Like many who create content for a living, I’ve experimented with GPT-4 to see what it can do. Part of this exploration is for self-preservation. (Will AI steal my job helping executives develop business articles?) The other is curiosity. (How might it save us time and effort?)
I’ve found that GPT-4 can make content creation easier and quicker, and I’ll describe how below. But it cannot compose anything original that adds to our collective knowledge for two main reasons.
First, generative AI (GenAI) tools can only generate content based on what others have written, and busy execs aren’t interested in advice they’ve already seen elsewhere. Producing unique insights on a topic still requires original research or an expert practitioner.
Second, GenAI tools don’t do a great job of generating content based on what others have written. Douglas Hofstadter, the author of Gödel, Escher, Bach, has a story in The Atlantic about a GPT-4-generated description of why he wrote the book that’s wrong, despite the detailed description at the beginning of the book’s 20th-anniversary edition.
That said, there are some things GenAI can do to help make creating B2B content easier and less time-consuming. The most useful I have found so far are:
Summarize a transcript. Interview and panel transcripts run 8,000 to 10,000 words for an hour of conversation. That’s a lot of words to wade through to produce a 2,000-word article. GPT-4 (equipped with a plug-in such as AskYourPDF) can summarize the salient points and trim out the less important stuff. It tends to lose interesting details, but if you ask it to give you all the content relating to a particular point, it will pull all the relevant sentences and tell you the page they’re on so you can decide what to include. This can all save a lot of time.
Analyze technical documents. For a recent research study, we had to scour FDA review documents for new drug submissions. These can be 150 pages long and are inconsistently structured, so the answer to questions such as “What PROs (Patient-Reported Outcomes) did the sponsor use?” can take one or two hours to find. A typical GPT-4 answer was “The sponsor used the following PRO instruments: PedsQL (Pediatric Quality of Life Inventory) and EQ-5D-3L (EuroQol-5D-3L). These were used to assess health-related quality of life (HRQoL) [Page 61]. GPT-4 was stunningly complete and correct in doing this.
Draft routine copy. If you want an article that, for instance, summarizes the key issues in nonprofit marketing, as a friend recently did, GPT-4 will write a perfectly good one. It won’t say anything new or rise to the top of search results. But if, as in my friend’s case, you need to show you are in that market and can help visitors get up to speed quickly, GPT-4 does an acceptable job quickly for little effort.
Write headlines. I asked GPT-4 to write click-attracting headlines for an article about six things a company can do to bring patient care into the home. Here are two of them:
-
- “Transforming Healthcare: Discover the 6 Key Strategies for In-Home Patient Care”
- “From Hospital to Home: Mastering the 6 Vital Techniques for Exceptional Patient Care”
They are a little more melodramatic than our usual style, but I asked for click-attracting, so they are good starting points that could be winners with a little editing.
Create social media posts. I had GPT-4 write some social media posts for a recent article. They, too, were a bit breathless the first time around, but it toned them down when I asked it to. Here’s the result for Twitter, for example: “Discover the future of healthcare with remote patient care and hospital-at-home programs. Learn how health systems leverage digital solutions to deliver convenient, high-quality care. #RemoteCare #HealthcareInnovation [Link to the article].” It needed some tweaking, but it’s on-topic and perfectly usable.
GenAI machines are word manipulators. GPT-4, which I’ve worked with is excellent if the text it’s given is bounded. It excels at analyzing complex documents to which it’s constrained. The minute you allow it to access its vast database of ingested knowledge, you lose control, it can wander anywhere it wants, and the output can be banal and wrong.
GenAI cannot produce original, insightful content based on the unique experiences and opinions of people who know things the internet doesn’t. I’d never use it to generate final copy. On the other hand, if you use GenAI to support content development, you’ll be more efficient and produce better content than those who ignore or dismiss it.
