Guide to AI for Publishers

We’ve recently published a guide to working with AI, specifically LLMs, for publishers. In this post we explore the content of the guide and some of the key take aways for anyone looking to use AI in their organisation.

Guide to AI for Publishers

LLMs can aid content generation and enhance various aspects of the publishing workflow but there are a number of considerations and limitations associated with using LLMs in this context. Almost every industry is currently buzzing with the potential uplift in productivity that LLMs can offer. At Baytree, we see huge opportunity for LLMs to improve the output of anyone involved in digital publishing, with potential for the technology to replace some of the more time-consuming aspects of publishing and support editors and journalists in their day-to-day tasks.

What does current adoption look like?

27% were currently using LLMs in a limited ad hoc fashion without any guidelines or defined process in place for their application.

67% were not using LLMs but intend to do so in the future.

19% will not use LLMs based on an artistic belief that they devalue their editorial output and or pose a threat to their publication from a regulatory or compliance standpoint.

  • How they are are using LLMs

  • CRM content tagging and management (42%)

  • Article generation (53%)

  • Email and newsletters content creation (37%)

  • Audio and video management (31%)

How can and how should publishers be working with LLMs? There are a number of areas where LLMs can aid publishers.

  • Changing Style and Tone of Articles

  • Headline and Metadata Generation

  • Topic Research

  • Statistical Analysis

  • Use of LLMs to Suggest Next Best Articles (recirculation)

Our guide covers these topics in more detail but one important point to note is that depending on the application, it’s important to see the output of an LLM as providing “inspiration”. Like Watson’s insights to Sherlock Holmes, the results of an LLM should be viewed as an aide to spark further investigation. They will most often require editorial oversight.

So what are the key takeaways?

  • You can use generative AI now in your organisation and a 1/3rd of the organisations we surveyed are already employing it.

  • There is no reason that you can’t be using generative AI at least as a ‘topic expert’ in your organisation. Think of this as an extra research assistant that your staff can go to to when an unfamiliar topic crops up and they need quick info on it.

  • OpenAI is the best provider in the market right now and has the lowest barriers to entry to use.

  • Ensure you are ready to leverage the power of LLMs by using a headless CMS that will allow you to “slot” in AI features as and when you need to. Monolithic legacy CMSs are extremely badly positioned for this trend; as they require your vendor to implement support for all this and are not as modular. 

  • Big opportunities are coming, with increased context windows allowing much more data to be analysed at once, and GPT4 allows interesting statistical analysis. Expect this to continue to improve. 

  • Open source LLMs potentially solve a lot of regulatory issues. If this is holding you back from using generative AI in your organisation, it is imperative that your technical leadership team are on top of these developments as the performance of the open source solutions have recently reached a tipping point of usability.