If you enjoy reading my writing as much as I enjoy creating it, please subscribe by clicking the button below:
(Note: In this essay, when I say GPT-3, that doesn't always mean that model. You can substitute other models or its successors. The models might change, the principles this article discusses don’t.)
Triggering GPT-3 in Lex was brutal. It felt like training my replacement before getting fired. I was excited to get Lex access, but I felt apprehensive for 500 words before triggering GPT-3.
My hands shook when I first triggered the AI with “+++”. A loading icon started spinning as the AI gobbled up and digested my writing. Dramatic as it sounds, those few seconds felt like minutes. My brain almost hoped the output to suck—so that I wouldn't have to worry about my future. The hope was futile.
But the circle stopped spinning. My robot
replacement assistant spat out accurate writing. And a jarring realization crystallized in my brain.
AI will transform writing and writers forever.
The above might sound like I hate AI. I don't. But it's changing everything. In this essay, we’ll explore how AIs transform writing itself and the profession of writing.
We'll discover its consequences and outline potential futures for (copy-)writers. We'll focus on utilitarian business writing like content, articles and copy. The principles might apply to disciplines like fiction and academic writing as well.
I want to share a nuanced perspective. I’m not a luddite longing for the good old days. Neither do I subscribe to Tech Twitter's utopianism where all technology is only great for everybody, always.
Every technological shift has up- and downsides for groups of professionals.
While VCs remove web3 off of their Twitter bios, pencil in AI/ML and request 🧵s from ghostwriters, let’s explore the future of writing!
Before exploring its consequences, we need to define artificial intelligence. The popular discourse misconstrues the term. AI has come to mean something like “Computers doing things I didn’t think computers could do”.
But artificial intelligence is machines doing anything that used to require human cognition. By that definition, the following are AI:
Chess computers (even if they lose)
Facebook friend suggestions
Photoshop’s magic brush
These aren’t exciting (anymore). But calculator used to mean a job, not an object. When you realize that, you see AI has been changing the landscape of work forever. But we move he goalposts of what qualifies as AI.
AI isn't new. What's new is that AI is coming for creatives who believed their jobs couldn’t be automated.
Let’s explore why writing is next. Here are a few properties writing AI has right now that make it so transformative:
It’s gotten good and useful!
Writing AI produces content indistinguishable from human writing. At least for simple topics with ample information online, GPT-3 is as good as many human writers.
It's also useful. Many technologies start as gimmicks before being useful. It's impressive that GPT-3 can generate an Emily Dickinson poem about washing machines.
Technologies usually decline in cost over time. But something magical happens when that cost approaches zero.
Think about graphic design. You could always get cheap design from overseas freelancers. But Canva let, anyone create good-looking graphics in minutes for free. So Canva template designs became pervasive in newsfeeds, event posters and Tinder profiles. That last one might only be me.
The same is happening with written content creation.
AI copywriting software Jasper charges $500 monthly for 700,000 words. Even if you found a writer to produce this amount at 1 cent per word (unlikely), that content would cost you $7,000 monthly!
While $6,000 yearly is cheap for most businesses already, few companies need 700,000 words per month. If you need a more reasonable amount, you can get your content needs met by an AI for less than $100 a month. Peanuts for most B2B clients.
GPT-3 is way cheaper than human writers. With AI solutions competing in the market, we’ll soon see the cost of content approach zero. That doesn't mean we'll pay nothing for content. But we'll likely pay for fine-tuned versions of models and integrations, not for words.
AI plummets the cost of content creation. And if AI tools improve , it’s safe to assume they'll soon deliver all the content you need at a tiny budget.
Near Zero Coordination
Writing projects often end up on the backburner. Website copy, content, SEO—these are often eternal projects reserved for "someday".
In most companies, the constraint for writing projects is rarely budget. It's coordination. More companies would revamp their website copy if they could press a button, get copy and pay an invoice.
Instead, they have to:
Find a copywriter
Sign a contract
Tell them about your company
Get feedback from team
Send invoice to finance
Much of this is redundant now that the magic button actually exists.
GPT-3 doesn’t only reduce writing labor, but coordination labor.
Revisions, hiring, interviewing, contracts, etc., are laborious. If you could skip those, more companies would execute their writing projects.
There you have it: AIs are cheaper, faster and often as good as human writers.
That begs the question: How will this impact writers? Let's first define why writing matters.
Every company is a combination of activities its team performs in pursuit of profit.
A B2B SaaS company might perform, the activities of shipping code, maintaining databases, designing workflows and interfaces...
The same applies to departments within companies. The marketing department is a set of activities performed by its members.
This lens is useful for sober analysis. Companies, departments, teams and even individuals are activities designed to produce an outcome.
This viewpoint is not comprehensive. Fuzzier concepts like culture and brand matter. But the model is useful to intercept the emotional reaction many have to automation. When we don't want something to happen, it's easy to manufacture theories about why it won't.
And writing is one of the most important activities in any marketing team.
This might sound crazy. Writers often earn little and have low status. Many see writing as an entry-level profession and escape for management or strategy. But writing matters more than most strategy or management activities.
Let’s make this concrete with an example. Imagine you’re in a mid-sized company and plan to publish a blog article. Here’s what the chain of activities may look like (excluding meetings, messaging, etc.):
Social media sharing
Regardless of pay and status discrepancies, these activities have a hierarchy. How might a solo founder tackle this? In most cases, the process would look like this:
Fewer resources force you to reduce to the essentials. The activities you can’t skip. Not writing and publishing is like multiplying by zero. The rest of the variables don’t matter—the result will be zero.
You can publish without a coherent strategy and get a result. You can publish unedited writing and you'll still have something to show. But your content strategy becomes worthless if nobody writes the content you planned.
This is shows how important writing is to marketing. Whether it’s a blog article, a new website, a social campaign—skipping writing sets it to zero.
If we analyze this activity chain, we can enrich the data. Each activity has a place it’s performed and a responsible person, e.g.:
Activity: Writing Place: Google Docs Person: Writer
Activity: Creative brief Place: Notion** Person: **Creative Director
A blog article is like physical items moving from factory to warehouse to retailer. Content creation is a process that happens in various places and people. This is what I called knowledge logistics in Work is Software:
Knowledge work may not involve physical shipping, but it involves logistics.
The above example is for a blog article. But it works for any marketing effort.
The changes writing AIs will bring will show up as a change in the activity chain. It will change who performs the writing activity and where. Writers are currently core to these processes. But as these chains change, they'll affect the day-to-day of professional (copy-)writers.
Let’s outline the most likely scenario:
Content marketing isn't only one place where writing is mission-critical. Lots of business software becomes useless when you have no copy/content to feed it:
Email marketing provider
Social media marketing softwares
What happens when we can AI-generate content for all these? Many believe less skilled, folks armed with AIs will replace writers. They'll generate writing and paste the copy over to these tools.
In our value chain, this would mean the person would change to somebody less well-paid while the place in which the writing activity is performed would change to some AI tool.
I don’t believe that will happen. Instead, the changes will look different. The writing activity will disappear and the person will change to whoever manages the channel, while the editing activity will be performed on auto-generated copy within the platform.
Instead of introducing a new place, an adjacent activity absorbs writing by integrating AI within their platform. As Evan Armstrong points out, fine-tuning is an important part of using (writing) AI.
Software providers are most likely to capitalize on this. They can fine-tune the AI to produce output for a specific use case.
Imagine you’re running an email marketing tool ActiveCampaign. You can fine-tune GPT-3 to produce copy of a certain length and format. Then you feed it the customer’s best-performing emails as source material. Now you bundle it into your service or charge extra as an upsell.
Tools like ActiveCampaign already have preset automations. Imagine those auto-generated copy when you used them. You'd only have to approve or edit the copy. If you assigned this to a writer, you would wait for the copy, then request revisions and wait again. With AI, your automation would be live in hours, not weeks.
In the value chain, this would condense the process. Writing would join the editing phase and be done within ActiveCampaign.
Email marketing is one example. You can imagine writing AI in every software that needs writing:
Your SEO software could generate content for any keyword you’re exploring.
Your social media suite could auto-generate content based on your existing content.
This is a brutal scenario for writers. AI would remove every trace of them from the process. Former writers either become editors overseeing AI or become channel managers.
That doesn’t mean writers will vanish. But the writing profession will change.
Many low-end writing jobs will go to AI, not humans. But even outside of lower-level work, (copy-)writers need to adapt to changing realities.
The value chain dictates that your best bet is to elevate yourself to an adjacent stage. Creative direction and nuanced branding decisions are harder to automate.
AI might reduce total writing gigs. But automation doesn't tend to create mass unemployment. Automation tends to change the types of employments.
These are a few ways writing professions may morph into:
Before AI, writers performed different functions within companies. That won’t change with AI. But the types of activities writers perform will change. Here are a few roles AI-augmented writers will play.
Good writing is only useful if the strategy is. Great SEO is useless if your chosen keywords don’t align with business objectives.
Writers need to understand how their work plugs into strategy. You need to know what you say and why you say it before how you say it.
Good copywriters have long added value by helping out with strategy. AI makes executing a strategy easier and cheaper. This enables more writers to capture the strategy activity.
When creating content is more accessible, more people do it. This means lots of people will create content without the skills to make it effective. Pre-AI (Copy-)writers will understand the thinking behind the copy. They'll advise companies on messaging strategy and oversee the AI content generation process.
The Source Material Creator
GPT-3 generates generic business writing well. Here's a snippet with the prompt “how to create an effective marketing strategy”:
“To create an effective marketing strategy, you need to understand your target audience, your product or service, and your competitors. You also need to set realistic goals and objectives.”
This might be enough to fool Google crawlers, but it’s and boring. Differentiated brands wouldn’t use this. It’s not that GPT-3 is bad. The prompt sucks. If you want better results, you need to give AI good source material to work with.
Don't fire off a generic question like “How to create an effective marketing strategy?”. Provide more context:
Who is your audience?
What do they care about?
What features are unique to your product?
What medium is this for?
AI results improve when you feed them more specific info:
Good source material requires knowledge of the industry, company and its positioning. Complex, data-rich prompts are hard to create. They flow from the messaging/strategy piece I describe above.
Writers who can tell good writing from bad will be best at defining what the AI should react to.
AI makes writing almost free and instant. Even so, it won’t eradicate human writers. In the age of 3€ IKEA mugs, people still pay 30€ for a handmade one.
As Nathan Baschez points out, writing is changing. But we won't completely remove writers from all processes.
IKEA didn’t eradicate ceramics studios. McDonald’s didn’t kill home cooking. And GPT-3 won’t exterminate human writers.
But mass production did reposition the handmade mug from a necessity to a luxury. Our emotional experience with handcrafted goods differs, even when features are the same. It’s not about utility, but the story we tell ourselves.
The same will be true for hand-crafted copy or content. Some companies hire humans to write bespoke copy and content, even if it costs 10x as much as an AI-driven team.
But these artisans will need to offer more than words in a Google doc. This will revolve around helping founders feel connected or as a flex (signaling).
Humans will also write for complicated industries and new categories. Simplifying complex science is harder for AIs than rephrasing 7 ways to start a business.
The Prompt Engineer?
Some are predicting companies will hire legions of “prompt engineers” or “prompt creators”. It seems obvious. New tools demand specialists who maximize their utility.
General AI rarely gives you the precise output you want. That’s why people use long prompts. There are even prompt marketplaces! When you buy a prompt, you get a set of AI inputs you enrich with specifics.
But prompt engineering won't be a common job in most companies. The UX of copy-pasting 30-word prompts is awful. Instead, there will be fine-tuned front ends that have prompts baked it.
Some former writers may be engineering prompts. But they'll likely work for AI software providers, not companies with AI in their workflow. Writers can best judge the AI's output, which is why some may flock to this.
These are all examples of shifts in the activity chain. As I explained in the beginning, this is a reductionist, but useful take. But shifts like these always have a human component.
I recently went on a date with an illustrator. Her main AI worry wasn’t about income or job security. Instead, she surprised me when she said this: “When these AIs can illustrate, who’s going to need me?”
Automation started with farming, continued with manufacturing and keeps going with knowledge work. Centuries-old fears of automation-driven mass unemployment have never manifested, but they create distress.
“Who’s going to need me?” goes deeper than a paycheck. Whether it’s writers, illustrators or developers: Humans want to feel needed. We want somebody to shed a few tears if we vanished tomorrow.
We want to feel valuable, appreciated and competent. If society tells us we're not needed anymore, a coding boot camp won't fix that.
Former typesetter Anastacia Moore writes on Quora about the advent of computers:
As a typesetter, I also developed excellent typing skills (speed), and a very extensive knowledge of the 'inner workings' of these various typography machines, thus making the transition to computer typography seamless. What a jump from those old methods of rubber cement, wax, vertical and horizontal cameras, negatives, stripping, masking, platemaking with the advent of computers! I LOVED IT . . . . . . . until I came to the realization that NOW, anyone that could figure out how to run all these wonderful programs like Pagemaker, Quark, Photoshop, Indesign, Illustrator, etc. was making the 'profession' of the typesetter/graphic artist much less of an 'art', thus overall lowering the wage for having a skillset that few had mastered. This new age where just any old Joe can go out and buy the aforementioned programs, slap them on their computer and suddenly call themself a graphic artist has, in my personal opinion lowered the standard for a profession that back in the old days required an extremely wide range of knowledge and technique.
Note how money isn’t the focus and only mentioned in passing. It’s the skills she’s learned, the competence she felt and the status her skill conferred!
It must be crushing to spend your career perfecting a skill, only for that skill to become a party trick.
When a core source of self-esteem evaporates, it's a disaster for mental health. We’re already living through a mental health crisis. Making hordes of tech workers feel useless only exacerbates this.
Many writers will find other careers or level up to write things AI can't (yet) produce. But for every success story, there will be a freelancer unable to pay her bills. There will be a copywriter close to retirement who has a hard time learning new skills.
We can't stop the rise of AI in writing and other fields. But let’s be mindful of the people it impacts.
As Anastacia Moore concludes her Quora response:
'We' the professional typographer . . . are dinosaurs now, and for those who have not adapted to modern technology, have more than likely gone the way of the dinosaur . . . never to be heard from again.
That’s a wrap! If you enjoyed this essay, please collect it below to show your support and add it to your collection! Or, if you’re reading this in 2030, tell the AI assistant that read this to you to collect it.
If you enjoyed this essay and don’t want to collect it, please help it reach more people by liking and retweeting this tweet: