Writing better with AI
Why can AI output feel 'meh'?
It's because it's trained on the internet. As the saying goes, 'garbage in, garbage out.'
The garbage of the internet floats to the top because companies pay for it to be there.
This is the great "enshittification" of the internet, and big tech more broadly.
“Enshittification is a term that was coined by British-Canadian writer Cory Doctorow in 2022 to describe, essentially, the process of something becoming more and more shit.”
How do you spot AI-generated content? Here are a few things I look for:
Generic openings: "In today’s fast-paced world...", "As technology continues to evolve..."
Too much enthusiasm: "Unleash your potential...", "Transform your life/work..."
Lack of examples: AI is terrible at adding real-world examples unless you force it. If everything is vague, that’s a bad sign.
Large language models (LLMs) are trained on internet content that prioritizes clicks and ads over genuine insight. So if we rely on an LLM to do our writing, the output is kinda garbage, or at best average.
If you don't push past the average sludge when you're working with AI, you'll get average results.
While AI tools are getting better, I still follow a few steps to make my writing more like me. That is, until I have an AI bot that is Jas-ified :)
Until then, let’s explore how we can find balance:
Finding the balance
Somewhere between “I’ll never use AI” and “AI writes everything” is the spot you want. It's where efficiency is balanced with your unique voice—the optimal AI integration sweet spot:
The AI sweet spot for writing. Created with Napkin AI
The goal is to use AI to sharpen your work, but also keep your voice.
The steps I use are:
Step 1: Define your voice
If you don’t know your voice, AI sure won’t.
To define your voice, write a style guide. It doesn't need to be long or complicated.
Here's an example of a voice and tone guide I use when editing my writing with AI:
Use short sentences.
Be skeptical, ask tough questions, but not too cynical.
Avoid hype and marketing buzzwords words like “unleash” or “transform.”
Add a little humour, but don't go overboard.
No emojis
Be direct
Sound human
Either add these points into the prompt when you're asking AI to edit your writing or into the tool settings such as in ChatGPT:
ChatGPT Personalization Settings: Add custom Instructions in the traits section
Step 2: Write the first draft yourself
Don’t let AI think for you.
Start messy. Struggle with the blank page. Organize your thoughts.
Build—and rebuild—your outline.
AI can’t replace this part. It’s where your real ideas come out.
And yes, it's painful. It's uncomfortable. We all sit there, and squirm, and procrastinate by doing a load of laundry or paying the bills or watching one more YouTube video.
But for best results with AI, the first draft should come from you.
Ann Handley, writes in her book "Everybody Writes":
“..embrace The Ugly First Draft. As painful and depressing as it might be to write badly—you're writing! You're getting the mess out of your head and onto the page! Then when you get back to it, you can start shaping it into something more respectable. My own first drafts are like that: littered with typos and half-written phrases, like I'm typing with mittens on."
Embrace that Ugly First Draft! It'll get better later, and you're doing a valuable exercise in the process.
Step 3: Go deeper with research and examples
To do this, add stories, examples, opinions—things only you can say. This will help you make AI work for you, and not lose your unique experience in the process.
For example, recently I was responding to a sales RFP for a Microsoft 365 solution where they had 80 technical requirements that required a couple paragraphs each to create a full response. That's a lot of writing!
I could've thrown the whole thing into Copilot, and even though it’s ‘seeded’ my past RFP work, the output is still too generic. It’s still trained on internet content after all.
Instead, I took extra time with the prompt to add my own lessons and insights from client projects. For each prompt, I added a few words of the client work and insights I wanted to use in some of the requirements. Copilot then weaved those points into the general output. It still saved me time, and also strengthened the answer.
If you don’t have examples or research top of mind—which let’s be real, is most of the time!—use AI to get the research.
These days, I'm experimenting with AI tools to get better research results. One example workflow I’m inspired by from Vicki, a YouTuber I follow is:
Start with an AI tool for research papers like Elicit. It uses AI to search and extract data from millions of academic papers.
For stats and charts, use something like Our World in Data.
Gather related videos, podcast transcripts, links that spark ideas
Save the all of this including the research reports, links and transcripts to an AI tool like Google's NotebookLM to explore all the data faster.
Google NotebookLM for gathering sources and using the AI chatbot to interact
The important part here is not the tools. The tools will likely change next month or next year.
The important part is to go deeper with research—when it matters to the piece you're writing.
Step 4: Edit and Refine
The last step is to edit the insights and points in your writing.
Once I have a draft, with my deeper research, this is the point where I find AI really saves me time without losing my voice.
Here's an example of what I ask it to do before pasting in my draft:
I'm writing a piece and I'd like your feedback.
Suggest cuts. "What seems unnecessary or tangential?"
Run a tone check: “Does this sound too formal or robotic?”
Get a second opinion: “What’s missing here?”
Then, if you're really disciplined, fix it yourself. AI is still, at the time of this writing, your intern, and not your writer.
Of course, use AI to rewrite sections or smooth out hooks between section. It gives helpful suggestions for these tasks.
Final Thoughts
Writing with AI is possible. But it’s work. You need to stay in control with your voice, your ideas.
Otherwise, you’re just publishing the same generic sludge everyone else is.
Try different tools to see what works best for you!
References and more learning:
Lecture from Cory Doctorow: Enshittification
Video by Vicki Zhao: Research Workflow with AI tools