How Marketers Are Using AI for Writing [Survey]

There’s a lot of debate about how AI can be used in writing. Is it the future, or is it ruining the internet with bad content? While the debate continues, we decided to gather data.

We surveyed over 50 marketers, from freelance writers to C-suite executives, on how they’re using AI to produce marketing content today.

We received extremely detailed responses about the processes they’re using, the strengths and weaknesses of AI writing, the tools they rely on, and more.

Below is an analysis of those responses.

You can also download a spreadsheet of every single response by signing up for our newsletter. If you’re already subscribed, just check your email.

Survey demographics: Who we asked & selection bias

We sent this survey to our email list and shared it on social media. We estimate that it reached an audience of over 5,000 marketers. We received 53 responses.

That’s a roughly 1% response rate from “saw it” to “filled it in”.

Why does this matter? Because it’s extremely likely that the 1% who responded are more enthusiastic about AI than those that didn’t. This is known as survey selection bias.

As a result, our data likely over-represents marketers who like AI, and under-represents marketers who either haven’t tried it or maybe have tried it and didn’t like it.

Of the respondents, 84% identified as being in marketing roles.

The roles were fairly evenly distributed across freelance writers, employees, management, and executives.

A majority of the respondents were in the U.S. or Europe:

Most writers and marketers are using ChatGPT

We asked respondents about the AI tools they use in an open-ended question, allowing them to list multiple tools, list obscure AI tools, or self-built models.

We manually categorized each response into one of the tools below:

The results are overwhelmingly dominated by ChatGPT — 90% of the responses mentioned it. So the data below reflect marketers in various roles, mostly in the U.S. and Europe, primarily using ChatGPT, followed by the other two members of the current AI “big three,” Claude and Gemini.

83% of marketers have published AI-assisted content

The first and most fundamental part of this survey was to understand how much marketers are using AI assistance in writing.

We asked: Have you published AI-assisted content at all, ever?

Benji and I were shocked that this number was so high. We can’t say what number we expected, but it definitely wasn’t 83%.

This means that, in 2024, when large language models and AI tools are still in their early stages, the vast majority of marketers are already using them — not just for fun, but for actual published content.

This is significant. It means AI writing is happening. It’s here — whether we like it or not. And if it’s 83% today, what will it be next year or in 3 years?

But let’s be careful. This stat doesn’t mean a few things:

1. It doesn’t mean 83% of words in marketing copy are AI-generated. As we’ll discuss below, marketers are using AI in a variety of ways, from seeding ideas to solving tricky wording issues. Most are not just asking ChatGPT to write entire blog posts.

2. It doesn’t mean this writing is good. Below, we discuss how the biggest complaint — even of these 83% ‘yes’ respondents — is the hours of editing required to turn it into something publishable.

Most have published many AI-assisted pieces, not just one

We asked: How many pieces have you published that were assisted or written by AI?

This question aimed to verify whether we’re talking about marketers using AI assistance regularly or just dabbling in it once or twice.

Over half of the respondents had published more than 5 pieces with AI assistance.

This shows that marketers aren’t just dabbling; they’re consistently using AI assistance in their writing.

Those who aren’t using AI assistance cite poor writing quality

Next, we asked the 9 respondents who said they aren’t using AI: why not? The majority pointed to concerns about the quality of AI writing.

Here are some examples:

• “The writing style is that of an eager grad student. And oh so bland.”

• “AI produces content that sounds ok but lacks the depth needed to really address pain points and needs in an engaging way, so that you really feel heard and understood.”

• “Mostly, it’s pretty but meaningless marketing blah blah.”

• “Also, I find it unreliable. For instance, in one case the AI invented an interview partner for a customer case study who doesn’t exist, even though it was fed with the transcript of a real customer interview.”

These concerns get at the heart of the matter with AI writing:

1. It’s not great.
2. It can outright hallucinate (a well-documented issue).

Finally, one particularly notable response cited pride in their craft, which we resonate with. Using AI to help write feels like it defeats the whole joy of writing:

“I’ve been a professional writer for 16 years, so I have a process. I also enjoy the challenge of building a piece of content from scratch. Although I’m sure AI would make my process faster, where’s the fun in that?”

Regardless, this perspective only accounted for 17% of respondents. For the 83% who are publishing AI-assisted content, let’s see how they’re using it.

How marketers are using AI for writing: ideas, outlines & drafts

This is where we dive into the specifics of how marketers and writers are using AI assistants for content creation.

First, we asked: How exactly did you use AI to help write? What was your process like?

Many responses were incredibly detailed. One writer even shared a six-step process:

1. Interview with client — AI (otter) makes a transcript.

2. Claude fixes the transcript so it’s easier for me to read/scan.

3. I load my notes/strategy docs/instructions/tone of voice to a Claude project (or use one that’s preloaded)

4. I consider the angle I want, plan strategy and messaging, etc. Then I tell Claude to write me an outline based on those directions. I make edits.

5. I might also have Claude use the interview to answer a series of questions.

6. Depending on the project, I’ll have Claude write a first draft. I make edits.

We went through the responses manually and bucketed them into themes:

• Draft long-form content
• Outline/ideas/keywords
• Editing
• Writing headlines or meta info
• Summarizing text or interviews
• Writing social posts
• Marketing site copy
• Other business use cases

We then tallied how many of the nearly 50 responses addressed each theme.

Note that a single response often touches on multiple themes. For example, if someone says they use AI to outline and draft, that would add one count in both the “outline” and “draft” categories.

Drafting was expected to be popular

Drafting long-form content is exactly what we all expect to be the most common use case for generative AI writing assistance. Drafting a full blog post is time-consuming, as we know first-hand! If AI can speed that up, even a little, it’s worth it.

As it turns out, for these respondents, it’s worth it even if AI isn’t a great writer, even if it produces a lot of fluff, and even if you have to spend time editing (we’ll get into this later).

Which begs the question: If LLMs are useful to help write now, despite producing fluffy writing that needs heavy editing, how useful — and therefore ubiquitous — are they likely to be next year? Or in 3 years?

Plus — and we’re going to make a bit of a judgement here — most companies produce mediocre to bad long-form content anyway…with humans. We’ve been writing about this problem for years, long before ChatGPT came out (e.g. Mirage Content, first published in 2017). So, if a writer or marketer’s clients or bosses are expecting “mid” content anyway, what’s the harm in using AI to help produce this?

Editing and outlining was a surprise

What we didn’t expect was how many marketers use AI to help them with outlining and editing. We assumed they were feeding outlines and arguments they created, then editing AI’s drafts.

We thought the process must be:

Because if a human is going to produce an article, they must know about the topic, right?

Turns out, no.

Based on these survey responses, the average use of AI in the writing process looks more like this:

Each step could involve humans, AI, or a mix of both. Marketers are using AI in various ways at different stages of the process — some for research, others for outlining, summarizing human interviews, drafting, or editing a human’s draft.

Here’s a symbolic example of how extensively AI is used in the writing process:

“I use several AI tools to help generate ideas and outlines. I guide the tools to create something very close to what I am trying to convey. I use other AI tools to confirm and/or add more.”

We have to admit, this isn’t very “G&C”. We’re all about interviewing subject matter experts who know the topic better than your average reader, so this approach makes us uncomfortable.

In our view, if you’re asking ChatGPT to outline your arguments, where’s the originality in your article? The same goes for editing. How can the LLM edit without understanding the key arguments or your brand’s position on the topic? We haven’t seen any responses suggesting that folks are training custom GPTs on their brand material to write or edit. Most are using out-of-the-box ChatGPT.

But, as mentioned earlier, most clients and brands don’t prioritize originality.

If you’re producing top-of-funnel Google Research Papers anyway, ChatGPT can help with both the initial research (replacing your Google research step) and generating an outline, often in one go. It’s a very convenient process. We’ll demonstrate this with an example in the next section.

In fact, this response from a separate question of the survey perfectly summarizes the point:

But for the clients that were most demanding in terms of quality, I found this doesn’t work as well. My best writing comes from not using AI at all.

But most clients don’t want that high of quality. Most clients want something that is generic and familiar with a little extra on top. Most agency clients don’t know (or care that much) what kind of content actually produces results for their clients.

Overall, we found the detailed responses to this question to be a fascinating look into the workflows of different marketers using AI for content generation. Again, you can sign up for our newsletter here to get the full list of responses.

Why did marketers start using AI writing? Speed and research

To better understand respondents’ pain points, we asked why they started using AI in the first place.

Was it to solve a real problem, or just FOMO from the ChatGPT hype?

The following seven themes emerged from their responses:

• Speed/saving time
• Research
• Improving writing
• Ideation or outlining help
• Cost savings
• Staying cutting edge
• Overcoming writer’s block

It wasn’t surprising that speed and saving time were the main reasons people first tried writing with LLMs. Along with 14% citing cost savings, 68% of responses highlighted the obvious benefits of using generative AI for writing.

However, what we didn’t expect were:

• Using AI to improve writing
• Ideation and outlining
• Overcoming writer’s block

Using AI to improve your writing

We were surprised by how many people said they first started using AI to improve their writing — this was the second most common response. We’d assumed AI writing was seen as inferior to human writing, but using AI to improve writing? That’s a surprise.

To be fair, some referred to “editing” as small tweaks, like trimming sections or reworking sentences:

• “I needed help trimming some of my content so as not to be cut off in Google search results. I’ve also used it for research.”

• “…to help rewrite a sentence in an article…”

But others cited larger reasons, like this marketer struggling with product copy and using ChatGPT for ideas:

“I was writing about a product line from a company whose website was full of marketing guff; I had attempted to rewrite it from scratch several times, but it just wasn’t clicking so I finally plugged it into ChatGPT to see if it could give me a better angle.”

While we don’t know if ChatGPT’s output was useful or how much editing was required, the fact that a marketer turned to AI after several failed attempts to write what they were looking for shows how much marketers are already valuing LLMs for writing assistance — even if most aren’t using the copy verbatim.

Research, ideas, and overcoming writer’s block

The combination of AI being used for research (22%), ideation and outlining (22%), and overcoming writer’s block (8%) stood out as notable themes.

Remember, this question is about why they started using AI in the first place. Adding up those three percentages (though not entirely fair, as several mentioned multiple themes), more than half of marketers in this survey said their gateway to AI writing was getting help with the foundational legwork behind a piece.

This suggests that AI is helping produce the “Google Research Paper” we referenced earlier, but on steroids. Before LLMs, writers would use Google to research a topic, then build their arguments.

What’s compelling about using LLMs is that they assist with both research and outlining or writing — all within the same tool.

For example, say we are tasked with writing a piece on trimming trees. Sure, you can Google it and read a bunch of these articles:

You can also watch a bunch of videos like this:

But that takes time.

You have to consolidate the information yourself (take notes, organize bullet points, cite sources) and then start formulating a structure or outline for your article on tree trimming.

This is hard work, but it’s what writing is all about: organizing information, deciding what’s worth keeping, how it flows, and playing with the structure. Authors like John McPhee and William Zinser have written extensively on this.

The words are the end result, but it’s the arguments and their flow that matter most. And yes, that’s tough. It’s like solving a challenging puzzle — it taxes your brain. But that’s the point. It’s the mental equivalent of lifting weights. The whole point is that they’re heavy.

While this mental weightlifting is the hard-won reward of writing, it’s also the allure of shortcutting this process with LLMs. ChatGPT can handle much of this hard work on its own. It may not reach the level of a great writer, but the appeal of having a computer tackle these mentally taxing tasks is too strong for many — especially if your client or boss can’t tell the difference.

With AI, you provide a prompt, a topic, and other details, and it’s trained on data related to that topic (like tree trimming). It’s as if you’ve read 10,000 articles on tree trimming. It handles both the research and outlining in one go, freeing you from juggling all the ideas in your head.

Let’s continue with the tree trimming example in ChatGPT:

We asked it to draft the post itself, not just the outline, but you can see how using LLMs first for research, then for outlining or brainstorming ideas, can easily lead to eventually asking it to draft the post.

And off it went, drafting this article in just a few seconds:

Not to mention, it sold our fictitious tree trimming company so well:

We just weren’t aware we were offering a 10% discount. Read More