This week, we held our ‘AI Special — Q&A’ webinar, in which the fourteen contributors to the AI Special we ran in the November / December issue of InPublishing magazine answered questions about AI.
There were lots of insights and good advice — I have paraphrased one from each of the panelists below:
- Creating entirely new content or data sets to power products is very resource intensive, so using AI to repackage or to reframe existing content into new products and services makes sense. With that in mind, you need to start with a clear understanding of what data and content you already have and then think about how you might layer in information from different content types or external data to add depth, context and new perspectives. (Jennifer Schivas, CEO, 67 Bricks)
- We’re moving from articles being a mainstay of how people consume content and moving much more towards the all interactive chat, question / response paradigm... if you have a directory that’s well maintained and current, then it lends itself really well to answering questions, and it lends itself really well to that kind of interactive chat. (Markus Karlsson, CEO and founder, Affino)
- Using AI images is a way of replacing the kind of pedestrian stock art, stock images and clip art that we used in the past and replacing them with highly arresting, highly engaging life-like images which convey the subject in a way that we simply weren’t able to do before. (Tim Robinson, editorial director and publisher, National World)
- When using off-the-shelf LLMs to auto-tag articles, you run the risk of overly generic tags. The best approach is a combination of fine tuning the AI, training it on domain-specific data, and prioritising the most relevant and impactful tags, using, for instance, whitelists and priority scoring. (Brian Alford, founder and CEO, Bright Sites)
- When speccing out your new ‘AI chat discovery’ tool, a good place to start is to write down sample good conversations. Deciding what you want the desired response to be will help you later evaluate how effective your system is. (Ben Tregenna, chief technology officer, Content Catalyst)
- Productionising is essential. The goal is to integrate AI into your systems so people don’t have to go somewhere else. Copying and pasting into tools like ChatGPT can be super clunky, so connected systems make workflows faster and easier. (Katja Eggert, head of strategic development, Immediate Media)
- Once you’ve decided on the problem you’re trying to solve from a content perspective, that’s where it becomes very interesting and useful to involve AI. If, for instance, you want to make your content more user needs centric, AI can be deployed to help you understand what is the underlying motivation that makes someone read a story. The analysis, when applied across a publishers’ output, can help editors fine tune their content strategy. (Aliya Itzkowitz, manager, FT Strategies)
- One of the things we’ve seen really working well is creating a sort of shadow set of tagging and topic assignments that allows content to have this completely different set of tagging that is not visible to the end customer, but it really informs search. (Stewart Robinson, managing director, Full Fat Things)
- People are moving away from generalist platforms and starting to use specialised agents and specialised tools for specialised jobs. We’re going to see more frameworks to build your own agents and more ways to embed them into your workflows. (Thomas Lake, director of product and technology, Infopro Digital)
- When using automation to edit and enhance images, where it gets really interesting is in the area of cut-out processing. A job which used to take hours now takes maybe 10 to 15 seconds for a very detailed cut-out. Scale that up across a publisher’s output, and the time saving is substantial. (Derek Milne, commercial pixometrist, Pixometry)
- Prompts, for instance, which enable publishers to create articles from press releases can be written by editorial teams, but they need to be optimised, centrally maintained, integrated into workflows and reviewed periodically. (Peter Dyllick-Brenzinger, head of product and engineering, Purple)
- While, ultimately, productionising AI processes is the goal, there’s still an enormous amount of value in picking up these tools and encouraging your workforce to work with them. The key thing is figuring out how you get people to experience the tools, allowing them to work with them, encouraging experimentation and having the right kind of lightweight governance to make that possible in your organisation, just as a starting point. (Ian Mulvany, chief technology officer, BMJ Group)
- A great way of creating sales leads is to build the ultimate local business directory, using AI on top of a solid crawling engine to gather all the available data about local businesses in the publisher’s area. A comprehensive and engaged business directory is an amazing door opener, helping sales teams prioritise leads and enjoy higher conversion rates. (Christian Scherbel, CEO, Smartico)
- When using an AI for copy fitting, the AI should not rewrite the whole article, but instead give the editors suggestions with shortened sentences, so the editors can then decide whether or not to accept the changes. The results have been remarkable! These and other AI tools should be integrated into workflows, not sit external to them. (Tom Pijsel, VP product management, WoodWing)
In ninety minutes of in-depth Q&A, there was loads more advice and insight, which you can catch by watching the recording here. (You will need to fill out the registration form and then you’ll be directed to the recording.)
You can catch James Evelegh’s regular column in the InPubWeekly newsletter, which you can register to receive here.
