Writing Good Content

2860 words, 17058 characters

I started writing full-time in September 2022. And then two months later, ChatGPT came out.

Naturally, I’ve felt self-conscious ever since.

After all, how can I compete with models that were trained on the entire internet? How do I stand out as a writer when AI-generated content is just an API call away?

AI models are a major breakthrough for content as a marketing channel. Companies can quickly rank for long-tail keywords through cheap, SEO-friendly articles. For instance, with just a few prompts, a trucking startup for dispatchers could generate a series of blog posts with the title “[City] to [City] Route Truck Stops.”

They say that if you can’t beat them, join them.

So I played around with different models, fed my writing into smaller models, and tried using different prompts to tune the writing style.

It didn’t work.

Each time, I’d end up trashing the work and starting over from scratch. The models were producing content — just not good content. Not the type of content that grabs you, keeps your attention, and leaves you with some compelling takeaway. The output had all the right keywords, but it wasn’t anything memorable or worth reading. [0]

So while AI content might be the future of marketing, the models aren’t yet capable of producing good content. To understand why, let’s first walk through what we mean when we talk about “content.”

What is content and why is it important?

I think of content as blog posts, website copy, documentation, twitter threads, and anything involving the written word. The specific niche that we focus on at Quill is long form content, usually 1,200 to 2,500-word articles. [1]

On the enterprise side, the rise of product-led growth (PLG) has also brought content to the forefront of marketing. Instead of hiring salespeople to do outbound and hunt for contracts, PLG companies leverage their sales team to convert and upsell existing customers. The best way to win a deal isn’t an elaborate sales process. It’s showing that 20% of the prospect’s company is already using the product.

But to gain that foothold with potential customers, PLG companies turn to content. For instance, Hubspot has a blog that’s become the default source for sales / marketing advice. Auth0 and Snyk have both leveraged content to engage millions of developers. And Cloudflare’s blog has cemented their engineering reputation.

For an individual, content is just a matter of jotting down thoughts and publishing them to the world. For companies, content can take on broader objectives: lead generation, sales enablement, and branding. [2]

Lead generation content

The goal of lead generation content is to build awareness. When companies say that they invest in content marketing, this is often what they mean.

They create blog posts, educational guides, and landing pages that target specific keywords and search queries. A trucking startup might write content around “load management,” “bills of lading,” and “factoring invoices,” while a fintech startup might choose to focus on “spend management,” “accounts payable,” and “fiat-to-crypto onramp.”

It works. Zapier has a page for every integration that they offer. Search “Webflow to Notion” and the first result is Zapier’s “Connect your Notion to Webflow integration in 2 minutes.” It even ranks above Webflow’s own pages. Today, roughly 40% of Zapier’s traffic comes from organic search results. [3]

Great SEO is probably the single most important PLG metric; it means your content is useful and authentic, and that web sentiment is positive. Great SEO “automates” the “recommendation by a trusted friend” pillar of PLG, so that Google becomes the trusted friend in your prospect’s early discovery and evaluation.

  • Jon Gelsey, Auth0 CEO (2014-2017)

Sales enablement content

Sales enablement content is meant for conversion. It includes case studies, documentation, and seminars. Not a lot of visitors reach these pages, but when they do, they’re often qualified buyers who need that last push to decide.

For example, Stripe’s documentation includes a page on product tax categories. Seeing that Stripe has a distinct tax code for Cocoa Mix (40040003) as opposed to Baking Cocoa (40040009) gives me confidence in the rest of Stripe’s product. It implies that the Stripe team has thought of everything (and is also serious about chocolate.)

Today’s enterprise sales cycle is also longer and more complex than ever before. That’s where content comes in. A product manager might see a vendor case study solving their exact pain point and develop the confidence to champion the purchase.

Content allows enterprise companies to adapt to the changing procurement process without increasing the burden on their internal sales teams.

Brand content

Brand content is about taking a stand. It includes thought leadership, deep dives, and industry reports. These pieces plant a flag in the ground about the company’s (or the founder’s) stance on a subject and generate disproportionate impact.

The front page of Hackernews from the past few days includes Clerky’s Startup Incorporation Handbook and AirGradient’s price to accuracy comparison for Air Quality Monitors. Neither article targets a specific keyword, nor do they nudge a reader to become a user or purchase their product. Yet, they’ve been successful in attracting attention and adding to the company’s “brand.” That then leads to downstream effects, like being top of mind when a young entrepreneur wants to start a company or when New York has the second-worst air quality in the world. [4]

The other side of brand content is non-revenue focused writing. Clipboard Health has a great blog post about how they ship features. I can’t imagine hospitals or professionals interested in Clipboard’s product framework (Note: There’s a good chance I’m wrong) but new hires might care. And maybe their content results in hiring an awesome new engineer.

At Quill, we mainly write brand content. Our focus is taking a company’s thoughts and turning it into company content that effectively explains who they are and what they do.

What is good content?

We’ve defined content, but how might we define good content? Well, good content has three distinct pillars: ideas, prose, and narrative.


As Isaac Asimov one said, writing is really just “thinking with your fingers.” Good writing needs ideas that are clear, original, and valuable to your reader. Those ideas are rare.

Most good ideas are some variation or extension of past ideas. For example, the iPhone is the combination of three different ideas: a mobile phone, a touch screen device, and an internet communications device. Separately, these ideas aren’t original or exciting. But when we cleverly combine them, we create something new and valuable.

As I write, I tend to doubt my ideas. There’s always that nagging voice in my head asking if my idea will actually matter to anyone. Part of being a writer is learning when to silence this voice and when to listen. It’s a constant push-pull of reworking ideas, throwing out the bad ones, and thinking of new ones ‒ in other words, writer’s block.

Unlike me, AI doesn’t get writer’s block. In fact, AI can come up with good ideas, really good ones. Trained on the internet, AI models have access to more ideas than anyone else. With good enough prompting (which is key), AI models can bring together distinct ideas in new, interesting ways. [5]

So, a company can call on an AI multiple times and extract the unique ideas. For instance, if I wanted to write an article about “how to sell a pen,” I could ask “how do I sell a pen quickly/on the internet/for $1,000” to find the right ideas.

But as we’ll see, good ideas are just one piece of the puzzle when it comes to writing good content.


Prose refers to the type of language you use in writing. Which words do you choose to convey ideas? You might have learned about it in English class as “diction,” but it means the same thing. Prose underscores a key idea: there’s the grammatically correct way of saying things, and then there’s the stylistic method.

As an example, let’s look at the title of this article, “Writing Good Content.” In theory, we could rewrite the title to “Good Content Writing” or “Content Writing Done Well” or “Well-Written Content.” With small differences, they all have the same end meaning. Yet, “Writing Good Content” rolls off the tongue in a way that the other variations don’t. It just sounds better.

Prose is what makes ideas stick. Part of why I got into writing is this quote: “Man decays, his corpse is dust. All his kin have perished; But a book makes him remembered through the mouth of its reciter.”

We could simplify the quote to “Writing lasts a long time.” By traditional metrics of concision and active voice, our shortened version is better writing. But it lacks the power and longevity of a phrase like “man decays, his corpse is dust.” That’s an image that will stick in your brain ‒ for better or worse.

Good content requires engaging prose to get its ideas across and keep the reader thinking.

Again, this is something that AI models can do. With a bit of fiddling, such as “Please rewrite to the style of [insert famous author]” or “Rephrase to be more bro-y,” AI models can produce good prose. It still needs clever prompting and a pair of human eyes to read over and verify the prose, but it’s much easier to judge and edit good prose than to write it from scratch.

So we’ve established that AI can generate ideas and write them out in engaging prose. But there’s one more key ingredient that’s needed for good content.


For the most part, AI is incredibly useful in creating lead generation content. Good lead gen content can get by with just ideas and prose.

But useful lead gen content is only a small portion of the overall content picture. AI content doesn’t extend well to sales enablement content or brand content. Both of which require narrative to function effectively.

I spend around 30-40% of my total writing time on the first few paragraphs. I write and delete and rewrite and edit and tear out my hair. I brainstorm that perfect opening that sets the tone for the rest of the piece. When I finally find the correct words, the rest of the piece just flows out.

Those first few paragraphs are the start of a piece’s narrative. When people say that a book could have been a blog post or that a blog post could have been a tweet, they’re saying that the content lacked a narrative. The writing might have had good ideas and great prose, but it lacked that beginning-middle-end narrative thrust that keeps readers’ interest.

A good piece of content has a bunch of smaller ideas that are packaged into a single essay. Crafting a narrative enables the writing to flow from one idea to the next and pound out a path made of ideas. Great narratives then connect the ideas so that each idea builds on the next.

Narratives are crucial in both sales enablement content and brand content. Good sales enablement content tells a story. You must lay out the customer’s problem, explain why the problem matters, show how the vendor’s product solved this problem, then emphasize what that solution meant for the customer. It sounds formulaic, but the difference between a great case study and a bad case study can make or break a deal.

Brand content is a bit more self-explanatory. If readers can’t stick around to the end, then the content failed at its job. It’s no different from a movie that you turn off after 30 minutes. The writer and director weren’t able to lay out the character’s problems in a compelling way. So of course you aren’t going to stick around for the solution to this character’s boring conflict.

Unfortunately, AI content doesn’t have a narrative. The default response for most AI models when asked an open-ended question is to write bullet points. Each of the bullet points is a distinct idea (that might be good), but they lack the connective tissue to piece them together.

More importantly, I’m not sure if AI models will be able to improve. If we look at OpenAI’s ChatGPT and GPT-4 AP scores, the AP English Language and AP English Literature scores didn’t improve with GPT-4.

While AP English exams aren’t exactly the best proxy for good writing, it shows that most AI models are struggling to figure out how to understand and create narrative-driven writing.

According to the College Board, the express goals of AP English Lang are “understanding of how written language functions rhetorically: to communicate writers’ intentions and elicit readers’ responses.” And AP English Lit aims to “examine the choices literary writers make and the techniques they utilize to achieve purposes and generate meanings.”

That’s pretty much the definition of narrative.

Content, AI, and the future

The PLG approach to growing a company tends to emphasize prioritizing investments in product and software. It’s a build-once-benefit-often model, where a company’s initial investment pays off again and again over time.

Content works in a similar way. A good piece of evergreen content can draw readers’ attention and respect for long periods of time. This article from Uber’s engineering blog about GPS was written in 2018 and reached the front page of HackerNews in 2022. For startups building for a technical audience, good writing is often the only repeatable method for reaching technical folks.

Until now, companies have had a tendency to fall on the extremes of the content spectrum, either investing heavily into a full content calendar or just writing one to two blog posts every year.

AI is changing that. Companies are incentivized to use GPT-4 and other models to rank for long-tail keywords. With just a few hours of work and a couple of dollars, companies can build an entire content library that continually draws in organic inbound leads. All things equal, it doesn’t make sense to not use AI content.

That leads to two predictions. First, the amount of enterprise content is about to explode. A subject that might have been addressed by just one or two relevant blog posts will now have dozens to hundreds of different posts addressing those ideas, often in similar ways. Companies will compete to adopt the newest tools and have increasingly sophisticated SEO and keyword targeting. Amid this arms race, companies will likely chip away the efficiency gains from AI.

The second trend is that companies will bifurcate into those that optimize on the quality of content and those that focus on the quantity of content. Having a baseline set of lead generation content becomes trivial with AI. When people are inundated with more formulaic content, content with a strong narrative will cut through the noise. Content that goes viral will be a function of narrative and not a function of idea or prose.

At Quill, our mission is to write good content. After my experiments with AI, I’ve shelved the models. They aren’t yet able to create engaging, narrative-driven content ‒  and may never be. So instead, we write everything from scratch. It’s slow. It’s painful. But if you need good content, it’s still the only way to create it.

Shameless plug, if you liked my writing, I run a content agency. We work with startups and firms that have a unique perspective on the world. Learn more about us at our website or email me at richard@hirequill.com .

Special thanks to RC, N.M., M.S., and A.L. for reading over early drafts of this piece.


[0] Caveat here. If we get AGI, then this whole blog post is moot. But, we’d also have bigger problems than AI models being able to write good content.

[1] Content is more broad, it includes stuff like long and short form videos, webinars/conferences, images, and minor tools (like Shopify’s logo maker). Its official definition is - “information made available by a website or other electronic medium.”

[2] There are other frameworks for content. Such as Lenny’s framework by source and outcome -

or Buffer’s framework by inputs -

[3] For more reading on how Zapier built such a strong content engine, Ryan Berg wrote a great article on how they tactically did it and Tanay Jaipuria has a broader roundup on programmatic SEO.

[4] Hackernews isn’t exactly the most representative source for viral content. But for developer-focused companies, it’s the best marketing channel (when successful). You could also run the same process on Twitter, LinkedIn, or any other platform.

[5] I normally don’t use AI in my writing. However, when I asked Claude+ about how it thinks AI models will generate new ideas, the response was: “For example, an AI model can analyze romance novels to learn the style, themes and tropes of that genre. It can then generate a science fiction story using the same style - this style transfer can yield new ideas as the AI combines genres and styles in unique ways.” I’d love to read a sci-fi novel written in the style of a romance novel. Maybe that’s a space opera?

· Writing