Using AI for thought leadership writing: how to complement - not replace - your creativity
More and more, I’m being asked by potential new clients: “Do you use AI?”
The answer is yes - I think AI can be a powerful way to enhance my productivity. But having taught myself a lot about AI - and tested various tools extensively - I can see the huge problems that you can run into if you don’t use them correctly.
And it’s clear that there are too many who are using them in a way that leads to mediocre content - or worse. This is a problem among both freelancers and in-house marketers. But I think it’s particularly problematic for freelancers selling content-related services to use these tools badly, or fail to be transparent with their clients about how they’re using them.
So, I want to be transparent and share how I use these tools as an aid to create high-quality, original content. As well as share advice that can help other content marketers get the best out of AI.
I’ve written a detailed article that explains how - and how not - to use AI for original content that achieves thought leadership.
The biggest AI content creation problems
Hallucinations
“Hallucinations” - in other words, when AI makes stuff up - are a huge problem. One study indicates that around 46.4% of AI-generated texts have factual errors, while Vectara’s Hallucination Leaderboard points to 1 to 30%. Whatever the true number, it’s undeniably a very common issue.
Fortunately, most marketers are well aware of this issue. According to a Salesforce study, B2B marketers say accuracy and quality are their top concerns with generative AI.
As AI models advance, hallucinations should become less frequent - but it’s not clear how long this will take, or if it will ever be solved altogether. Recently, OpenAI claimed that GPT 4 was 40% more likely to produce accurate responses compared to GPT 3.5 – but research then showed that the later model is more likely to spread damaging misinformation. So we have some reason to be skeptical about whether these models are really becoming more factually accurate.
This problem is made worse by the fact that AI tools sound so confident in their answers. They can produce plausible-sounding nonsense, which can throw even the best critical thinkers among us.
For this reason (and others that I’ll explain later), it’s a bad idea to use generative AI to create whole chunks of text from scratch. Sure, you can go through and fact-check AI-generated text. But this is a time-consuming process (when the point of AI is to save us time), and still runs the risk that some inaccuracies will slip through the net.
I’d go as far as to say that fact-checking AI is a completely different task from fact-checking a human writer. While humans make mistakes when we write, the pros among us don’t just make stuff up on the fly. AI, on the other hand, will state something totally made up with complete confidence, and it requires a lot of focus by the editor to spot this.
2. Robotic tone and ChatGPTisms
At some point, after reading enough AI-generated content on LinkedIn and elsewhere, I started to feel like I was hearing the same, robotic voice over and over again.
When you use generative AI to generate text from scratch - without following best practices that I’ll describe below - something about it just feels off. It has a tendency towards overly enthusiastic or cheesy copy that feels inauthentic (as well as a funny obsession with 🚀 emojis).
There are even certain words and phrases that AI tools seem to use repeatedly, which I like to call ChatGPTisms. “In the ever/rapidly-evolving landscape” is one particularly common example.
As with hallucinations, this is something that can be partly fixed with better prompting, as well as editing AI writing. But even the best AI-produced copy doesn’t have personality - this is difficult to imitate, it turns out. It also can’t convey a strong sense of relating to your target audience. The best writers can get into the minds of their audience and write in a way that really speaks to them, but as AI doesn’t have empathy, it can’t do this.
A big problem I see too is that not everyone has the ability (or the time) to spot ChatGPTisms and tailor the tone of voice sufficiently. I’ve seen experienced (and otherwise skilled) marketing and comms professionals post content that’s clearly written by AI. This suggests to me that taking AI-generated content and making it feel human and compelling is easier said than done.
Spotting this problem of robotic language can become even harder when you rely too heavily on AI. I think there’s a big danger that when you delegate too much to AI and stop flexing your writing muscles, you lose both the ability to recognise truly good work and to create it.
Below is an example of the kind of copy AI tends to generate. I asked ChatGPT to generate this LinkedIn post from scratch, but unfortunately, it’s full of buzzwords, generic, and even has the telltale 🚀 emojis.
I had AI create this LinkedIn post, and it’s packed with buzzwords and generic language.
3. Lack of originality
Perhaps the biggest problem with using AI to generate whole pieces of content from scratch (rather than as a helpful tool) is that this content is, by definition, not original. Let’s remember: generative AI models predict the statistically most likely combination of words in response to your prompt. This is an impressive tech capability, but it obviously can’t constitute anything like real thinking or expert insights.
There’s so much content already out there competing for your audience’s attention. You don’t need to add to the noise by publishing something generic pasted straight out of ChatGPT. Rather, you want to create content that demonstrates authority, expertise and personality. You do this by:
Interviewing subject-matter experts and shaping their insights into interesting content.
Using original data/research to add something new to the conversation.
Talking to your customers and crafting content that’s genuinely useful for them to solve their problems; content that they can’t find elsewhere.
AI can be a useful tool to help you do this better and more efficiently. But you can’t use it to replace real thinking and insights. After all, if you can’t be bothered to write it, why should anyone be bothered to read it?
AI analyses the context and calculates probabilities for each possible next word, selecting the most likely option. This is a very useful tech capability, but is also clearly nothing like original human thinking.
Effective ways to use AI as a productivity tool, not a creativity replacement
Some writers might take issue with me saying this, but I don’t think that all aspects of writing are particularly creative or interesting.
AI can help here, speeding up more mundane writing tasks like summarising information and checking grammar. This frees up time for writers to focus on more creative and strategic aspects of writing. For example, crafting a distinctive point of view and story, strengthening our arguments, and improving the flow of the piece.
Summarising long articles
Taking large amounts of text and creating neat summaries is something that AI is very skilled at. Doing this can save writers a lot of time and improve accuracy, as it allows us to spend longer on fact-checking.
For example, let’s say that you’re looking to summarise a relevant news event in your content. You could share the link with your generative AI tool, using a prompt like “Summarise this information in 50 words or less, aiming for clear and concise language. Make sure to include the most critical details, and if you leave any key details out, explain why you have done so”.
As with anything AI does, you’ll want to cast a critical eye and check it. But from my experience, AI tools are very accurate when it comes to summarising information. This is because they are applying quite simple logic (summarise, shorten) to a small dataset you’re inputting in the form of the text to be summarised, rather than generating responses from their huge training dataset.
2. Analysing transcripts and extracting quotes
AI’s summarising capabilities are useful for helping to extract and categorise key information and quotes from calls with subject-matter experts. To do this, provide the LLM with the call transcript and a prompt that clearly explains:
The topics and objectives of the content you’re writing
The context of the call (who you’re speaking to, what the conversation is about)
Who you’re writing for
Provided your prompt has these things, it can do a great job at organising the call content into key themes and pulling out quotes that are good at getting key messages across. And tools like Fathom and Otter.ai have AI integrated into their workflows, making it really easy to summarise information from calls.
Of course, you’ll still want to review the call recording and transcript yourself (particularly if you weren’t present on the call) to check for hallucinations. And as ever, you should cast a critical eye on what AI says are the most interesting points. Often, the way AI summarises the call information feels formulaic and generic. So the skill of the human writer lies in taking this basic summary and crafting it into a more captivating story.
3. Reformatting/repurposing content for other channels
Again, the ability of AI to summarise and adapt information can be useful to content marketers who want to distribute content across multiple channels (as they absolutely should be doing).
For example, let’s say you’ve created a blog post packed with subject-matter expert insights. You could then feed this blog post into an AI tool, asking it to break the article down into 3 LinkedIn posts, a Thought Leadership Ad and an email newsletter. I’ve found that when you feed AI the right raw material in this way, it’s quite good at repurposing it for different channels.
Outcomes will vary and you’ll want to bring some sharp editing focus to make sure that the repurposed content meets your high standards. But overall, this is a very good use case for AI and one that can help marketers nail the crucial distribution aspect of content marketing.
4. Checking grammar and improve readability
Grammar checking and language refining are by far the most common ways that I use AI in my writing. Compared to the autocorrect and grammar checkers of only a few years ago, new AI-powered tools like Grammarly and ProWritingAid do a fantastic job of picking up things like missing words and subtle grammatical errors.
They’re great at spotting the kind of mistakes that you can so easily miss when self-editing, even when passing over your text for the second or third time. They also have features that will rewrite sections of text according to instructions like “Make it clearer” or “Make it more engaging”. When you’re staring at a sentence for ages, wondering how to get it to look right, this can be very useful.
These tools don’t get it right every time, so you’ll need to take the suggestions with somewhat of a grain of salt. For example, sometimes we might want to use a word like “really” or “actually” when writing copy with a conversational tone, but Grammarly always marks this as an error.
I’d also recommend that you turn these tools off when you’re getting into the flow of writing your first draft. Why? Because it’s a mistake to edit while you write your first draft, as this inhibits you from writing freely and blocks your creativity. Having your AI tool turned on during your first draft encourages this bad habit.
But overall, there’s no doubt that these tools help to make my writing significantly clearer, with fewer errors.
5. Providing ideas for titles, meta descriptions, captions etc.
I do believe that there’s an art to writing great titles, and it’s certainly worth spending time to get right. The right title can greatly increase click-through rates and engagement with your content.
But AI’s summarising skills can be really helpful when it comes to generating title ideas, which you can then refine. You can feed AI the article that you’ve written, and say e.g. “Provide me with ten ideas for titles that give a clear sense of what the article is about, are engaging, and motivate the reader to continue”.
You can then take the ideas it’s provided and hone them as needed. Often none of the titles it provides will be up to scratch, but it can spark your creativity and serve as a useful form of brainstorming. The final result will likely be better for it.
6. Brainstorming - but to refine ideas, rather than as the main source or impetus
AI can help you in the brainstorming process. But I’d see it more as an assistant that helps to refine your existing content ideas. Or expand and repurpose them into new variations.
This is because the ideas that AI generates are unavoidably generic. It’s just going to tell you to create the kind of content that everyone else in your industry creates. And when you take this approach, you become a thought follower, not a thought leader. So you don’t want to rely on it too much to generate new ideas from scratch.
Instead, you want a content strategy that’s centred around:
Insights from your internal experts. For example, a bold take from your CEO about where the market is heading.
Real understanding of your customers, their problems, and how you can speak to them. You get this from actually speaking and listening to customers.
Publishing original/internal data and crafting a strong narrative around it.
Having a distinctive point of view in your content so that your audience understands how you approach an industry problem differently (and better) than your competitors.
This is a content strategy that gets you leading the conversation. It speaks to your customers’ problems before your competitors do. It highlights why your expertise and product are distinctive - and better.
You can use AI as a useful sounding board when refining how to package your expertise. For example, let’s say you have some interesting internal data that shines a new light on an industry problem. You could use your AI tool to provide some suggestions about the angle you might take when telling a story about the data in the report you’re writing.
But original thinking should be the impetus behind your content ideas - not AI.
What are my favourite AI tools?
My favourite AI tools differ from task to task. Many are designed for a specific task, making them easier and better for that job than a general model like Gemini or ChatGPT.
This list reflects my personal opinion and experience on their performance, not rigorous testing, but it should help those at the start of their AI journey.
For brainstorming, summarising, and repurposing - I find that the paid version of Claude performs the best.
For recording and analysing call transcripts - I use fireflies.ai.
For research and searching for sources - I use Perplexity.
For checking grammar and readability - I use ProWritingAid.
Use AI as a tool to aid in creating original, high-quality content - not as a shortcut
As I’ve set out, there are several good use cases - like summarising information and generating title ideas - for AI in writing high-quality B2B content.
If there’s one theme that unites all of these uses, it’s this: AI should be used as a tool to automate mundane processes and compliment your creativity, not as a lazy shortcut. AI can completely automate some basic writing tasks, but more sophisticated writing requires a lot of human involvement.
Some will ignore this advice and use AI to crank out massive amounts of generic content. But this is a shortsighted strategy, and one that will damage the brands that engage in it. Clever B2B marketing teams are realising that quality over quantity is the way to make their brands stand out. Personality, real subject-matter expertise, and deeply understanding your customers are becoming much more important in content marketing.
In other words: AI can automate the kinds of content that we used to write - but this has forced us to get smarter and look for ways to stand out.
No doubt, as AI capabilities continue to evolve, I’ll be updating this article. But I’m confident that even as AI advances, we won’t be able to use it to hack or shortcut our way to content with real originality and value.