-
AI imperfections
-
Quality depends on the data
-
AI still needs humans
-
Value of AI-generated content marketing
Will content marketers and copywriters become redundant in the AI age?
With its instantaneous ability to create text and images, generative artificial intelligence is transforming how we work. And content creators are in the front (firing) line.
As AI ethics adviser and researcher Ravit Dotan observed: “Jobs that are more likely to be replaced or negatively impacted include editing and copywriting jobs, as generative AI is already relatively strong in these domains.”
A 2023 study conducted by US researchers just a few months after the November 2022 launch of ChatGPT appeared to confirm such predictions. The study, reported by the Financial Times, showed that copywriters and graphic designers on major online freelancing platforms suffered a large drop in the number of jobs they got, and even steeper declines in earnings. “This suggested not only that generative AI was taking their work, but also that it devalues the work they do still carry out,” the FT journalist noted.
Marketing content is particularly well-suited to AI automation, noted a separate 2023 Financial Times feature. It enables marketing materials to be “generated at scale in a tone that maintains brand and message consistency. These can also be easily translated into numerous languages.”
One in three American business owners surveyed by Forbes Advisor in 2023 planned to use ChatGPT to write website content, while 44% planned to use it to write content in other languages. Gartner was predicting in early 2023 that 30% of outbound marketing messages from large organisations would be synthetically generated by AI by 2025.
The reality for content copywriters and other marketers though may be less apocalyptic.
AI imperfections
AI is not without its problems. Just ask New York lawyers Steven Schwartz and Peter LoDuca.
In perhaps the most famous AI gaffe to date, Schwartz and LoDuca were lampooned, and fined $5,000, for submitting a legal brief that included six fictitious case citations and opinions generated by ChatGPT.
Gen AI’s tendency to produce inaccurate or false information – so-called “hallucinations” – creates serious trust questions, leaving users and information consumers unsure of what is true. The technology is unable to tell the difference between fact and fiction. Rather, it creates content by autocompletion, using likely word sequences to build sentences based on statistical plausibility, not factual accuracy.
To highlight the hallucination issue, FT journalist and technology commentator John Thornhill asked Microsoft Bing’s Copilot to tell him the world record for crossing the English Channel on foot. Christof Wandratsch of Germany, who can apparently walk on water, was said to have completed the crossing in 14 hours and 51 minutes on 14 August 2020.
“Handily, it even provided a citation for this fact,” Thornhill noted. “Unfortunately, the reference turned out to be an article posted last year highlighting the hallucinations generated by ChatGPT.”
Quality depends on the data
Which points to a second problem: AI is only as good as the data it’s trained on.
Large language models (LLMs) can’t make judgement calls. If they’re trained on inaccurate, outdated, biased or toxic datasets they will spew out similarly flawed content. And unless you’ve trained your LLMs on your own carefully curated data, which is hugely expensive and impractical for all but the largest corporations, how will you know inaccuracies and biases haven’t crept in?
Worse, it risks creating a vicious cycle, where misinformation populates future datasets and permanently contaminates them, as in the English Channel walking on water example. Given the rate and scale of content creation, checking and expunging such errors will be impossible.
Intellectual property and copyright questions are a further concern. The use of IP-protected data to train AI models has created a legal minefield. As Baker & McKenzie lawyers wrote for the 2024 World Economic Forum annual meeting, does training generative AI models on copyright-protected third-party content constitute copyright infringement, exposing developers to related claims?
In addition, noted Deloitte’s IP Advisory team, where Gen AI produces “outputs that incorporate or are otherwise derivative of protected works,” will the model users – be they company employees or freelance workers – face legal repercussions if the outputs are used in commercial work?
AI still needs humans
AI programs are improving rapidly and researchers are working to reduce hallucinations. John Thornhill at the FT pointed to a Google DeepMind paper that proposed a new search-augmented factuality evaluator (Safe) methodology. This breaks down AI-generated responses into constituent sentences and cross-checks them against Google Search for factual correctness.
“In the next few years we will be able to fact check the output of large language models to some good accuracy,” one of the paper’s authors told Thornhill.
But we are not there yet … and may never be. In the meantime, the potential for errors and biases in AI-generated output – and the reputational and legal risks they bring – means human oversight and quality control over the content (fact-checking, editing, proofing, even re-writing to avoid plagiarism charges) are essential.
The need for human validation though may eliminate the initial efficiency benefits gained. As Gartner observed, “end users should be realistic about the value they are looking to achieve”.
Whether AI is even sustainable, given the vast computing power and associated electricity and water it consumes, is a whole other matter.
Value of AI-generated content marketing
Accuracy and logical sense should be the minimum readers expect from the content they consume.
Above that, content marketers should be striving to inform and entertain. Here again AI risks falling short.
In my experimentations with Gen AI, I’ve found the output can help provide the foundations of a blog/article, but the text is dull and repetitive. That base needs considerable work to build out the messaging, juice up the language and improve the flow. Unedited AI-generated content may suffice for clickbait or to pad out SEO strategies. Anyone who wants to seriously engage their audience though needs to go further.
This is particularly true for thought leadership content that has any aspirations to add value. Thought leadership is hardly thought leadership if it simply uses AI to regurgitate content already out in the public domain. Thought leadership needs an edge. It should provide considered (ideally novel) insights from subject matter experts, structured and expressed clearly, and targeted at a defined audience to land with maximum effect. That, to me, requires human savoir-faire.
Gen AI, in short, can be a useful writing aid in content marketing. But it cannot replace the value that a trained, thoughtful creator can provide. For now, at least.
Of course, as a content copywriter I would say that! What do you think?