knowing What You Want to Say
Before you can share your thoughts, you need to have some idea of what those thoughts are.
Are there ways AI can help you if you’re stuck at this stage?
Yes. It’s not hard to get AI to generate something that looks like a brainstormed list of thoughts.
But you have to be careful here, because you may actually have less work in front of you when you’re staring at a blank page than if you first have to sort through 20 AI-created suggestions where 19 and a half of them are just garbage.
And then, after you do the work of figuring out which 19 suggestions are useless, what do you do with that half-suggestion that might work? Is it really what you wanted to say? Or is it kinda-sorta-adjacent to what you wanted to say? Except for a few small tweaks, which then turn into an intensive rewrite because, actually, this was farther from what you wanted to say than it initially seemed?
That whole process can end up taking more time, more effort, and producing a lower-quality final product than if you had just given yourself some time to stare at that blank page.
The first part of any writing process happens inside your head.
And it can be difficult for anyone, including AI, to know what’s going on in there unless you put it into words.
Which is writing.
Step 2: Saying What You Want to Say
This step can include everything from getting an outline on paper to creating a first draft.
The point is that the thoughts inside the writer’s head are coming out into the world where the writer can now work with them in some way.
Here, too, there are ways that AI can help a writer who’s feeling stuck. But here, too, you have to be careful that AI does not take the ball and run with it in a direction that doesn’t reflect what you’re trying to say.
As an example, here’s how I used AI in this stage for this very piece of writing.
As I often do while writing on a specific topic, I’d been writing down thoughts as they came to me in a long, stream-of-consciousness Google Doc.
Some of these thoughts had already ended up in the video I was making. The thoughts that were left over started to seem like a pile of jumbled, repetitive notes. I was having trouble figuring out how these thoughts came together, and had a hunch that some were completely off topic.
Looking at these leftover notes exhausted me. As with the AI-generated list, it was clear that more words on the page did not mean less work ahead.
In an attempt to figure out what was worth using, I turned to Claude (Anthropic’s AI) and typed in this prompt:
I’m going to paste in some brainstormed notes for a YouTube video on how AI can and can’t help with the writing process. Use these notes to come up with a framework I can share that goes step by step through the writing process and explains how AI can and can’t help. Use as many of my original ideas as possible, but you don’t have to use all of them. If something doesn’t fit the outline and you leave it out, tell me at the end what you left out.
The best thing I can say about this is that it absolutely got me unstuck.
AI organized my information into five sections, each of them complete with subsections on “how AI can help” and “what AI can’t replace.”
It also provided some categories of ideas it recommended leaving out, most of which I agreed with, and one of which it entitled, Bad AI Advice Rant, which was accurate.
I didn’t end up using the organization that AI recommended, but seeing my notes organized this way helped me realize there was a way to organize them, and that they contained some ideas worth using.
So, was AI helpful in organizing my thoughts? Yes. Did it get out in front of my thoughts and start to make me say things I didn’t quite mean? Also yes.
Most of what AI produced was close to my original thoughts, but didn’t match them exactly. In fact, there were a couple of moments where AI came up with a really good way of phrasing something, and it was only when I tried to elaborate on that phrasing that I realized: this is not exactly what I mean to say.
In one instance, it took me several reads to realize that a whole chunk of AI output not only wasn’t what I wanted to say, but was leading me in the wrong direction. And it was because it was so well-phrased that it took me a while to notice.
This brings us to one of the big new challenges of AI writing: Decisions are work.
This is an issue that shows up in AI coding too, where people are “vibe coding” projects that initially seem to work but can cause real damage without vetting.
The issue also shows up in science, where AI can suggest magnificent-sounding research projects that could cost hundreds of thousands of dollars and years of someone’s life.
Decisions are work.
Exercising judgment is work.
Narrowing down possibilities is work.
And writing is just one decision after another.
The thing to emphasize to students here is that every decision in your writing should be made by you.
You know how sometimes you’re sending a text message and autocomplete gets involved and finishes your sentence?
And sometimes it’s a version of what you wanted to say, but sometimes it’s really, really not what you wanted to say?
It’s probably worth thinking of AI as an autocomplete that can write three pages for you.
You need to be even more careful to make sure you’re processing all of that.
With a three-page document that has had a version of autocomplete involved, if your hands are not on the communication wheel, you might be in a situation where you are not really writing at all.
Step 3: Making Sure You Said What You Want to Say
What you’re really doing in this final stage is trying to read your own work through the mind of your reader; you’re doing your best guess of how your work will be received.
This is where you ideally catch any issues with spelling, punctuation, and grammar.
But on a more important level, this is where you’re doing one of the things that writers have always done on behalf of readers: they try to put themselves in the reader’s perspective and imagine how the work will land.
Will the reader understand it? Are there details that need to be explained?
And again: are these words, strung together in this order, what the writer wants to say to this reader?
Does this mean you cannot use AI for editing help? Not necessarily.
AI can often like a first set of eyes on what you wrote. It can sometimes give students personalized feedback that a teacher simply doesn’t have time to provide. And it can definitely help catch some of those “own goals” in terms of spelling, punctuation, and grammar.
But AI can also take your original thoughts and original phrasing and turn them into something that sounds kind of like you, but not quite.
It can phrase your ideas in a way that is kind of what you want to say, but not exactly.
There’s just no way around circling back to that final question: Is this what I want to say?
And if it’s not quite right, you’re back at step one: What exactly is it that I want to say?
Handing Off Your Thoughts to the Reader
Boiled down to its essence, the writing process is really cycling through the three steps above over and over again until you’re ready for the fourth and final step: handing off to the reader.
This can mean turning in an assignment, hitting send on an email, or publishing an article.
But in all cases, this is where the writer’s mental processing of the material ends.
At the hand-off point, the writer turns over the effort but they also turn over the control.
It is now up to the reader how they’re going to take in this message.
If the stakes of miscommunication are high, this is where the writer may find out that their attempts at communication failed.
Which why students are learning writing in school.
That communication muscle needs exercise to make it stronger, and teachers are kind of like the personal trainers who help students do that.
In keeping with that metaphor: AI can sometimes act as a spotter.
It can help you do a level of heavy lifting that a writer can’t quite do on their own.
And it’s probably worth telling students: if there’s a roadblock that’s going to keep you from completing a piece of writing in the first place, it’s worth seeing if AI can boost you over that hurdle so you can keep aiming for your goal of getting your message across.
But please don’t let AI do the exercise for you: There is no substitute for the thinking process that gets those thoughts out of your head and into a form that makes them usable to another person.
As I mentioned above, this series on teaching in the age of AI took more time and more effort than I originally anticipated.
That’s probably because this is something we’re all right in the middle of, and I’ve been working through my own thoughts on this topic as I’ve been writing about it.
My hope is that that effort comes through in a way that is valuable to you as the reader.
But that is for you to decide.
Because the moment I hit publish on this blog post, my part of the writing process is over; I’m handing it off to you.