To Write Better, Use AI to Write Faster

What’s more valuable: Writing fast, or writing well?

Thanks to GPT and other generative AI tools, you don’t have to choose. In fact, you may find that writing more quickly (with the help of Al) lets you produce much better writing.

 

Turning speed into quality

The secret lies in knowing how to turn speed into quality. When you bleed for a 1,200-word draft, or spend hours and hours to write a single page, it’s incredibly painful to admit that something’s not working, that a paragraph you sweated for needs to be cut, or that your whole post or article has headed in the wrong direction.

When you can toss off a draft very quickly, you can bear to change it, or even to start fresh. You can take the time to set your work aside, and return to it with fresh eyes; you can listen to tough criticism, and take feedback that calls for major changes, without wanting to weep. You can even write multiple drafts of the same document, and then pick the best parts from each version.

That’s what I’ve learned from my many years as a relatively speedy writer. It takes me about an hour to write a 1,000 to 1,500-word first draft, and it’s not unusual for me to write 5,000 words in a day.

But what’s really useful about being a fast writer isn’t measured in word count. It’s measured in the way speed translates into steady improvements in each draft, and in your own writing style.

And that’s what generative AI makes possible.

 

Expand your toolkit

To make the most of faster writing, you need the right tools. My latest for The Wall Street Journal  recommends complementing your trusty word processor with specific apps for idea tracking (Coda and Evernote), outlining (MindMeister) and long-form writing (Scrivener). I particularly recommend Coda and Scrivener as apps that will help you make the most of faster writing, because they make it easier for you to slice up and rearrange an imperfect first draft.

And of course, you’re going to need access to a generative AI tool like ChatGPT. Spring for ChatGPT Plus, because access to its latest model (GPT-4) will give you dramatically better results.

 

Write badly

Yes, I mean that! To write faster, give yourself permission to write a lousy first draft—or give that permission to GPT.

Don’t get stuck staring at a blank screen until you can write something really good, and don’t get hung up on tweaking a single paragraph until it’s perfect. Just puke up the lousiest first draft you or the AI can muster.

 

Revise, revise, revise

Writing badly only works if you then move onto making your first draft better. Do not use the robot’s draft as your final draft; for now, GPT’s writing tends to be stiff, jargon-filled or (if prompted to write casually) a bit wacky.

AI can speed up this part of the process, too, by making it easier for you to just relax.  Don’t worry if you’re adding too much, or if your arguments are in the wrong order, or if your grammar and spelling are incorrect; all of these are easy for GPT to find and fix.

Instead, focus on the edits that simply move the work forward.

Do you have a personal anecdote or hypothetical example that would clarify a portion of your argument? Add it!

Did your AI generate a particular word or phrase that doesn’t sound like you? Replace it with a sentence you write yourself.

And of course….

 

Kill your darlings

(but bury them in a shallow grave)

“Kill your darlings” is a famous bit of writerly wisdom. It means that if you find yourself really charmed by a phrase, paragraph or section of your own writing, you should strongly consider cutting it—because that attachment may be a sign you’re not seeing own work clearly. And if you’re charmed by something the AI wrote for you, run that phrase through Google to make sure you’re not accidentally plagiarizing an existing work.

It’s a lot easier to cut viciously if you hold onto all the cut bits. If you’re writing in a word processor, create a parallel file that has the same name as your draft doc, plus a suffix like “cuts” or “dumper”. (Use the same suffix every time, and you’ll always be able to track down text for future re-use.) If you’re using a slicing-friendly app like Scrivener or Coda, create a sub-section just for all your cut-out pieces.

 

Start fresh

Once you develop the habit of writing quickly, or using GPT to generate quick drafts for you, you can open the door to a very liberating possibility: Throwing out your drafts and starting over. It’s often easier to start anew (even without GPT!) than to keep revising what doesn’t quite work.

 

Ask for an edit

In all my years of fast writing, nothing has helped me as much as the opportunity to work with great editors. As I have told all my favorite editors, an editor is like a therapist who pays you, instead of the other way around!

But you don’t need to work with a professional editing team in order to experience the power of a second set of eyes. Once you think you have a pretty good draft, give it to someone else for feedback—and the first “someone else” can be GPT.

If you really want to level up your writing over time, I recommend that you work with some human readers, too. Find someone whose writing you like, and see if you can become one another’s first reader—the person you show your work to for initial feedback. The more you write and rewrite with the guidance of a second opinion, the more you will improve.

 

Prompt advice

To ensure that GPT (which most people access in the form of ChatGPT) makes you not only a faster writer, but a better writer, you need to power your multi-draft writing approach with the right prompts.

You can say something as vague as “draft a blog post about work-life balance”, but you’ll get better results with a specific request like…

draft an 800 to 1000-word blog post about work-life balance for a corporate blog, focused on the productivity benefits of improved focus when people get enough sleep and exercise, creativity that comes from exploring outside interests, and reduced distraction when people aren’t worried about whether they’ll be able to get home to their kid, dog or band practice on time.”

In either case, you can expect a draft in under a minute, and particularly if you’re using GPT-4 (the latest model) you might be surprised at how decent that draft can be.

Here’s what I’ve found effective at each stage.

 

The first draft

If you’re using AI to get underway, take no more than five minutes to jot down whatever you think you want to include, and then feed it to ChatGPT with a request for a first draft.

It helps to prime GPT with examples of your past work, so it knows your voice; I’ve got a single chat in my ChatGPT history that I return to whenever I’m working on a newsletter, because I’ve already trained it on a few examples of past newsletters.

That past training means I can now feed a few very rough bullets to ChatGPT and get a decent first draft, just by framing those bullets with a prompt like:

You are a ghostwriter aiming to write in the voice of Alexandra Samuel (author above). Your goal is to write a newsletter that most closely mimics her voice and perspective. The title of this newsletter is, “To Write Better, Write Faster: How AI unlocks the key to better writing”. It should include the following points…”

And you don’t have to settle for the first first draft: I often revise my prompt several times, adding in points or telling GPT to avoid certain missteps (like “avoid business jargon”), before I get a draft that feels like a worthy starting point for my own efforts.

 

The third draft

Once you’ve crafted a second draft by expanding on, correcting or amending GPT’s first draft, you can ask GPT to feed it back to you as a more coherent third draft. Consider prompts like…

Revise this draft to 1,000-1,200 words, but don’t cut any specific anecdotes or examples.

Suggest three ways I could restructure this draft so that the argument flows more logically. You don’t need to provide the complete draft for each version: Just suggest options for how I can re-order the subsections, annotating each subsection title with a line explaining why you think the logic might flow better this way.

You are the marketing executive in charge of an e-commerce site. Your goal is to edit this submission to correct all grammar and spelling errors, and to revise the style and details to maximize purchases and social media re-shares.

The fourth draft

Yes, I said fourth. Once you’ve got a robot army, you don’t need to be afraid of going to draft four (or further, if you prefer to iterate rapidly and see what sticks!)

Start by looking at all your previous drafts, and thinking about what works in each one. In some cases what works might be something big picture, like the overall voice or a key part of your argument. In other cases you might have specific phrases you liked, or anecdotes that work. The fourth draft is the “greatest hits” draft that takes everything that’s working—and cuts everything that isn’t stellar.

Feed all these ingredients back to ChatGPT as a new prompt, bearing in mind that it may take several messages to give it all the context it needs, because GPT chokes if you feed it too much text at once. I like to use “STAND BY” as an instruction so that waits for all my bits and pieces. (I have no idea if capitalizing makes a difference to GPT, but it helps me spot the sections in my own prompts.)

So for example…

You are a ghostwriter combining three draft blog posts into one blog post on the subject, “Why hybrid work is here to stay.” It should be 1000 to 1200 words. I will feed you the three drafts in turn. Please stand by between each draft and return a new draft only when I say “PROVIDE DRAFT NOW”. Combine the drafts as follows: DRAFT 1: Use the insurance office anecdote plus the argument about competition for talent.

DRAFT 2: This has the closest voice to what I want for the final draft. Be sure to use the phrases “we’ve seen a seismic shift in how we perceive our professional lives” and “balance isn’t a luxury anymore”. But do NOT use the phrase, “in conclusion”, or do anything else like that which signals the sections of the post in a heavy-handed way.

DRAFT 3: This was basically the right set of arguments but it had too much business jargon. Make sure all the key points in this draft make it into the new draft.

Now, STAND BY FOR DRAFT 1.

DRAFT 1

[Paste draft in here]

STAND BY FOR DRAFT 2

ChatGPT will probably tell you it’s standing by; if it starts to respond, hit “stop generating” and then enter DRAFT 2 as your next prompt, pasting in the text of your second draft and finishing that prompt with STAND BY FOR DRAFT 3. Then do the same thing with draft 3, finishing with PROVIDE DRAFT NOW.

You may still need to do some editing on this fresh draft, but you’ll probably be a lot closer than if you kept working away on one of its predecessors.

 

The final edit

When you think you’re ready to hit “send” or “publish”, it’s time to treat GPT like a professional editor. Yes, there are editors who will just return a revised draft to you, but the best editors simply raise questions or point out places where you need to clarify your work. So ask GPT to do exactly that:

You are the editor of a business publication who is responsible for training and cultivating new contributors. Your goal is to teach contributors to edit and improve their own work, rather than rewriting it for them. Take the following draft blog post and return an annotated version that uses code commenting or [square brackets] to…
– Flag any outright grammar or spelling errors, indicating the recommended correction
– Insert queries about any sections where the content is confusing or needs clarifying
– Suggest potential cuts to reduce word count without losing voice or content
– Suggest three possible approaches to drafting a conclusion

I used this very approach to get GPT’s input on this draft post, and its parenthetical annotations and suggestions were far more useful than what I get by asking it to simply revise a draft. You can thank GPT for sparing you from a pretty rotten pun—and you can blame me for rejecting its suggestion to replace the phrase “puke up” with something “more professional”.

The end of the beginning

Once you embrace the idea of writing as an iterative process of gradual improvement, it not only changes how you approach each individual piece of writing; it changes how you look at the journey from one writing project to the next. Just as we aim to improve with each draft, we can aim to improve with each project — and given the rapid rate of AI advancement, we can aim to improve the way we collaborate with AIs.

 

But what about the humans?

If all this back-and-forth drafting with AI reminds you of how we used to go back-and-forth with our colleagues, well, hey! You’re right to wonder about how this kind of change in our work processes might affect the future of organizations and employment.

The recent piece in the Journal dug into this larger question with comments from a range of tech and business thinkers, including me. In How AI Will Change the Workplace, I wrote that…

As AI takes over routine tasks, there will be a temptation to cut the whole tier of entry-level employees: Summarizing documents, answering routine emails, writing basic computer code and solving simple logistical challenges are all tasks that AIs can do about as well as an inexperienced human, and at much lower cost.

But employers still need an on ramp for new hires.

You can read the rest of my comments—as well as thoughts from Sherry Turkle and many other thoughtful folks—in this article.

This post was originally featured in the Thrive at Work newsletter. Subscribe here to be the first to receive updates and insights on the new workplace.