6 ways businesses can use AI to create content more productively

Not long ago we explored how AI has the power to boost translation productivity. But what about content creation – specifically, the process of writing? Large language models (LLMs) such as OpenAI's GPT, Google's Gemini (used in Bard), Meta's Llama and others have wowed us with their fluency and ability to reflect different styles of writing. Given how quickly they can operate, the question naturally arises: can they help businesses create content more productively? 

Before answering this question, note that I won’t be addressing the more contentious question: can AI replace content creators altogether? Right now, I think the answer is ‘no’. I won’t say ‘never’, but for now the generative AI models (i.e., LLMs) most suited to creating content are also most vulnerable to hallucinations and biases that are difficult to mitigate. They’re also always in danger of losing relevance as they fall behind evolving human knowledge and values, unless continually retrained or designed for continuous learning (neither of which is simple to achieve). 

So for now, let’s look at AI as a tool to help human content creators be more productive. Let’s assume that we’re talking about using secure AI models that can be trusted with your content, and that you’re originating content in English (because if you’re not – and especially if you’re using a low-resource language – it’ll be harder to get good results). Here are six ways AI can help.

If you’re struggling to come up with an idea for a title or punchy introduction, or to organize your thoughts for a piece of content through an outline, an LLM is an easy way to generate something to start with. Even if it’s not perfect – and possibly not very original – it will spark additional ideas and hopefully help you get to your solution more quickly. And if all you’re looking for is inspiration rather than a substantive piece of work, you’ll likely get it without having to put a lot of effort into formulating the perfect prompt.

You can use the LLM to include a specific keyword in your title, intro or outline. This does assume that you’ve done SEO research to validate the keyword you’re using, because an LLM isn’t an SEO research tool. 

LLMs can also help you cover every angle of your topic to boost the relevance of your content. Two ways to do this are to prompt it to: 

• Find topical gaps in an outline 
• Generate a list of popular questions related to a keyword, that would interest readers of your topic

Even in organizations with in-house content teams, most people involved in content creation aren’t professional writers. They may be topic experts who struggle to express themselves clearly for a non-expert audience and are inexperienced in applying brand or writing guidelines. They may not have a talent for spotting inconsistencies. Or they may not be native English speakers. 

AI-powered editing tools – hopefully integrated with the application used for writing – can go well beyond basic spelling and grammar checks to give these content creators a helping hand. For example, you could ask the right AI tool to rewrite content to improve its readability for a specific audience. Provided with the right training or input, the right tool could also help with brand consistency by suggesting preferred terminology or the appropriate style and tone. 

But remember that the AI model doesn’t understand the content. If it’s a complex or technical topic, or the content is poorly structured or the language unclear, then the tool’s ‘simplification’ or ‘improvement’ might not accurately reflect the original intended meaning. It might even make a grammatical error. So it’s always best for authors – or a professional writer or proofreader, depending on the project – to check any AI output carefully. If the author is a non-native English speaker, they might actually get better results by writing in their own language and using a quality machine translation solution.

Similar considerations apply when using an LLM to create a different version of content, such as summarizing a longer piece or creating a derivative asset from a primary asset (an email or social post about an article or report, for example). 

Here your results may depend not only on the quality of the original content, but also on the quality of the prompt. If you want a specific tone or style, you need to include that in the prompt. Same for any word or character limits, or the target audience. If you want your summary to include a set of short takeaway bullets – no more than five – then say so. A good prompt reflects a fine balance between being specific enough to give the AI model the context it needs to achieve your objective, and short and clear enough not to send it off track. 

It may take a fair bit of experimenting to get to an effective prompt. This can be frustrating if it takes longer than just creating the new version yourself, but over time you should get better at it and see results faster. The aim isn’t usually to get the AI model to generate something perfect. It just needs to be good enough to save time for the author overall.

So what about getting AI to generate content from scratch, such as a standard report from a set of data (turning weather data into a forecast, for example) or a case study from interview notes? 

Here the potential time-saving is much larger than for generating short derivative pieces from existing content – but so are the chances of it going badly. You’ll typically need to put a lot more effort into prompt engineering to get good results, and you’ll want to give the AI model good examples of what you’re looking for. 

While there’s nothing to stop you asking AI to write a script for the creative, emotive advert that you hope will win awards, it’s not the most obvious use case (unless used for inspiration). The biggest potential for AI content origination comes from content types that are: 

• Formulaic or heavily templated, where you’re looking for consistency more than originality and there’s benefit in having the AI model reproduce content that is similar or even identical to previous examples. 
• Produced often enough that the effort to get the prompt right will see a good return on this investment over time. 

Again, the intention isn’t to absolve humans from responsibility for the finished article – both its accuracy and its ability to meet the objectives of the content – but to save them time so that they can focus their expertise and creativity where it’s most valuable.

Most content creators will know what it feels like to have written something in the past that could be reused for their current task – if only they could remember when and what it was. Or they may suspect that someone, somewhere in their organization has done something like this before, but they can’t be sure and don’t have the time to try find out. 

Finding content to reuse is just one example of a more general information findability problem faced by organizations. Organizations have been solving this problem with the help of different types of AI, but the advent of LLM-powered chatbots can simplify findability for people even further. Retrieval-augmented generation (RAG) is an approach that combines enterprise search with the power of LLMs, potentially even allowing for an integrated AI model to prompt authors as they write with relevant information found in the organization’s repositories. 

Remember, though that ‘garbage in, garbage out’ matters to any use of AI. LLMs can’t magically make sense of information repositories that are full of out-of-date or inaccurate information, or that contain different versions of content with no indication of why they exist. The fundamentals of good information and knowledge management remain critical, which is why organizations continue to pursue a structured content management strategy that helps them eliminate content duplication and optimize content reuse for different purposes. 

For organizations using structured content, AI has two further ways to help improve content reuse. One is through smart tagging, where a semantic AI model suggests tags to authors to help organizations categorize their content consistently, thereby further improving findability. 

The other is to use generative AI to convert unstructured content into structured content, which could save a lot of time if you’re just setting out on your structured content journey or want to expand its scope. The idea is that you’d feed the LLM a Word document (for example), and it would return a fully-tagged XML document back to you. As with so many uses for LLMs, though, you’d need to put in the work training and prompting the AI to get reliable results, and check the results for accuracy.

It will take a lot of experimenting for businesses to settle on the best AI tools and approaches for the huge variety of content they need to create. Word users can try Microsoft Copilot. There are writing-focused AI platforms such as Jasper or Inkform. Or perhaps ChatGPT is good for some of your needs. 

If you’re using Tridion Docs to author structured content or Tridion Sites for web pages, you’ll already be familiar with smart tagging, among the platform’s other semantic AI features. As we continue to develop Tridion’s AI capabilities to improve content reuse and findability, we’re also actively exploring whether integrating generative AI into the platform can meaningfully help our users. We’ve already developed two use cases, one using retrieval-augmented generation to improve information discovery, for instance for customer support (so not aimed at content creators) and the other to help authors and editors make content more readable. Both are planned for an upcoming release of Tridion – evidence of our continuing commitment to an ambitious AI roadmap designed to bring genuine value to our clients.