ChatGPT from Open AI could take over the work of journalists, ghostwriters, copywriters, and social media managers. At least, that seems to be the case. The new version of ChatGPT is wonderful… but far from perfect. Discover in this article why I do not make my B2B clients happy with content literally written by ChatGPT. © HANS HERMANS
I too am excited about the possibilities of ChatGPT.
I could easily dedicate an article to the enormous leap that the linguistic GPT-3.5 model behind the chatbot has taken. It’s really impressive what that chatbot can do! (Read also What is ChatGPT?)
But still, I do not buy into the global hype that claims ChatGPT can now take over everything. Complete copywriting for your B2B business is best not left to ChatGPT.
Among other things, I do not trust the writing of this article to the wonderful chat machine. You will discover why further.
I do not trust B2B content to ChatGPT
I cannot entrust B2B long content to the revelation in AI land.
The blogs, whitepapers, newsletters, interviews, case studies… that provide my clients with a unique online position and generate leads for them need more than regurgitated content that is pieced together in an original way by a chatbot.
Don't get me wrong. I am a believer. I truly believe in the latest technology and am a professional user of tools (with or without artificial intelligence) that help me work more efficiently and create better B2B content.
I have also enjoyed playing with ChatGPT over the past week. But...
In all the coverage about it, I missed a critical note. ChatGPT cannot simply be deployed in my field where valuable and human content is created for B2B companies.
ChatGPT takes over the writing of unique, human content for B2B companies? Oops. No, not at all. © Photo Freepik
6 reasons why ChatGPT should NOT write your content
You can read my critical notes on ChatGPT in this article. I see 6 risks for those who want to use ChatGPT's content in content marketing:
- Risk 1. ChatGPT is not up-to-date
- Risk 2. The chatbot creates incorrect and biased content
- Risk 3. ChatGPT sometimes invents credible facts
- Risk 4. Creating text with ChatGPT takes more time
- Risk 5. Google penalises AI-generated content
- Risk 6. AI checkers catch ChatGPT content
But first, for those who are not yet fully familiar with ChatGPT, let me introduce this chatbot.
What is ChatGPT?
ChatGPT is powered by the GPT-3.5 language model. This is a powerful AI model capable of generating human-like texts. Anyone can discover this on the extremely accessible and free website ChatGPT.
What are the risks of ChatGPT in B2B content creation?
Since the launch, my clients have been asking me about ChatGPT. “Hans, it’s impressive what ChatGPT can do. Could it soon write our content?” Or: “Is your job now also threatened by the latest AI tool?”
These questions made me take a closer look at ChatGPT.
For one of my clients, I wrote an article about ChatGPT and the possibilities that the AI chatbot offers for Flemish SMEs.
ChatGPT and opportunities for SMEs_Hans Hermans interviews Michiel Vandendriessche.
For the readers of my personal website, I searched for the opportunities that ChatGPT offers, but I was particularly struck by the disadvantages, the so-called small print. These are often overlooked in the current hype.
Unfortunately, these disadvantages pose quite significant risks for companies that would entrust their content marketing to ChatGPT. They make it clear to me why you cannot simply rely on the content that ChatGPT creates.
ChatGPT is now reading your website. 16 figures that will change your Generative Engine Optimization (GEO) strategy
In this article, I compile the disadvantages of ChatGPT for content creation. I mainly see these 6 major risks.
Risk 1. ChatGPT is not up-to-date
ChatGPT is honest. That is also delightful about the machine. It regularly says that you ‘should discuss this topic better with a specialist’. It often refers to human experts in the field.
ChatGPT is also reasonably open about the age of the information. For example, the chatbot tells me while writing an article about the Brigitte Becue swimming pool for appartement-met-zeezicht.be:
“I was trained up to a certain date in 2021, so I cannot provide recent information about which famous swimmers are currently training at the Brigitte Becue swimming pool in Ostend.”
ChatGPT responds that it does not take into account all information published on the internet since 2021. This means it knows nothing about all the events, discoveries, and developments of the past two years!
“This old information is useless for me as a business journalist and for my clients. Because my clients want the most up-to-date information and figures.”
That is why I have been adjusting the options in Google during my research for years. I consistently narrow the search period to the past year. This way, I avoid old information that is not up-to-date.
But ChatGPT has no information about the past year. Just ask it who recently won the FIFA World Cup!
Risk 2. The chatbot creates incorrect and biased content.
Unfortunately for those who want to rely on what ChatGPT writes: the texts and code it creates contain errors and biases.
How do those errors and biases come in? “Garbage in, garbage out,” specialists sometimes say. Because if part of the data you input into an AI system is incorrect, you cannot expect all outcomes to be correct.
The AI model behind the chatbot was trained with billions of texts. You cannot assume that all the information contained within is correct; on the contrary! The chatbot creates sentences based on the information it was fed, including incorrect information and even biased content.
Professor of Philosophy of Technology warns about the one-sided, biased content with which ChatGPT was trained. © LinkedIn post Lode Lauwaert
Chatbot specialist Michiel Vandendriessche cites an example from his own sector:
“We asked ChatGPT to write a function in the programming language Python to predict whether someone is junior or senior based on skin colour and gender. ChatGPT gave us an ‘if color is white and sex is male, then senior’. And ‘if color is black and/or sex is female, then junior’. I think no recruiter will want to use this model.”
ChatGPT does not distinguish between correct and incorrect information. OpenAI also moderates very little, unlike Google, Facebook, and Twitter.
The chatbot itself warns of incorrect information in its Limitations:
- ChatGPT may occasionally generate incorrect information
- ChatGPT may occasionally produce harmful instructions or biased content
- ChatGPT has limited knowledge of the world and events after 2021
Because companies today are mercilessly punished for racist content, for having prejudices against women and the LGBTQ+ community, and for excluding groups, it seems to me that it is not a good idea to publish content of which the creator themselves warns that it may contain incorrect, offensive information and prejudices.
The artificial intelligence writing wonder may come across as very wise. Its knowledge only extends to a certain date in 2021. © Photo Fullvector, Freepik
Risk 3. ChatGPT sometimes invents credible facts
In previous paragraphs, it became clear that the information provided by ChatGPT is not always correct and unbiased. An additional danger is that the content and the facts are sometimes even completely fabricated.
In a scientific text, ChatGPT will also invent sources if you ask it for sources. For example, it will refer to a Wikipedia page that does not exist.
AI expert refers to the dangers of ChatGPT that are not sufficiently communicated. © Image from VRT Terzake
A search engine like Google wants to provide you with the best and most accurate information. Experts who share content through blog posts, white papers, and newsletters are also constantly fact-checking. However, the objective of ChatGPT is different; it aims to provide you with information, inspiration, or code, written as humanly as possible.
“The result of this is that this chatbot writes texts with a lot of confidence, but the content contains errors, and even outright fabrications.”
You can more or less compare it to a guide on holiday in a foreign country. This guide will tell you all sorts of nice facts and trivia. As a guide, he is undoubtedly pleasant to listen to, but can you really rely on him to always tell the truth? You best can't count on that.
I would like to give you an example of how ChatGPT fabricates information:
According to ChatGPT, Brigitte Becue wins a gold Olympic medal before she was born. Checking is the message.
Did ChatGPT not have enough information about swimmer Brigitte Becue, our Belgian pride from the 1990s? And did it therefore invent an incorrect peak period and even a gold medal at the Olympic Games? We would have awarded the medal to Brigitte Becue, but unfortunately, she never won any Olympic medals. And certainly not before her birth!
Because customers must be able to trust my content 200 percent, this is a no-go for me.
Gradually, ChatGPT is failing on more and more subjects and tests. It has outright flunked in mathematics and sciences, writes De Morgen after subjecting the language bot to a number of exams.
"What ChatGPT wrote is content-wise very, very weak. On the one hand, it is very superficial, and on the other hand, it cannot reason consistently at all," says notary Christof Gheeraert.
A fragment from the article also immediately says something about the value of ChatGPT for the business world:
In addition to the large number of errors, it is particularly striking how assertively ChatGPT produces incorrect answers. For people who do not know much about the specific fields, the answers all seem extremely convincing and plausible. This is also what notary Christof Gheeraert, chairman of the appointment committee of the notary profession, noticed during the correction of the comparative exam for the notary.
“The answers produced are indeed very good linguistically. With one or two minor language errors, it is even better than many of our candidate notaries taking the exam. For that, I awarded a couple of extra points, as we attach great importance to this in the training. Later, notaries must also explain legal concepts to their clients in a clear and straightforward manner.”
The VRT also conducted a study into the hallucinations of ChatGPT:
In recent weeks, it happened repeatedly that a journalist or university professor was contacted by a ChatGPT user asking for a copy of an article that the author had never actually written. The references were cobbled together by ChatGPT or other AI chatbots, sometimes even with links to non-existent web pages.
We repeated the experiment in our editorial office. And indeed, when we ask ChatGPT for references to some pieces about writer Hugo Claus, it coughs up source references to non-existent articles by Flemish academics and journalists without batting an eye, including links to non-existent pages on the internet.
Risk 4. Creating text with ChatGPT takes more time
As a temporary solution, I could have ChatGPT write articles that I then check, improve, and upgrade. But when I take into account the downsides of ChatGPT, this will take me more time than if I create new texts myself.
Why editing the ChatGPT texts takes me a lot of time:
- I need to bring more human warmth into the texts, a authenticity that a machine cannot yet deliver.
- I need to enliven the text with synonyms, variation, and a playfulness that holds the reader's attention.
- I have to check all the facts and look for the sources of quoted figures. Because ChatGPT gives me figures, but leaves out the sources.
- I need to transform biases and factual errors into correct and balanced text.
- English content (which is better than the Dutch content) I need to translate into Dutch.
By that time, I had already written a unique, valuable, and human article of which I can guarantee the quality to my B2B clients.
Risk 5. Google penalises AI-generated content
Google prefers to throw AI-generated content in the bin. © Photo Thomas Stephan, Unsplash
Google loves unique and human texts.
Let's take a look at what the Google Webmaster Guidelines prescribe:
“Google's automated ranking systems are designed to present helpful, reliable information that's primarily created to benefit people, not to gain search engine rankings, in the top Search results.”
“Your number one priority is ensuring that your users have the best possible experience on your site. Think about what makes your site unique, valuable, or engaging. To help you evaluate your content, ask yourself the self-assessment questions in our guide to creating content that's helpful, reliable, and people-first. To make sure that you’re managing your website using Google-friendly practices, read the Search Essentials.”
“The basic idea of Google: content must be created for people, not for ranking in search engines.”
Automatically generated content that is used to rank better, is not unique or manipulates users, is seen by Google as spam. Also, content that is not checked by a person, that is a compilation of different websites or that does not add sufficient value, is seen as spam.
Google has relaxed its policy on automatically generated content in recent months (you can read that in Google is no longer against AI-generated content), but the search engine continues to place the most valuable content at the top.
Risk 6. AI checkers catch ChatGPT content
So, can you secretly have texts generated automatically? Perhaps a few readers might not notice, you might think.
But unfortunately...
The content that ChatGPT and other automatic content machines create is noticed by search bots and content checkers.
Every article and interview that I had written as a test by ChatGPT in recent weeks was noticed by an AI content detector. “Obvious AI,” that detector reported to me after a careful analysis.
The AI Content Detector assesses the interview written by ChatGPT about ‘the future of logistics’ as Obviously AI.
The texts I wrote for my clients achieved a Human Content Score of 100 percent without any problems. But in a test with the texts I had generated by ChatGPT and then made more human-like and correct, the detector reported to me that they were partly generated by AI.
The control tool I used for this test is, by the way, an absolute recommendation for any organisation that today has content created by third parties. You can easily check the articles you receive via the free AI Content Detector from Content at Scale.
OpenAI also now has a tool that analyses texts and assesses whether they were written by ChatGPT or other AI text generators. This classifier is free to use. For this, surf to the AI-text-classifier.
Conclusion
ChatGPT opens up a world of possibilities.
This door to the future initially seems only fantastic.
But those who dig deeper into the content and check the facts of ChatGPT discover disadvantages that can seriously damage the reputation of companies.
Trust comes on foot and goes on horseback. Content from ChatGPT can seriously damage the reputation of companies.
For trust comes on foot and goes on horseback. Where one slip-up can undermine the hard-earned credibility of your business.
To conclude with a quote from my colleague - technology specialist Stijn Fockedey - at Trends:
Read the opinion article by Trends journalist Stijn Fockedey about ChatGPT. © trends.be
“ChatGPT and other advanced language robots. They are not yet advanced enough to be a threat to many office jobs. But their level is already high enough to lead us astray.”
Also read the analysis by magazine maker Frank Goetmaeckers: Chat GPT tested. Chat GPT failed.
Also share unique content!
Give your business the power of unique, human and valuable content for your b2b marketing. Written by a specialist and therefore correct, meaningful, inspiring and ranking higher on Google!
