There’s been a huge amount of hype lately around AI article writing software and copywriting AI tools. Everyone and her wife seems to be voicing an opinion about ChatGPT, and I’ve had not one but two clients send me blog posts generated by artificial intelligence writing software to ask if they’re okay to post. The short answer is no, they’re not.
But, I’m a copywriter, and clearly going to be biased on this subject. Many are saying the days of copywriting are over and AI will soon have my job. I beg to differ. While I admit the advances made in AI writing software are impressive, the technology is still nowhere close to being able to do the job of a trained and experienced copywriting professional.
And in order to prove this point I decided to pit myself against ChatGPT and see who the Google gods favoured more. To that end, I charged ChatGPT with a simple task: to write a blog post. I gave it the parameters needed: a title, a keyword to optimise for, and the desired length. I then wrote my own version of the same blog post, published my version on this site, and the AI’s version on Acrylic Digital’s site (that’s the marketing agency I used to work at, who were happy to oblige me in running this little experiment).
I’m going to give it a few weeks and see which performs better.
If you want to see what happened while I was using ChatGPT, I suggest you watch the video. If you want to see the various conclusions I have drawn about the software since having it write this post, and comparing it to a post I wrote myself to the same parameters, read on.
Yep, I used an AI article writer so you don’t have to, and understand why you shouldn’t. Here are 13 reasons why ChatGPT sucks…
Skip To The Best Bits...
#1 An AI Article Writer Can’t Create A Compelling Voice
While AI has made significant advancements in natural language generation, it still faces limitations when it comes to creating a compelling voice in article writing.
AI models have access to vast amounts of data and can retrieve information quickly. However, they lack real-time and experiential understanding of the world. They may struggle to grasp current events, cultural references, or the latest trends, making it difficult for them to create articles that resonate with readers on a deeper level. A compelling voice often requires a deep understanding of the subject matter and the ability to relate to the reader’s experiences.
Plus, AI article writers lack personal experiences, opinions, and subjective viewpoints. They rely solely on patterns and data from the training data they were exposed to. This absence of personal perspective can make the writing feel detached and impersonal, failing to connect with readers on a more human level. A compelling voice often reflects the author’s unique perspective, which can be difficult for AI to replicate convincingly.
While AI has made significant strides in generating coherent and readable text, creating a compelling voice in article writing involves a level of creativity, emotional intelligence, and contextual understanding that current AI models have yet to fully achieve. Human writers, with their subjective experiences and ability to connect with readers on a deeper level, still excel at crafting articles with a captivating voice.
#2 AI Writing Software Lacks Emotion
Another lack in AI models lack is emotional intelligence and understanding. While they can generate text that appears coherent and grammatically correct, they often fail to grasp the nuances of human emotions, which are crucial for establishing a compelling voice. Emotional connection is a powerful tool in writing, and without it, the writing may come across as flat or lacking in authenticity.
#3 Your AI Writing Tool Can’t Optimise For Search Naturally
In theory, an AI article writer can optimise for search effectively if it is programmed or trained to do so and has access to relevant data and algorithms. However, there are a few reasons why an AI article writer will struggle with SEO.
For starters, AI models are typically trained on large datasets, but the training data may not include up-to-date information on SEO best practices. SEO techniques evolve over time, and search engines frequently update their algorithms. If the AI model’s training data does not include the latest SEO strategies, it may not be able to produce optimised content.
A larger issue is that effective SEO involves understanding user intent and creating content that matches those intentions. AI models excel at generating text based on patterns in training data, but they may struggle to grasp the context and intent behind a search query. Without a deep understanding of user intent, it becomes challenging to optimise content effectively for SEO.
SEO optimization often requires access to real-time data, such as keyword trends, search volume, and competition analysis. While AI models can process and analyse data, they need to be regularly updated with the latest information to make informed SEO decisions. If the AI article writer does not have access to real-time data or is not programmed to utilise it, its optimization capabilities may be limited.
While AI models can generate coherent and relevant content, they may lack the creativity and intuition that human writers possess. SEO optimization often requires a balance between satisfying search engine algorithms and engaging human readers. AI models may struggle to produce content that strikes this balance effectively, leading to suboptimal SEO results.
Search engines like Google regularly update their algorithms to improve search results and combat spam. These algorithm updates can significantly impact SEO strategies and ranking factors. AI article writers may not be able to adapt quickly to algorithm changes without updates to their training or programming.
#4 Keyword Stuffing For The Win
I asked ChatGPT to write a blog post with the title ‘top tips to building a brand story that converts’ and optimise it for the term ‘building a brand story’. While it was spitting out words I noticed it included “building a brand story” several times, in speech marks. What I didn’t notice until I uploaded it was that the final sentence of the first paragraph read:
In this blog post, we will share some top tips to help you build a brand story that converts, optimising it to rank for the keyword “building a brand story.”
Yes, really.
Not only did ChatGPT fail to realise that instruction wasn’t relevant to the blog and shouldn’t be included in it, the AI’s idea of optimising a post for that keyword was to literally stuff in “building a brand story” as often as possible.
If you know anything about SEO or have been following me for any length of time you’ll know that keyword stuffing is an archaic practice that will NOT help you rank highly. It will, in fact, lead to you being penalised by the search engine gods for poor practice.
#5 Your Copy Will Lack All The Relevant LSI Terms
Again, in theory, AI can be programmed or trained to write copy that includes Latent Semantic Indexing (LSI) terms for a given keyword. Based on my experimentation with ChatGPT this isn’t the case – there was no evidence of LSI terms used, either in the blog post I asked it to write or various other tasks I had it perform to test its limitations.
LSI is a technique used by search engines to identify the relationship between words and concepts in order to understand the context of a piece of content. An AI model can be trained on a large corpus of text data that includes LSI terms related to various keywords. I suspect that someone will create an SEO-specific AI model that is capable of fully writing and optimising a piece using LSI terms, but I’ve yet to see one. If you use tools like Yoast or Rank Math they have the beginning of this already, but they are far from being developed enough to write a full piece from scratch and optimise with LSI.
Theoretically though, the capability is there. By learning patterns and relationships in the training data, the AI can generate copy that incorporates relevant LSI terms for a given keyword. For example, let’s say the keyword is “digital marketing.” The AI model can analyse the training data and identify LSI terms that are commonly associated with digital marketing, such as “SEO,” “social media marketing,” “content strategy,” “online advertising,” and so on. It can then use this knowledge to generate copy that includes these LSI terms, enhancing the relevance and context of the content.
The problem is that the effectiveness of an AI generating LSI-rich copy depends on the quality and diversity of the training data. The AI model needs exposure to a wide range of text sources that cover various topics related to the keyword. Additionally, the AI model should be regularly updated with the latest information and trends to stay relevant and accurate in incorporating LSI terms.
As it stands, while AI can assist in generating copy that includes LSI terms if it’s been programmed to do so, that AI still cannot match a human, and would need checking and editing by a human (at the very least) to make it a coherent piece of writing that flows naturally and delivers value. And that’s if the AI has access to the right training data. Most of them don’t.
#6 You Can’t Rely On The Character Counter
When you provide a specific word count to an AI writer, it tries to generate content that meets that requirement. However, without explicit constraints on the content structure or limitations on the generation process, the AI may tend to be more verbose or include additional information. It aims to provide comprehensive and detailed responses, often erring on the side of generating more words to ensure it covers the topic thoroughly.
AI models are also trained on vast amounts of data from the internet, which can contain biases, including the tendency for longer and more detailed articles. These biases might influence the AI writer to generate longer responses as it emulates the patterns it learned during training. In other words, you ask it to write you a meta description of no more than 160 characters, and it won’t understand that the character count is pivotal to your SEO. It will prioritise writing a description that includes more information. It may or may not include the keyword you ask for in that, and if it does it doesn’t know enough to make sure it’s in the first 160 characters.
The problem for many people trying to use AI to generate metadata is that they don’t know enough about SEO and metadata to recognise these issues. My OH asked for a meta description from ChatGPT last night and read out the response to me. The first thing I said when he’d finished was ‘that’s over 160 characters’. Her frowned, and told me he’d specified that it be under 160. I said, “Sure. But it’s not. That’s more like 240.” Turned out to be 242.
Needless to say he was impressed by my ability to judge the character count based on hearing the description. But that’s what happens when you spend a decade writing things that have to conform to a specific character count; you know when they don’t.
AI isn’t aware of any of that information unless someone has taught it, and even if it is, that doesn’t mean it will prioritise your character count even when you tell it to.
#7 Your Content Won’t Contain A Compelling Narrative
While AI can assist in generating story-like content or provide inspiration to human writers, it currently falls short in delivering the depth, creativity, and emotional impact that humans can achieve in storytelling.
AI models do not possess true consciousness. They lack the ability to deeply comprehend emotions, complex human experiences, and the subtleties of storytelling. As a result, they’re incapable of generating stories with the depth, nuance, and emotional resonance that human authors can imbue in their writing. They’re even less capable of infusing an effective narrative into an article, blog, or other piece of more formal content.
Storytelling requires creativity, imagination, and the ability to generate original ideas. Writing a good story involves maintaining consistency, logical progression, and narrative coherence. As they generate text based on local context and patterns without a broader understanding of the overall narrative structure, attempts to generate a narrative result in disjointed or incoherent storytelling.
Stories are also deeply influenced by cultural references, societal norms, historical events, and individual perspectives. AI models do not have a comprehensive understanding of these factors or the ability to incorporate them effectively into their storytelling. The result is content that lacks cultural relevance, authenticity, and an understanding of the human condition.
AI also struggles to evoke genuine emotions, empathy, or a sense of relatability because they lack the personal experiences and emotional intelligence that human writers possess.
#8 AI Cribs From Other Content
May not be identical to existing content, to the point it flags as duplicate content, but an AI can’t include anything about the topic at hand that doesn’t already exist on someone else’s post about that topic. The AI literally scans existing content and uses all the same information and concepts to answer the question you have posed, or deliver the content you have charge it with creating.
And how are you going to outshine the competition if you don’t have anything original in your copy?
#9 Longform Content Does NOT Come Naturally
I specified to ChatGPT that it should write a blog of 2500 words. It tapped out at the 600 mark. I asked it to elaborate on the post and write more words, and in fairness it obliged, but only to the tune of another 600 or so. Repeating this a third time got me to around 1700 but when I asked again, the poor thing melted.
By which I mean slowed down to a snail’s page because it was really struggling to find more to say on the subject. Eventually it just screamed red and errored out.
The reason for this is simple: ChatGPT has a limit of 2048 characters per response. It just can’t write more than that in one go. Likewise it’s not able to absorb more than 3K words of information at a time (that’s on the free version, the paid version is 25K). You can physically input more information than these word counts, but the AI will only ‘read’ up to its word limit. Anything input after that will be ignored.
#10 It Can’t Include Cross-Links For You
One thing it’s super important to do while adding fresh content to your blog is cross-link to existing content. This requires you to work references to subjects you’ve written on before, that are relevant to your current topic. You then use the text that best represents the subject of another article as anchor text to link to that article.
Including several cross-links in each of your blog posts not only boosts the SEO of that blog, it also gives the posts you’re linking to an extra boost. Most importantly, a comprehensive cross-linking strategy ensures Google and other search engines can easily crawl and index your whole site and all its content.
Not only do you have to add any cross links manually to a post generated by an AI writing tool, you’ll also need to add the small references throughout the piece that allow you to naturally link out to that content.
#11 Statistics CANNOT Be Trusted!
In the course of testing ChatGPT I asked for ideas on blog topics, to see if it could be useful as an ideation tool. As it turns out, it’s super helpful in this regard. I ghostwrite a weekly post for Portugal News for one of my clients, a financial investment firm, and honestly coming up with a topical subject that’s relevant, current, and related to finance each week is seriously tough. If there’s something happening in the news that impacts finances it’s fairly easy, but on slow news weeks it’s genuinely tough; everything’s already been written about at this point.
So, I asked ChatGPT to come up with topic ideas, and it did, including one on Portugal becoming a cashless society. I did actually end up writing a blog on this, as it was an interesting topic, but I researched and wrote it myself (for all the reasons on this list). The idea the chat spat out, however, included a statistic. I didn’t save it, but it was something like ‘80% of Portuguese consumers no longer carry cash, is Portugal ready for a cashless society?’. I thought the topic was interesting, so I asked for the source of the statistics so I could research it. That is when ChatGPT informed me that it made the statistic up.
Not joking!
Here’s how that convo went:
Me: “What is the source for the stats?”
ChatGPT: “I apologise for the oversight in not including a source for the statistic mentioned. Unfortunately, as an AI language model, I do not have access to real-time news or statistics. I generated the statistic as an example based on hypothetical data. Please note that any resemblance to actual data is coincidental.”
Yep. Really. When this thing spits ‘facts’ it’s making them up!
#12 Their Ability To Respond To Feedback Is Limited
AI article writers operate based on pre-programmed algorithms and patterns learned from training data. While they can make adjustments based on specific instructions, they struggle to adapt to real-time reader feedback or engagement. A compelling voice is often responsive to the audience’s reactions, adjusting tone, style, and content accordingly. Without this adaptability, AI-generated articles may lack the dynamic nature that makes writing engaging.
#13 There Are Potential Ethical And Moral Issues
AI-generated content can raise ethical concerns, such as the potential for misuse, propaganda, or spreading biassed information. The lack of a moral compass or the ability to differentiate between what is appropriate and what is not limits AI’s ability to create a compelling voice that aligns with societal values and principles. Authenticity and ethical considerations are important factors in establishing a trustworthy and persuasive voice in writing.
So, Can Artificial Intelligence Write For Us?
There is no possibility that an article writing AI could replace a human copywriter. At least, not in their current state. Whether they ever develop to the point that is possible remains to be seen. Technology is advancing seemingly exponentially, it seems likely that, at some point, an AI capable of emulating human emotion, humour, personality, storytelling capabilities, intuition, and natural cadence will be developed.
So, if you were thinking you don’t need to hire a copywriter because you can just let AI write you copy and content for you, I’m afraid you’re mistaken. I mean, you can. It’s physically possible to use ChatGPT and other AI tools for content writing to create everything you need. The content will exist. But will it be any good?
The answer to that is subjective; to some it will be decent content, to others it will be profoundly dull and uninteresting. Neither camp will regard it as inspired, but do we all need inspiring content? Perhaps not. Regardless of how you personally feel about the content created by your AI, the salient question isn’t how much you enjoy it, but how well it performs.
You’re running a business, after all, your content creation has a purpose. So, the real question is, will AI generated content perform? Or, more specifically, will it outperform a piece of content written to the same parameters by a professional copywriter?
Well, I decided to test the theory one step further. After all, it’s one thing to compare the two and for ME to say ‘mine’s great, the one created by ChatGPT is shit’. But let’s see what the Google Gods have to say.
I’ve posted one post on this site, and the other on a different site. Both are roughly comparable in terms of authenticity and trust, both have been optimised for exactly the same search term.
Let’s see which ranks higher.
