In communications, we have over the last year flogged the promise of AI in a thousand press releases and read predictions from a thousand pundits.
But the time of those vague platitudes – ’AI has the potential to completely change the economy’ or ‘We are completely unprepared for the radical shift to working alongside AI’ – is rapidly coming to a close. Instead where we are headed is a demand for what can be done short-term with AI, and how specifically this can align with current challenges.
When we at Cognito surveyed 100 professionals globally in financial services and technology, we frequently heard that there’s a pressing need now to understand how advancements in artificial intelligence will lead to actual business transformation. These are starting to emerge – BT’s plans to layoff thousands as it automates customer service, and education service provider Chubb’s stock crash after it became apparent ChatGPT threatens its core business model.
We also heard from executives that the biggest opportunities for artificial intelligence in marketing and communications is by increasing the amount of content that the marketing function can create. That’s great – but are we entering an era where we sacrifice quality for quantity? I decided to dive and see what’s possible with today’s tools.
Experiments
I tried to see if I could coax something a bit more distinctive out of a chatbot. Using the free version of ChatGPT (3.5) I ask the text to come up with a piece exploring the “challenges of implementing ESG data metrics into investment processes, with a particular focus on the United Kingdom.”
Here’s an example sentence from the result: “While the adoption of ESG data metrics brings numerous benefits, it also presents unique challenges. This article explores the hurdles associated with implementing ESG data metrics into investment processes, with a particular focus on the United Kingdom.”
To be frank, this seems more book report than an article. The writing level is reasonable – for a teenager. It’s nowhere near what I would think would work professionally.
I then tried to mold the text using prompts. I pushed the bot in a particular direction by asking it to act as a journalist at the Financial Times, writing a Big Read. I encouraged it to adopt a professional yet engaging style, and encouraged the use of several keywords and unexpected wordplay. Here’s what I got:
“One key obstacle standing in the path of seamless ESG integration is the maze of disclosure frameworks and reporting standards. As the United Kingdom forges its own regulatory path post-Brexit, market players find themselves confronted with a diverse range of reporting initiatives, such as the EU’s Sustainable Finance Disclosure Regulation (SFDR) and its British counterpart. This intricate web of standards, while aimed at enhancing transparency and comparability, can create confusion and hinder efficient evaluation of ESG performance.”
The first sentence is OK but it also doesn’t really reflect the level of analysis, or tone that you actually get from a piece of exemplary reporting. Here’s an example from an actual Big Read from earlier this year:
“Christine Lagarde’s husband has asked her to stop changing jobs, the European Central Bank president likes to joke, because each time she begins a new role a major crisis seems to follow. Soon after becoming French finance minister in 2007, she found herself handling the global financial crisis for which she eventually won plaudits.”
Finally, I spent a series of time trying to get the bot to try and approximate the style of a famous writer. I choose John McPhee, a longtime New Yorker correspondent and a personal hero. This really didn’t work when asking the bot to revise existing material, but things got much better only when I allowed the bot to “tell a story”. Here’s what it came back with:
Amidst the muffled hum of traders, a seasoned investor named James, with salt-and-pepper hair and a weathered countenance, leaned forward. He began to recount an encounter that would forever change his perspective on the intersection of finance and sustainability.
“I found myself in the rolling hills of Yorkshire, engulfed by the serene beauty of an idyllic countryside. It was there, on the grounds of a centuries-old estate, that I met Jonathan, a young farmer who had recently embraced sustainable farming practices,” James recalled, his voice tinged with intrigue.
Little did James know that his voyage into the realm of ESG would lead him down a winding path, strewn with challenges and contradictions. From grappling with disparate reporting frameworks to deciphering the reliability of data, he would embark on a journey that would test his resilience, knowledge, and beliefs.
That’s really not bad. But what to do with it? James, salt-and-pepper hair and all, is completely fictional. It seems in order to get AI to have a personality, you have to give it free reign to depart from the facts.
A discussion on grammar tools
A second area where style intersects with writing is through grammar editing tools. A decade ago these only flagged spelling mistakes. Then, through tools such as Grammarly and built in tools in Microsoft Word, tools started to identify errors in agreement, flag missing words and incorrect prepositions. These are now integrated into apps and writing tools, and transcend what I would call hard rules into areas of judgment. GrammlyGo doesn’t just shorten sentences but will offer to write the entire email, document, in different tones. Expect all editing tools to offer something like this by the end of the year.
Constantly follow the advice of grammar bots and you wind up with clipped, short sentences that are always in the active voice. In general, this makes sense. But I worry we are reaching a place where experimentation is always underlined, where reaching for a slightly uncommon vocabulary word or phrase is seen as a mistake.
I believe we need to use these tools in much the same role as editors. Sam Barber and Andrew Marshall, my erudite colleagues in New York, have made many suggestions on this draft. Many I accepted but I retained the ability to reject something that while a bit of a flair, was critical to the way I wanted to construct my argument. We should make people writing material understand this, and offer training on their effective use.
Cybernetic editing
AI offers the opportunity to reverse pressures around content creation. This will be some sort of ‘hybrid model’, where human and machine work together, with a firm need for some sort of experts. Any material that’s been generated in part with AI – down to the phrase level – needs to be re-worked and analyzed by a person.
The idea that ChatGPT will produce first drafts quicker and then allow us to spend more time on revision is tempting, but I think we need to be prepared to input quite a bit to get that draft. Perhaps we can enter our notes or a series of key phrases into the model, so we can incorporate our original ideas into the text.
There’s also the possibility of using AI to ‘optimising’ content as part of post-production for specific channels, particularly for web. Once you’ve written an article, ChatGPT can easily generate the meta title and description for SEO. Here’s what it did for this article:
Meta Title: “Unlocking the Power of Artificial Intelligence: Solving Style Challenges in Communications”
Meta Description: Discover how artificial intelligence is revolutionizing the way we approach style in communications. Learn about the potential for business transformation, the impact on marketing and content creation, and the role of grammar editing tools. Find out how AI can optimize your content and delve into the future of thought-provoking and engaging writing.
I worry that we are entering an era of flat, workmanlike prose designed simply for browsing and speed reading. We need to ensure that there is some friction, ideas and phrases that make people stop and think. This means ensuring we actually learn things by reading instead of simply endlessly consuming more and more pieces of writing.
Jon Schubin is a director in Cognito’s London office