ChatGPT and other large language models (LLMs) have now been around for a little while. Most of us have now experimented with at least one. So how can you use these models, collectively part of the generative AI family, to support thought leadership?
There are two parts to that. First, should you use it to write your articles? Second, can it help you to think through your ideas?
ChatGPT as a writing tool
ChatGPT is very good at writing. It takes information, and it turns it into a coherent form. However, it doesn’t know anything. If you ask it to write on a particular topic, you will get a fairly superficial take on existing information, which is probably not worth publishing. However, there are other ways.
Ethan Mollick, an academic at the Wharton School in Pennsylvania, asked ChatGPT some questions, and prompted it to improve on what it had written. He included the paragraphs it produced in his article, and it’s hard to tell the difference from his writing. Others have tried giving ChatGPT some bullet points to use as the basis for an article. They found that with a bit of editing, the article was publishable. This is therefore worth trying—but reviewing and editing is crucial.
ChatGPT as a ‘thought partner’
The other question is whether ChatGPT can help you develop your thoughts. We tried this, to see if it could help us come up with a series of blog posts. Here’s what we learned.
-
Treat the process a bit like a conversation
If you ask ChatGPT to suggest a series of blogs on a particular topic, it will do that. However, that may not be the best list. As with brainstorming, it’s worth thinking around the topic. Use questions like “How would you persuade…?”, and “What factors are most important to stakeholders…?”. Some of the alleys you go down in this process may turn out to be dead ends—but that’s OK. Doing this will give you the broad ‘state of understanding’ of the field, and help you to target the level of your blog posts more accurately.
-
Give ChatGPT as much context as possible
One expert on AI has described LLMs as being like very bright but naïve interns. This feels about right. When you give your intern more information about what you want, you get better results. With an LLM, more specific prompts will give more specific results. You can also dig down into the initial results to get more information on the bits that interest you most by asking questions like ‘Tell me more about x’.
-
Language matters
When asking multiple questions, you can use phrases like ‘from your first list’, or ‘from the list above’ instead of setting out your requirements again. However, you get better results if you ask it to tell you more about ‘topic x from the first list’ than ‘#6 on the first list’. It is also better to avoid ‘negatives’ such as ‘tell me about x but not y’, because the model seems to appreciate that ‘x’ and ‘y’ are important, and tells you about both.
-
Case studies and examples are probably best treated as a starting point for research
ChatGPT will give you examples if you want. However, they are fairly limited. Most of the examples it provided had been widely documented in the news. Others were broadly consistent with the company’s reputation. However, on digging deeper using a standard search engine, it was hard to find a source for the information. The ChatGPT site contains a disclaimer about the accuracy of information about people, companies and events. It is therefore probably best to fact-check any examples and find your own source.
-
Don’t underestimate how long the process will take
The process of coming up with a series of blog posts using ChatGPT is probably shorter than brainstorming it by yourself or with a colleague. The outcome is also probably better judging by our test. However, don’t underestimate the time required. It isn’t a five-minute job, and still needs thought.
-
You are the final arbiter of what you write—and original thinking matters
We are used to thinking of computers as infallible. However, that’s definitely not true of ChatGPT. It can ‘hallucinate’ or make things up, including sources and references. Remember that you are the final judge of what will work for your audience. Think of ideas from ChatGPT as a starting point for your own thinking, and work from there—and crucially, fact-check everything.