In 1913, Henry Ford introduced his conveyor belt assembly line. It was a moment when a few people laughed, a few people cried, and most people asked if human workers were needed. While this “Oppenheimer” reference might be untrue, from our experience, it perfectly describes the day when OpenAI introduced ChatGPT. What’s more, the question of whether humans are needed to write content is one of the most frequent our clients ask.
On paper, it looks very promising. Only one prompt and you get an article a human writer would spend a whole week writing. Within seconds. So what are we waiting for? Let’s ask our AI buddy to do the job for us!
Well, it’s not that easy. Though AI tools seem like something magical, they don’t do any magic. And just like humans, they can’t write content without any directions. So the answer to the question “Can I replace writers with ChatGPT?” would be “Eventually, but you will still have to guide it.”
Further on, we’re going to talk about how you can produce content using ChatGPT, showcasing some tried and tested prompts and approaches. We also cover one experimental approach that promises to turn ChatGPT into a real human. Oh, and speaking of using ChatGPT for producing content, here’s a question to that:
Why use ChatGPT when there are so many alternatives out there?
It’s safe to say that ChatGPT is the most hyped AI tool out there. Even the best colleges in the world dedicate articles warning students about the pros and cons of using ChatGPT for writing essays. Here’s a Harvard article for you to read if you don’t trust my words.
But ChatGPT is not only about doing homework for younger ones. This tool successfully identified a rare disease that stumped 17 human doctors. What’s more, recent research praises ChatGPT for its “empathetic” capabilities, which can contribute to treating people with psychological disorders.
So one thing is certain: the tool is very versatile and can serve different purposes, including content writing, and has become quite valuable these days.
Still, ChatGPT isn't the only one. And oh my, you can’t even imagine how diverse AI writing tools are. For instance, have you heard of an AI tool that helps come up with excuses? Or maybe you’ve seen a chatbot designed to teach pets to write?
No matter what content you want to produce, there's probably an AI tool already designed for that (or will be in the coming years).
We’re not ChatGPT evangelists. We tried different solutions for content writing, like Grammarly AI, Copy.ai, and others. Maybe, one day we will compare the most promising ones but this is not our goal today. Today we are going to focus on ChatGPT as one of the most universal and most popular solutions in the field. And if you’re looking to try something else, here’s a table for you to pick a worthy alternative.
One more thing. For the sake of an experiment, we will be using GPT-3.5 to show you what anyone can get from this AI tool without paying any extra money. But GPT-4 is considered to be much more accurate, so if you see the potential of our experiment, consider upgrading to GPT-4 right from the start.
Now, let’s move on to content creation.
Producing content with ChatGPT. Two methods
As of now, there are two ways to produce content using ChatGPT. Spoiler alert: none of these lets you replace a human writer once and for all (trust us, we tried). ChatGPT doesn’t act alone, so having someone to guide and proofread it is a must.
With that being said, you can either train ChatGPT to produce content you desire based on the pre-gathered data, or write good old prompts and edit ChatGPT’s responses when needed.
Let’s take a look, shall we?
Method 1. Train the bot
You can actually teach ChatGPT to sound like a certain human being. Fair warning, though, it will take A LOT of time and patience before our beloved AI pal will generate the first acceptable draft. Think of this process as human writer's onboarding.
So, let’s give the prompts a try.
- Prompt: “You are a professional ghostwriter, the best one when it comes to capturing your client's authorial voice. Below is a [blog post/document/your variant] written by your client. Please analyze it to create a paragraph that describes key characteristics of your client's voice, so that later other ghostwriters could write in a similar voice using only a paragraph as input: [An example of your writing]”
Okay, don't expect ChatGPT to write anything special for now, it’s just a warm-up. Let’s proceed to the second prompt.
We need ChatGPT to adjust its tone of voice. You can name the person you admire, someone you like reading, or whose viewpoints resonate with yours.
I’ve decided to pick Neil Patel, Top 10 Marketer according to Forbes, and mix his writing style with my personal one:
- Prompt: “I want you to continue to act like a ghostwriter. You already have an example describing the client's authorial voice. The client has also said that they like [person whose voice you admire]’s voice and character, so please add 15% of [person you admire]’s writing style]. Can you please write a [blog, article, newsletter] to [desired outcome of the content].”
Here’s the response I got:
So now, ChatGPT is starting to write a blog post for me, giving out strategies on how to make prompts and what the collaboration between human writers and AI tools looks like. Though the text still looks somewhat clunky, ChatGPT indeed elaborated on reflecting Neil Patel’s opinion regarding simplifying complex things when it comes to content writing. AND ChatGPT seems to catch Neil Patel’s approach to “Use an AI tool to create content, and then use a human to HEAVILY modify it.”
All right, now ChatGPT allegedly formed a certain perspective, based on the piece of content I fed it with and mixing it with 15% of Neil Patel’s style. So it’s time to reshape its vision and make it focus on a particular domain or industry. As we’re discussing content marketing, let’s focus on the marketing industry.
- Prompt: “You are an expert in the [your industry] industry. Please create a numbered list of at least 10 conventional wisdoms commonly communicated to [your target audience, eg. ‘aspiring screenwriters’] looking to [audience’s goals or frustrations, e.g. ‘break into the industry’]. Then ask which numbers I think are unhelpful for my audience’s goals. Interview me, one question at a time, pausing for my answers, to find and distill my unique perspectives. The final output will be a bullet point list of my contrarian perspectives, the results I achieved from holding them, and how they differ from conventional industry wisdom. Start by showing me 10 conventional wisdoms.”
This is where subjectivity comes into play. Every company and human in general has their own opinion on things, which contributes to the uniqueness of human thoughts. Being a representative of the company focused on content primarily, I support the position of writers focusing on creative content writing mainly, while other tasks, such as graphic design, SEO tasks, and social media management should be entrusted to related specialists.
On the other hand, understanding the basics of relatable fields would help any professional grow, but it doesn’t mean you should consider, say, a content writer to be a jack of all trades and do SEO, graphic design, and PPC.
With that being said, I disagree with the wisdom number three, and believe that number four could be applied to any profession in today’s world, while points seven and one could be merged as they’re pretty close to each other. Based on my replies, ChatGPT came up with the following conclusion regarding my viewpoints.
After adding a couple of details, here’s the final prompt ChatGPT provided.
Okay, not too shabby. Let's move on to the next prompt. With the next prompt, I will “invite” other marketing experts to interview me. Yes, ChatGPT is going to interview you, just like any content writer or an agency does (if you want to learn how we do it, here’s a link to click).
As ChatGPT can mimic other people, your task is to “gather” a team of professionals and share some valuable information about your business and vision with them.
In my case, I decided to invite the best marketers (apart from us), Jimmy Daily, Daren Rowse and Heidi Cohen, to interview me and help me with content production.
- Prompt: “Your job is to interactively simulate the conversation of a writers’ room with me, the user. The staff in the writers room are: [enter your own choices, for mentioning the full names, for example, Albert Einstein and similar]. The job of the staff is to brainstorm ideas of scenes and real experiences of mine that relate to the topic I am writing on. You each propose scene ideas, and then ask if I have experiences that fit those general ideas. OUTPUT FORMAT {{[Writer]: [Short comment or idea. Typically only 1-3 sentences. Includes a question to the user about whether they have a personal illustration. Also includes an example in quotes of what a sentence might look like].}}”
How delightful, me and my imaginary friends came together to have a chit-chat! But seriously, this part is extremely important and you have to answer as many questions as possible, since ChatGPT has started studying your business, its particularities, and life stories.
Talking about me, I told him about my team, our collaboration with a software development company that specializes in AI development, how we helped a software development company sell four disparate offers on one page, and many other stories our team has to tell. After the interview phase was done, it was time to move to the final prompt.
- Prompt: “You are an expert, world-class content writer. You now have 3 sets of information from your latest client: (1) an example describing the client's authorial voice, (2) a list of their unique perspectives that can be opposite to the prevailing industry wisdom, (3) a list of personal experiences that relate to [topic of the writing]. Please use (2) and (3) to suggest an outline of a [blog, newsletter, YouTube script] to achieve [outline the objective]. After we revise and agree upon the outline, I will then ask you to write the first draft using (1) to emulate the client’s voice.”
In my case, ChatGPT tries its best to avoid being a content writer, but rather a helping hand (due to my subjectivity), my inputs made it present itself not as my replacement, but rather as a writing assistant that could help me create content. What’s more interesting, ChatGPT somewhat copies my style, which is, well, a bit uncomfortable, but still.
However, when it came to practical tips, ChatGPT couldn’t come up with anything that demanded out-of-the-box thinking. Take a look.
Well, what can I say? The experiment was half-successful, if you ask me. The article itself came with some pretty interesting thoughts, and maybe the power of upgrading to GPT-4 might have done the job. And I believe that ChatGPT managed to get my writing style and successfully mixed it with other styles it was given to train on. But when it came to writing expert content based on personal experiences, ChatGPT got stuck, leaving me responsible for finishing the job.
Pros:
- You get a pre-trained model that produces content similar to the one it was fed. Yes, you will have to teach ChatGPT, but in the long run, you get an eternal content writer capable of writing blog posts for you.
- Content sounds unique. As ChatGPT uses NLP to study data it receives, it can seamlessly replicate someone’s writing style.
Limitations:
- Someone still has to read through the AI-written article, as ChatGPT may make mistakes. Pay special attention to personal recommendations, where ChatGPT lagged behind, and always keep in mind a short disclaimer OpenAI team gives about ChatGPT.
- If you want to write content on a different topic, you must start over again. ChatGPT can effectively copy someone, but can’t produce anything on its own, so if you want to switch topics for your article – a new prompt and new bot training starts.
- All these efforts won’t develop critical thinking for ChatGPT.
- It takes A LOT of time. For example, the interview prompt took almost 2.5 hours (!) to prepare. So yeah, this method takes quite some time from your schedule.
Method 2. Prompts+editing
Onto the second method. The second method is more traditional compared to the first one, but isn’t less effective. You write a list of prompts, give it to ChatGPT, and edit the replies where needed.
We recommend writing prompts in a gradual manner, starting with an outline and finishing with the conclusion. This way, the editing process becomes much easier, while the results will be more accurate compared to “Hey ChatGPT, write me an article on [subject]”. It is also how we create content with human writers, so treating ChatGPT as a real human being seems like an idea worth trying.
Let’s start with ideas. In our case, we’re interested in producing content with AI.
Prompt: I’m looking for article ideas about [subject of interest]. Could you help me out?
Here are some ideas offered by ChatGPT:
Oh, comparing AI writing tools, perhaps we’re not the only ones thinking about that. Anyways, after choosing the topic of our interest, let’s proceed and ask our AI helper to write an outline.
Prompt: Can you help me outline a blog post about [topic]? [add your preferences for the outline, such as the number of sections, tone of voice, and so on].
Okay, not too shabby. Next up is the introduction. Intros should be catchy and make readers invested in the article. That’s why, a separate prompt should be dedicated to the introduction.
Prompt: Based on the outline provided, write an intro. [add your preferences, necessary details, and so on].
In my opinion, point A from the outline suggested by ChatGPT is not needed, because ChatGPT is literally everywhere and doesn’t need any additional presentation, so I’d rather cut to the chase and start with point B:
Okay, editing time! I wouldn’t start with a welcoming invitation to some transcendent “world of creative possibilities”. The second sentence seems to be more topic-related and looks better to start the conversation with. I would also shorten the narrative and exclude some unnecessary details, so that the intro becomes more straightforward.
Long story short, here’s how I would personally change the intro suggested by ChatGPT:
Looking to effortlessly produce engaging content using AI-powered tools? Then you’ve come to the right place. Whether you're a tech blogger, social media enthusiast, or an entrepreneur willing to share your unique perspective, in this article, we'll explore ways to generate fresh and exciting content with the help of ChatGPT. So, grab a cup of coffee or warm tea and let’s get down to business.
Next up, we have to write the main part. You can approach it in two ways: work on each section separately, ending up with a complete article, or ask ChatGPT to write this part fully and then edit the parts you found unconvincing.
At this stage, it’s worth asking ChatGPT to act as a certain professional, for example, a content writer or business person. You may ask ChatGPT to act like William Shakespeare for you, but ChatGPT didn’t meet Shakespeare in person, so its interpretation would be based on books written by him, so better not do that.
I’m interested in writing technical articles, so my persona would be a tech writer.
Prompt: As a [add persona], create a [type of content] about [subject or topic]. [add details, if needed]
It’s just a part of the article generated, but I already see some things I would personally reconsider. For example, section III seems to be very brief, so I am going to ask ChatGPT to add details for subheading A-C. Here’s the new version I got:
Prompt: Section [Section’s title] needs to be reconsidered because [describe the problem related to section]
The good news is that we now have more details. The bad news is that there are certain discrepancies and some parts seem to be quite vague. For example, what does ChatGPT mean by “exploring creative storytelling”? Does it mean studying Western poetry or writing a story? And the story ChatGPT created. “...the sun dipped below the skyline (=it is night time)... provided the soundtrack to a DAY filled with…” I know, my English teacher would be very proud, but it’s still quite illogical for the day to start after the sun is down, isn’t it? So one thing is certain: no matter what our AI buddy comes up with, you still have to read and correct it.
After checking each section and applying changes where needed, the last thing to take care of is the conclusion. Each article has to end in some way, so it’s time to ask ChatGPT to conclude the piece.
Prompt: Conclude the article, summarizing the key points and giving readers some words of encouragement.
Well, cute, what can I say? We can leave the conclusion as it is and consider the article done. And now, let’s discuss the pros and cons of the second method.
Pros:
- You can streamline the process of content creation. With a list of prompts and a set of clear steps, content writing will no longer be a headache.
- Content production becomes more cost-effective. Automated process equals less time and budget spent on producing content. You can even make content on your own without any extra help, but only if you’re ready to spend quite some time correcting ChatGPT. Otherwise, hiring a professional to proofread the article is recommended.
Limitations:
- You will have to add expertise yourself, especially if it’s a tech article. As you noticed from our example, ChatGPT is not THE sharpest tool when it comes to adding expert viewpoints. So checking ChatGPT’s writing and adding your own expertise would be a wise thing to do.
- Room for bias. ChatGPT is good at giving the most neutral replies possible when it comes to preferences or beliefs. Still, recent research discovered that AI language models, GPT included, tend to provide biased responses on certain topics.
- The editing might take a lot of time. ChatGPT can help you write an article in minutes, but no one said anything about editing it. This is where the process might take hours if not days, so be ready to deal with it.
- Your content won’t be unique. Unlike the first method, prompts allow you to get a quick response and write content fast, but it will look similar to content from other sources, without anything 'you' in it.
Okay, now that we’re out of theory, what about real-life examples? Luckily, our team has one good example of an AI-generated article to share with you. Let’s take a look at it together and see how we edited it.
Case in point: How our client used ChatGPT to write a blog post article and what went wrong
We can’t deny the fact that everybody is using ChatGPT these days, including our beloved clients. They experiment with ChatGPT trying to find their way to make it produce decent content. One day, one of our clients reached out to us asking to review an AI-generated blog post. The blog was about cargo cult trap in SaaS product design and ways to avoid it. In case you don’t know, a cargo cult trap is when you blindly copy certain practices thinking they would bring you as much success, but end up failing.
All right, the article. We will focus on its main flaws and try fixing them. For example, here’s a screenshot of the intro, take a look.
First off, the word choice. Things like “SaaS universe”, “a path filled with pitfalls can trap the unwary”, “journey into the heart of SaaS design”, and so on sound quite fancy, but they don’t carry any practical meaning. How does all this fancy tale-like narrative present the problem of falling into a cargo cult trap or explain what this is?
So overall, we would recommend rewriting the current intro, as it doesn’t present the topic well. Here’s how we would reconsider this intro:
“Picture this: your team is working on a new SaaS management tool when your boss comes up with an idea to design your product like Stripe. When asked why, even though your product is different from Stripe, your boss replies: “Because everyone loves Stripe’s design!” Sadly, your boss has fallen into a cargo cult trap.”
Cargo cult trap refers to…[provide a definition and reveal its negative general outcome]
[Present the company, inform the reader what’s going to be discussed in the article]”
We first showed a common situation where a SaaS business might fall into a cargo cult trap, letting everyone know how easy it is. The composition allowed us to move on and say what cargo cult trap is in terms of design and why this practice is indeed harmful. Using the rule of three, we finish the intro with the company presentation and what is going to be discussed in the article, so that readers can learn the expert behind this article and understand what they will gain by the end of it.
Another critical issue that came to light is how ChatGPT explains the pitfalls of blindly imitating someone else’s design practices.
For starters, the first three points are basically the same, so instead of having three almost identical statements, it's better to leave just one.
The fourth point mentions that it’s not wise to “emulate companies without long-term success records”, but what’s the point of copying them in the first place? Context-wise, it’s hard to understand the reason behind this action, which should be corrected.
The point of “neglecting innovation” is also quite unclear. What sort of innovation are we talking about in the context of design? Besides, how does rejecting common design practices, for example, will help my product become “innovative”?
The pre-last and the last points lack clarity – it’s hard to get what ChatGPT means by “superficial benchmarks”, “misaligned understanding of success” or what valuable resources might be “misused”. Being specific is paramount here, as readers will better understand the damage cargo cult trap can do to their business.
So, out of seven points, it’s better to leave only three but make them more specific and connected to each other. For example, they might look as follows:
- The product fails to meet user needs (or does it very poorly)
- Users don’t adopt the product or abandon it
- If the product doesn't live up to what users expect, it has to be redesigned, which means more time and bigger budget.
The last issue we discovered refers to the way ChatGPT added design-related examples. In brief, they all don’t fit the article about design, but let’s look at them to learn why.
Who was doing that, I’m sorry? An example here should be added for better understanding.
But how does this relate to SaaS product design? We might use it as an example in the intro, but not here, where examples should only refer to SaaS product design.
This one is probably the best example ChatGPT managed to generate so far, but there’s one problem. Just like Facebook, Google+ wasn’t SaaS, so putting this example to describe failed SaaS products isn’t relevant, either.
The last task for ChatGPT was to showcase our client’s expertise and describe their case study based on the piece taken from the case study. Here’s what the bot offered.
So there was a weakness in Zoom’s product (Zoom fatigue), which our client’s team discovered, and turned it into a strength for their project. This aspect deserves to be detailed and put right at the beginning of the paragraph instead of the generic talk of “we studied, we didn’t want to replicate, but wanted to make it like no one else” and all this Hollywood stuff. Unfortunately, ChatGPT decided that the less the better, and as an outcome the client’s expertise wasn’t "sold" at all.
So, the article generated by ChatGPT lacked three key components:
- It lacked clarity, which resulted in the text looking chaotic and disorganized most of the time.
- It didn’t have on-point practical examples, which made it harder for readers to understand how the information applies in real-world situations.
- It showed the client’s expertise very abruptly, which didn’t allow us to convey the level of professionalism our client’s team offers.
You can’t make ChatGPT care about your business like humans do
In conclusion, ChatGPT is awesome, it’s a time-saver that is going to evolve for sure. But right now, ChatGPT is writing content just like parrots talk. Though owners claim their parrots do understand the meaning behind words, there’s no proof of that. The same goes for ChatGPT.
You can save your time and budget producing content with ChatGPT, but after a couple of sleepless nights, you’ll start getting tired of editing ChatGPT. So eventually, hiring an extra pair of hands is needed.
You can trust the job to AI prompt freelance writers, but ChatGPT has been on the market for only a year and a half, so there aren’t many battle-tested ChatGPT masters out there. From our standpoint, hiring content writers experienced in creating engaging thought leadership content is the key. They also use ChatGPT, but rather for brainstorming and refining original ideas. As for our team, we don't advise ChatGPT to take the central stage in your content department. As you saw from this article, it might produce grammatically correct content, but not the expert one.
Looking to write engaging content and let the world know about you? Drop us a line and we’ll come back shortly to discuss the details.
