We recommend that brands carefully consider where, when and how to integrate generative AI into their content strategies, creation processes and management. The optimal approach will differ depending on brand, industry and organizational culture and priorities.
The risks of both extremes
There are two sides of the coin when it comes to either going all-in or completely avoiding the use of generative AI for content creation and operations. The risks of going all-in include:
- Content produced with generative AI can’t be copyrighted, and it technically violates the terms of usage of ChatGPT to publish content generated by the platform as-is.
- LLMs need human-generated content to maintain their functionality; in fact, studies show LLMs degrade significantly when trained on AI-generated content. So, as brands begin flooding the internet with content that’s been partially or completely created with gen AI tools, and the off-the-shelf LLMs use this data for their outputs, it could theoretically create a vicious cycle in the erosion of content quality and usability.
- LLMs are trained on huge volumes of content, so their output is essentially the most generic version of that content. But brands need content with a unique voice and point of view. Carefully crafting the prompts entered into the LLM (sometimes known as prompt crafting or prompt hacking) can help, but only so much.
Meanwhile, there are also risks to continuing with business as usual:
- An Adobe survey of marketing leaders predicted that demands for content will increase between five and 20 times by 2025. Meeting these demands with little to no automation will prove extremely costly.
- Gartner reports that 75% of CMOs are [already] facing pressure to ‘do more with less’ to deliver profitable growth in 2023.
- With many brands already embracing generative AI, competitor brands are likely to significantly outpace non-adopters in content publication velocity.
The path forward
Clearly, neither optional is tenable. Instead, brands need to determine their best course of action based on their own characteristics and the generative AI adoption modes available to them. As we recently described, businesses have two options for creating content with generative AI:
- Off-the-shelf tools such as ChatGPT and Google Bard, or SaaS tools, such as Writer. These tools have lower adoption and implementation costs but will require continuous and careful attention to crafting exactly the right prompts. And, while Writer will learn your brand and voice over time, ChatGPT and Bard, in their off-the-shelf formats, will always be focused on generic responses. For this reason, off-the-shelf solutions can be considered good enough for smaller companies with limited content needs.
- Custom-trained AI content generation tools. Essentially, this means building (or working with an implementation partner to build) your own generative AI platform on the foundation of an existing LLM. While this platform would require more upfront investment and development, it would be better engineered to draft content with your unique POV, and brand woven into every output.
Custom-trained tools are likely the better option for larger companies with vast content needs across business units.
Ensuring the right foundation
For companies that choose to build a custom solution, it’s essential to have a strong foundation of existing content for the model to train on. This content should be unique to the brand and offer a unique POV.
The content should also, ideally, be built in structured formats with robust metadata tagged to each atomic slice of content (essentially snippets of content that can be deployed independently into various content assets, formats and channels). The structure and metadata help the AI model better understand the content and its context.
Businesses already have this level of structure and metadata if they use a “headless,” CMS (one that acts as a content repository usable by a variety of front-end tools) or takes a modularized content approach to “create once, publish everywhere” with dynamic, defined business logic.