I tried writing with AI. The results were surprising.

Original Source Here

I tried writing with AI. The results were surprising.

Photo: Vlada Karpovich/ Unsplash

“Don’t like to write, but like having written,” is a quote that has stood the test of time for writers. I’ve written my own book Creative Doing, I write at my blog every day, and I’ve led teams of writers and editors for years, so I actually really enjoy writing. However, I totally understand where the pain comes from. It’s an expensive pain, which is why ghostwriting can be such a lucrative endeavor; just a couple of weeks before I wrote this, Business Insider interviewed a ghostwriter that earned over $200,000 in 2021 writing tweets for venture capitalists.

Ghostwriting isn’t the only solution; there are plenty of others. One of the most exciting ones is the rise of AI writers. One of the latest named Lex, made by Nathan Baschez, has over 26,000 people signed up for it in its launch week. Lex is the latest in many other AI writing software, including Copy.ai, Snazzy AI, and ShortlyAI, all powered by the GPT-3 natural language processor. There’s even AI writing for programmers; over 90 million people use GitHub Copilot.

It’s a trend that’s been developing for years. K Allado-McDowell has coauthored two books with GPT-3, Almira Osmanovic Thunström at Scientific American prompted GPT-3 to write an academic paper, and Vauhini Vara wrote a story about her sister’s death, a topic she couldn’t bring herself to write about for years.

Last year, I tried writing with GPT-3, and it was incredibly unpredictable. Amid a lot of incoherence, and even occasional insensitive and even hateful text fragments, one particularly exciting response was GPT-3 likening creativity to LEGO blocks, which brought me back to my childhood. (I won’t be covering AI’s capability for machine learning hate speech, though it’s worth mentioning.)


Whether you’re writing as a business leader, a team leader, or for your personal brand, your work should enable people to trust you more; it matters because you’re the person who’s expressing the thoughts they’re reading. Generative technology is useful, so long as it unblocks your creativity and enables you to make work that people want to hear from you — your stories, your experiences, and your insights.

That means generative work with GPT-3 would, probably, still require a lot of writing from you; perhaps you’re prompting it, then responding to the text that it generated. It would look like turning a conference talk into a blog post; while this blog post started as a voice dictation, I ended up actually typing most of the thing out.

Sometimes, it’s only in rewriting a bunch of stuff that you realize what you actually wanted to say in the first place.


In his memoir, Working, biographer Robert Caro explains that he writes his first drafts entirely in longhand. “You think with your fingers,” literary critic and writing teacher R. P. Blackmur had told him at Princeton. While that speed worked well during Caro’s career in news media, it was too fast for him to think with depth in his career as a biographer; so, he returned to handwriting. Caro didn’t think to write; he wrote to think, and he knew that doing it slowly would make his thinking better.

It’s a sentiment echoed by Paolo Belardi in his book, Why Architects Still Draw. “You start by sketching, then you do a drawing, then you make a model, and then you go to reality — you go to the site — and then you go back to drawing,” Belardi quotes architect Renzo Piano. “You build up a kind of circularity between drawing and making and then back again.”

In my writer’s craft class, my teacher would describe another effect of slowness. He called it temporal distance, the ability to review or revisit writing with fresh eyes. In this case, the whole point of doing and working by hand is the slowness; not necessarily to go faster. Time, care, and consideration are the bare minimum to making something another person worth spending their time with.


A few years ago, my friend and former client Robleh Jama had his business acquired by Shopify, and asked me if my team could help launch its product blog. We’d use a similar process when I worked with him; I would sit down with the clients, discuss their work with them, and coach them through the process. I implemented this into my editorial studio called Wonder Shuttle, where we help technology companies make additional hires at lower costs, through sharing their team’s knowledge at their engineering blogs.

This is a tried-and-true strategy; the main problem is, writing is slow, and most people on the team don’t have that time. For a particularly slow writer, a well-considered blog post could easily take 20 to 40 hours. Moreover, priorities change, and blog posts often fall by the wayside for something more urgent and important.

One of the solutions we propose involves collaborative writing, which involves the client, a writer, and an editor. Here’s the gist of how it works:

  • A client comes up with an idea, or we source an idea. The author fills in an intake form, which they answer some questions: What the headline is, what’s the point they want to make, who they’re writing for, etc.
  • A writer and an editor from my team interview the client for an hour, write up an outline, get their feedback, make edits
  • The writer writes up a draft, the editor edits it, they get the client’s feedback and make edits
  • The writer and editor revise the draft, get the client’s feedback, and usually it’s good from here. If not, the writer and editor do another round until the client is super happy
  • Then, the client publishes the post and my team helps promote it

Clients can spend as little as a couple of hours cumulatively to have a finished blog post, written in their voice and tone. While this process is straightforward, it’s also incredibly high-touch; the client tells a stories and shares experiences during the interview, the writer filters out the noise and stitches together a coherent piece of writing, the editor carves and architects each piece.

The process is still slow; it takes at least a few weeks to get this done. During those revisions — which are also slow! — the writer or editor might notice a story doesn’t fit in with the intention of the piece, or the client might realize their voice and tone might not sound as professional as they like (or too buttoned up). That’s the whole point; the thinking isn’t just happening in one brain, but between at least three people’s brains, each with a different specialty and role.


One of the most well-known secrets in writing is to write a shitty first draft. With AI writing, putting the words on the page — idea generation — becomes even easier; that means the rewriting, revising, and restructuring will be the constraint for more people. The worse the draft is, the more editing needs to take place.

While there may be a day when I implement AI writing into my studio work, so far I still see the best working relationship involving people. After Garry Kasparov experienced defeat with an AI opponent, he came to appreciate the advantage of collaboration between a person and a machine. It’ll be the same with the collaborative writing process; writers may start using GPT-3, and editors may start using Grammarly’s or AxiosHQ’s editing software, but the substance inside the piece will still involve experiences, stories, and insights that people put forward.

Herbert Lui helps companies find additional hires at lower costs through sharing what they know at Wonder Shuttle. He is the author of Creative Doing, a book of 75 prompts that unblock creativity for your work, hobby, or next career. He writes a newsletter that recommends three great books every month.


Trending AI/ML Article Identified & Digested via Granola by Ramsey Elbasheer; a Machine-Driven RSS Bot

%d bloggers like this: