A transparent account of my perception of generative AI and the implications of its use by organisations to create long-form copy such as articles, guides, white papers etc.
Earlier this year, I read a book in which a couple of crows learned to talk. On the surface, it’s a story about communication. But dig a little deeper (or, you know, read to the end) and you’ll find that it’s also a story about sentience. The ability to think and how we measure it.
You see, these crows could talk. But the only things they said were things they had heard elsewhere. In this sense, they were crows but they were also parrots. Or were they?
A character asks a crow a question, and they repeat the question back to her with an inflection. Or they reply with a quote from Shakespeare. Their reply works, in the context of the interaction (or at least, the other participant infers meaning from it), but it’s never quite clear if the meaning is that which the crows intended, if they intended meaning at all.
After all, they’re just crows.
Fast-forward to a few months ago, and the business approached me to get my opinion on generative artificial intelligence (AI).
- How much time could we save using AI copywriting tools?
- How should we position our messaging around them?
- How can we help our clients to use them best?
I wasn’t able to give an on-the-spot answer. Because, although I’d been watching the emergence of, and the reception to, these tools closely, I hadn’t made up my own mind yet on what I thought of them, much less considered our company line on the subject.
This wasn’t anything to do with the topic in question. I’m usually the person in the room who mulls things over. A listener. One of the last to speak. You’d think me quiet (perhaps even disinterested) but I like to know all the facts so that my decisions are informed ones.
It was also everything to do with the topic in question.
Eric, one of our co-founders, first came to me to chat about an AI writing tool in 2022. Eric loves a new gadget but I wouldn’t hear him because I wasn’t ready to consider what it meant for my future, one in which robots could, at the click of a prompt, mass-produce prose with creativity, flair, and a deftness of hand that would make Shakespeare weep. (Alas, poor Thomas!)
So I’d been all too willing to bury my head in the sand, if only to block out the seeming death knells of my chosen career path. You might imagine the Terminator kicking down my door and throwing me off my chair, its shadow stretching over me and the flaming remains of my laptop, a deceptively human voice ringing out: “I’ll be back”.
And it was. When asked again about copywriting AI a year later, however, my circumstances had changed. HubSpot had launched its own tool, embedded across its platform, and very soon all our clients would have access to it. Using it, they’d be able to generate copy for blog articles, social media posts, marketing emails, and meta descriptions, among other things. Suddenly I couldn’t afford to stick my head in the sand anymore, because the tide was coming in fast and, to the best of my knowledge, ostriches aren’t sea birds.
So I ran tests. This is how I like to approach anything new. And I read. Because the best way to understand something is to get a well-rounded view of it, and the best way to get a well-rounded view of anything is to read about it. And the more I tested, and the more I read, the more I wondered if the AI and the crows from the story had something in common.
What follows is how I ran those tests, my findings, questions that I put to you because they feel important and, as it turns out, not many other people seem to be asking them. And how a book about talking crows helped me to look at this new wave of AI tools in a different way.
- Glossary of terms
- Putting generative AI to the test
- The brief
- AI in action
- Reviewing the results
- The question not being asked
- What’s left when you take the writing out of copywriting?
- “Why do we write?”
- To write or not to write; that is the question
Generative AI — generative AI is a field of artificial intelligence that focuses on creating new and original content. Unlike traditional AI systems that are limited to performing tasks based on pre-existing data, generative AI has the ability to generate entirely new and unique content, such as images, music, and even text.
Prompt — prompts are fed to the AI so that it understands what to create. A prompt can be thought of as a brief and usually takes the form of a sentence with keywords. The more detail provided in a prompt, the better the AI will be able to process your request.
HubSpot content assistant — HubSpot content assistant is an AI tool offering a wide array of features designed to assist in the creation and optimisation of content, such as generating topic ideas, providing real-time SEO recommendations, and creating copy.
Existential dread — a profound and unsettling feeling that arises when an individual contemplates the meaning and purpose of their existence, such as that which might be experienced by a writer faced with a machine capable of doing everything they can, better.
Putting generative AI to the test
As an account of my process goes, what follows is mostly anecdotal, because there are much more interesting things I want to discuss here than, say, exactly how much time I might have saved, down to the minute, using the tool. You’ll get the gist of it, which is what matters.
My experiment was straightforward: use HubSpot’s new content assistant tool to help draft a series of blog articles over the course of three months. At the time, HubSpot content assistant offered a range of commands: idea generation, paragraph writing, article outlines, and an option for summarising content into a conclusion. I would test them all.
I wrote one article a week, and I chose an industry to which I regularly contributed. (We have a client in that space.) This gave me a solid baseline of understanding, which I could use to assess the copy being generated for me. When reviewing the copy created, I looked for:
- The overall quality of the writing. (Quality is itself a difficult thing to quantify, and a subject for another day, but I’m confident in my ability to hear it in a piece of writing.)
- The accuracy of the content. This includes any figures cited but also extends to any claims/statements made by the AI. My industry knowledge would help here.
- How much time, if any, I saved. Efficiency being the word of the day in discussions relating to the benefits of AI-generated content, I wanted to review this myself. This would be easy to quantify by keeping a record of how much time I spent.
AI in action
Before we dive into the results, let’s look at how useful I found each of the commands.
Command: “Generate paragraph”
Everyone faces blocks when they write. At times, I struggled to wrap my head around a subject. Other times, my creativity was simply flagging. This command gave me a steer when I needed it, not because it offered me copy that I could, well, copy (it didn’t; see the results, below), but because it revealed to me the general consensus on any given subject, at least in terms of how the majority of people were discussing it online. (At times, this was helpful for instructing me on what not to write about to differentiate my copy from the crowd.)
As well as a general steer, the command could be used to provide a jump-start into an article. Again, I always rewrote this, sometimes entirely, but in terms of getting the gears turning at the beginning of a new writing task, it was occasionally helpful.
Command: “Generate an outline”
I found little use for this tool, which scans all the available content on a subject to recommend a comprehensive outline for you to follow. Its scope was far too broad for the kinds of content I was creating, which needed to be very specific to a challenge faced by a fictional target reader. It would have been more helpful when writing a pillar page but only at the planning stage.
By the time I’ve sat down with my thoughts and started typing, my outlines are already planned (and, usually, tailored to specific keywords). Perhaps this tool would have been more useful if I hadn’t known my industry or my audience so well, but I can’t speak to that.
Command: “Generate ideas”
I got the most value out of this by using it to help me generate ideas for titles.
With so much riding on your choice of title, and so few words to play with, it’s very easy to agonise over titles for far longer than you probably should in search of The Right One.
The ability to request a selection of varied titles based on my keywords and other prompts gave me lots of ideas to play around with. Often, the suggestions were phrased in ways I wouldn’t have considered for myself. And by using my article in the prompts, I was able to make sure the suggestions spoke to the key points. I found this command really helpful.
Command: “Generate a conclusion”
I’m often asked how to write an effective conclusion. Do you just repeat your key points? Address the reader? Round everything up by using the space to push your product or brand? There’s no universal answer, and a lot of the time it’s going to depend on what you’ve written (and why). This command didn’t offer much in the way of a satisfying solution.
It appears to work by reading the rest of your article and reiterating the key talking points. There’s no added value. It implies that the reader hasn’t already read or understood the rest of your article (or worse, forgotten it from five seconds ago). But most of all, it offers a summary, not a conclusion, and for all that these words are often used interchangeably, there’s a real difference between summarising what has gone before and concluding it.
A conclusion should evoke emotions in the reader. It should play off the tension you’ve built up, or the anticipation, throughout the rest of your article. It should nod back to your introduction, where the story began. It should make the reader sit up, not switch off. My experiences of this command to date are that it can’t, as yet, live up to its promise.
HubSpot has since added a “Generate image” option. This wasn’t available when I tested.
Reviewing the results
I didn’t use all of the commands for every piece of content. Instead, I applied them in the way that I would if I were using them in a real-life work environment: not to depend entirely on AI, without having written a word myself, but to use it selectively. To fill gaps. To ride bumps. To break down blocks; in other words, to complement my tried-and-tested approach to writing.
1) I expected that, in so much as quality can be determined in something as subjective as a piece of writing, the copy generated would be poor.
When asked about a subject, the statements provided were generally top-level. Different questions often generated similar-sounding answers lacking specificity or any meaningful insight above and beyond the obvious. It was a challenge to prompt the tool to provide detail to the level expected by me and required by the business if we were to continue to provide authoritative content that spoke specifically to a question or keyword.
It quickly became apparent that the way in which the tool gathered the information it needed to generate copy put it at odds with my belief that content should provide new value above and beyond what other articles or businesses are saying on the subject in order to stand out from its competition. I foresee businesses depending on this tool to deliver quality content struggling to rise above the sea of content from which their own copy has been generated.
2) I expected the accuracy to be high.
Inaccuracies were few, but liberal use of adjectives throughout the copy meant that there was often more sentiment to it than I was comfortable with. To read a certain field described as “fascinating”, for example, left me wondering how the AI had deemed it such and how it had determined that I would agree when nothing in my prompt suggested that.
The truth, of course, is that the AI doesn’t find one field or another fascinating. The tool’s dependence on source material from which to generate its output means that human biases such as this still found their way into the copy. For all that the copy generated was original, these biases ran like racial memories through its DNA. Not only was I having to spend time fact-checking every statement made, but I also had to remove or rewrite this language, too.
3) I expected to save time by using the tool.
For the first of the three months, I was able to save significant time by using the tool. Specifically, I completed the tasks in around half the time it would have typically taken me. I attributed this in part to the “Generate a paragraph” command, which helped me to streamline the writing process and maintain my focus by providing me with what were effectively prompts when writer’s block would otherwise have interrupted my flow, and in part to the “Generate an idea” command when it came to refining my proposed titles.
The experiment largely confirmed what other people were saying about the tool and AI copywriting tools in general. I was more interested in what other people weren’t saying.
The question not being asked
Reading into the subject, I heard lots of arguments, counter-arguments, and sweeping statements around the use of the sorts of tools we’re discussing here.
Some were rational, considered, and based on tests of the kind that I had run.
- “Look how much more efficient we can be if we use these tools correctly.”
- “Look how much money we can save by using AI to generate our content.”
- “Look at how much more value we can get out of our writers with these tools in their arsenal.”
Some were optimistic, or apologetic.
- “Inaccuracy isn’t an issue — our in-house team fact-checks everything it generates.”
- “The quality isn’t there but we re-write most of it anyway.”
- “It doesn’t matter that there’s no narrative — we add that in afterwords.”
Others were more impassioned.
- “The first rule of using AI for content? Don’t tell anyone you’re using AI for content.”
- “Replacing your copywriter with one of these tools is like replacing your accountant with a calculator.”
- “If you need an AI tool to help you generate ideas, you shouldn’t be in the room.”
- “The use of AI writing tools will make you unhirable.”
Mostly, the conversation kept coming back to whether or not these tools could replace actual copywriters. I wanted to look at the subject from a different angle, and the implications that stem from looking at it in this way: not could they replace copywriters, but should they?
What’s left when you take the writing out of copywriting?
Copywriting without the writing is just copy.
You can see this in the way that generative AI crawls existing content on a subject in order to generate its results. It’s coming up with something original, in so much as it isn’t directly plagiarising what’s been written elsewhere, but it’s not capable of thought. The ideas it generates, the copy it produces, are all determined by information it’s gathered elsewhere. It doesn’t have an opinion and it isn’t able to examine a topic critically. It just copies.
And what of a copywriter, when you take away the writer? Because that’s what these tools are doing: removing the need to write, because writing takes time and it can be expensive. AI writing tools solve both those problems, right? All you need to do is edit what’s there.
Yes, writing can take time, and yes, it can be expensive. But how do you put a price on having someone in your business (or close to it) who understands your industry? Someone whose passion for your company spills into every word? Someone who speaks to the value in what you deliver and how your services change your customers’ lives, day after day? Someone who holds all these things inside them and, through a few well-placed, deliberate keystrokes, can express them to the rest of the world, too? No, not just express them; shout them from the rooftops: this is us. This is what we care about. And this is how we can help.
And suddenly you realise that by asking yourself if you should or shouldn’t use these tools, what you’re really asking yourself is a different question, one as old as the hills:
“Why do we write?”
I’ve long believed that we write to make sense of the world, or if not to make sense of it, to better navigate it, one story at a time. This isn’t just some flowery way of looking at fiction.
- It’s all there in the best examples of B2B copywriting: stories that place a company in a context that matters. Stories that bring a brand to life. Stories that create a sense of community and shared values. Stories that genuinely mean something.
- It’s there in the emails we write.
- It’s there in the stuff we post on Instagram and LinkedIn.
- It’s there in our customer complaints and our shopping lists
- and the Post-It note you leave by the sink because you didn’t have time to do the dishes but you will and you’re sorry.
So there’s another implication of using these tools that we must consider, and that is their impact on a generation growing up relying on them. Specifically, what happens when everybody uses these tools to write for them and nobody has to write anymore?
If writing helps us to make sense of the world, what happens to the people who, in their formative years, don’t have to make sense of it? If making sense of the world helps us to figure out our place in it, what of the generation growing up without this perspective?
Writing teaches critical thinking. (“What do I think about this subject?”)
“In today's information age, obtaining facts is hardly a challenge,” writes one research paper published by Cambridge University Press. “[...] Through online databases, books, articles, newspapers, and more recently through websites, blogs, and social networking interfaces, students have access to unprecedented amounts of information without ever leaving their study rooms. What remains a challenge, however, is the development of the skills that are needed to critique and process this easy-to-obtain information.” AI tools exacerbate this challenge by gathering all this information for you and presenting it in a format that, according to the vast datasets they’ve crawled, has a high probability of making sense.
There’s no need for the writer to find the information themselves, much less consider its relevance to their narrative/blog post/essay or make a critical argument for/against it.
It promotes empathy. (“How will that person feel or respond if I word my email this way?”)
“As a writer, you have the power to inspire, provoke and challenge your readers,” writes BBC Maestro. This is as true of fiction writing as it is of engaging content; copy that draws out an emotional response in the reader. It’s why storytelling is so effective in content; stories are transformative. We read or hear them and they move us. They shake us. They change us.
But “[t]o understand stories, we have to understand characters, their motivations, interactions, reactions, and goals,” explains Dr Raymond Mar, professor of psychology at York University in Toronto, in the same article. “It’s possible that while understanding stories, we can improve our ability to understand real people in the real world at the same time.”
So empathy is hugely important to us as professionals working in collaborative teams, as businesses trying to create and promote healthy workplace cultures, and as writers, too. Not writing deprives us of a powerful way to nurture and develop our sense of empathy.
It helps us to develop a sense of self. (“This is me. This is my voice.”)
Whenever we write, we hold up a mirror to ourselves. At 100 words, it might be the quickest flash. At 100,000, a long, deep, stare. The point is that through the act of writing, we come to know ourselves a little better. Not because we’re writing about ourselves; most of the time we’re not, but because we’re writing through ourselves. The words you choose. Your sentence lengths. The bullet points, apostrophes, that wall of text you couldn’t bear to split, and how much you care or don’t care about whatever it is you’re writing. How you sound. It all gives you away. It all helps you to know yourself; self-awareness on another level.
We’re not talking about copywriters anymore and how much time AI can save them. We’re talking about people, particularly young people, and the repercussions that come with not having to think. If we must bring it back to business, these are your company’s people.
Every email generated for them is copy they haven’t had to consider. Every social post, messaging without identity. Every webpage, drafted without thought for who your company is, the value it delivers, how it contributes to your industry or betters the customer’s life.
“There’s an old adage that good writing is good thinking. And the truth is, ChatGPT doesn’t think. It curates.” ‘Should you replace your copywriters with ChatGPT?’, IMA
But writing is hard.
“For years, tech companies have promised us magic wands that solve all sorts of marketing problems,” writes the team at IMA. “Analytics, automation, scheduling, etc. But now, a technology is being offered as a solution to a creative “problem” — namely, the problem of copywriting.”
Yes, writing is hard. You’re taking invisible, intangible thoughts from inside your head (honestly, have you ever thought about how impossible those are?) and you’re turning them into words on a page. Words that mean different things to you than they mean to the person reading them. And you’re hoping that the person reading your words somehow understands them — not just what they mean, but what you meant when you wrote them — the spirit of the thing — and then they’re supposed to write their thoughts back to you, and so on and so forth. And we all do this all day, every day. One word after another. Sentence after sentence. It’s like the whole world, the human experience, is one long game of Chinese whispers.
But this isn’t a “problem” to be solved. This is why we write: to understand and to be understood. You and I might never have met, but I reckon you’re getting an impression of me by virtue of what I’ve written about here. My hopes. My fears. The metaphors I’ve used, the stories I’ve told. My voice. What’s the point of writing when you take that meaning away?
What is copywriting without the writing? It’s just copy.
- Established copywriters can learn how to leverage these tools to their advantage. They might be able to save a little time using them. Cut a few corners, so to speak. In a way, the “Generate ideas” command I found so helpful replicates a time-old creative writing task in which you randomly place words together (an apt description of AI writing tools in general) in order to look at those words differently.
- You shouldn’t be using these tools to write entire articles for you. This includes using the tool to write an entire article for you, then passing it across to a copywriter to edit. As we’ve well covered, these tools can’t write. I know of one business where a department head began using ChatGPT to draft articles for her, which she then wanted a copywriter to tidy up. I saw the copy she wanted editing and it was plain white bread, devoid of anything relating to her brand’s personality, the department head’s expertise (she hadn’t written it, after all), or content that distinguished it from the thousands of other articles being published by their competitors using the same prompt in ChatGPT to deliver a variation of the very same article draft for them.
- The bulk of the content these tools create is generic, cookie-cutter copy that in no way reflects on a company’s expertise (let’s assume every company is at least trying to be an expert in their field) and could well detract from it if your prospects/customers find out that you (the expert they are trusting with their business) need AI to write an article on the subject in question.
- Businesses whose blog content can be described as simple articles without opinion or commentary could find some use with these tools. Of course, this approach does nothing to showcase your values, story, or expertise (or hone it, to call back to the critical thinking discussion) — all the things that give a piece of writing your brand’s fingerprints. If this kind of content sounds familiar, it’s not an AI writing tool that will give you value but a content strategist. I’d argue that even step-by-step guides, which on the surface could be produced quickly by these tools, are better written by a human with experience delivering the process and real insights into the pitfalls to avoid, the areas to focus on, the quick wins, and so forth.
- Skill and practice is required to unlock the full value of the tool, from learning how to write effective prompts to all the post-work needed: fact-checking, editing for tone, editing for regional spelling, editing to align it with your company’s brand and values, and editing to insert even the murmur of a story (a soul) where appropriate.
To write or not to write; that is the question
I asked HubSpot’s content assistant tool to describe itself to me. Nestled within the Wikipedia-style response it generated, I noted:
“Through an iterative process of feedback and refinement, [generative AI tools] continuously improve their ability to generate increasingly realistic and convincing content.”
So what happens when everyone is using AI to generate their copy, and the vast majority of what the AI is crawling in order to spit out its response is itself AI-generated content?
Lots of people have likened the recent rise of AI to The Terminator’s infamous Skynet. I wonder if comparisons can’t also be made with the Terminator itself. This machine clothed in human form. Real-sounding, yet unable to speak, to think, beyond the parameters of its programming. Which, when you think about it, makes it just as much a parrot as our crows.
The crows. In the story I read, these birds couldn’t just talk. They could also build things. But they could only rebuild things. They might rebuild them differently to how they found them. And you could argue that in doing so, they’re generating something new, but they’re not building anything from scratch. They can’t create something seemingly from nothing. Not in the way that a person can sit down in front of a blank page and by some miracle begin to type.
In other words, whether the crows are talking or building, they’re repurposing something. They’re dependent on what’s already there. On things they’ve seen or heard before. And though they’re capable of taking those things and turning them into something new, there’s nothing to suggest real intelligence. In that way, they’re really no different to a machine.