What we have learned about journalism in the AI era two years on

Two years on from the launch of ChatGPT what do we know about journalism in the new era of artificial intelligence? Glide Publishing Platform hosted dozens of the brightest minds in media at a series of roundtables on the place of AI in media and publishing at two major Press Gazette events in London and New York, here are some of the themes that emerged.

by Rich Fairbairn

Published: 19:39, 23 December 2024

Last updated: 20:12, 23 December 2024

A variation of this article first appeared on pressgazette.co.uk.

For the first year after the launch of ChatGPT it seemed like the publishing industry was behind the ball in the conversation about AI, grappling to get up to speed with the assortment of new tools on offer, and understandably focused on the threat presented by AI startups stealing their content and traffic-grabbing features, like AI Overviews which end user journeys before a publisher gets a click.

After recently having the good fortune to participate in two of Press Gazette's major publishing events in London and New York, I believe that is no longer the case with publishing organizations now putting themselves ahead of the AI conversation — at least in terms of how it can be used internally. 

Through hosting dozens of the brightest minds in media at a series of roundtables on the place of AI in media and publishing, I got first-hand insight into what's actually happening inside the major publishers and the consultancies dealing with media brands of all scales. The discussions revealed that while the tech is advancing at a rapid rate, so is the industry's understanding and usage of AI, borne out by the change in conversations we are having with media firms and editorial teams. Here are some of the key themes that emerged: 

Publishers are largely on board the AI train

Two years into a ChatGPT world, publishers have moved from being behind the AI conversation to driving it themselves. Most of the big titles we spoke with now have formal policies in place covering the use of AI, rapidly advancing internal expertise in place, editorial and business training programmes under development, and a better understanding of the threats AI use can present to publishing businesses. 

The roundtables were all off the record, but one senior figure from a New York-based title explained: "It was quickly clear that using AI was going to be an everyday thing, but we didn't have any expertise in it — really, no-one did.

"On one side [of the business] we started to figure out what it can do for content teams, while on the other side we had to think about things like copyright, who we wanted to deal with, what are our liabilities, what's our legal position. We've stopped using [AI] randomly, and started using it in a considered way which we can measure."

Private LLMs are where it's at

Using generic public ChatGPT tools already seems very old school, especially when so many publishers still have a cautious or even adversarial view of OpenAI.

The development of private LLMs using the publisher's own data is at the forefront of work at many large publishers — borne out by our own experience with media firms.

Top publishers are thinking far beyond using LLMs to help write things based on prompts, and are looking at how AI can analyse their libraries to produce data that only the publishers themselves have access to.

As one attendee put it, "If we all use the same LLMs, the chances are we all get the same answers. And why should we be helping [the LLMs] anyway?"

Work groups and committees are in vogue

Far from cutting headcount, so far AI seems to be the reason for many new jobs at publishers: teams that look at just how it can be used or defended against.

Nearly every major publisher we spoke to already has work groups looking at AI from a business perspective, and one said: "I was here when the internet first started, and looking back we spent years thinking about it in an ad-hoc way before we did anything sensible about it. The cycle of adoption was about a decade long.

"This time round, the reaction has been much faster and we all [as an industry] seem to be moving as one for the first time I can remember. We all share knowledge more freely, and [where I work] we are regularly given updates on policy."

Internal training - not leaving it to users 

The rapid uptake of AI tech has been mirrored by training programmes and education in using it. This seems to be for two reasons: one, to get the best from the tools on offer, particularly any created in-house; two, to enforce internal policy and remind users there may be a very good reason why AI is in fact banned. 

Blanket bans are rare but no unheard of, although one said "We toyed with a ban but it's unrealistic. Now we say, like any other work you do, check it and take responsibility for it - so in that regard nothing has changed from before the time when AI was available."

SEO and intelligent paywalls 

Workflow tools are already seen as basic. What is really exciting to publishers are the possibilities offered by AI tools which purport to boost revenue, such as improved SEO and intelligent paywalls. 

As one said, "We are already in the content business so AI which makes content isn't really transformational. But AI which can help us get more subscribers... that's interesting!"

Customer-facing AI products are rare

Overall, there are relatively few customer-facing AI products or streams of content. Almost every roundtable attendee was aware of initiatives like the Washington Post's climate chatbot — of great interest — but also of the errors made by some publishers in sending AI content to readers. Caution still rules here and AI-to-consumer is a slow road.

AI content creation shunned by serious publishers

AI can create content, but right now it looks like serious publishers are not using it for that much at all — reaffirming previous cautions to keep people between audiences and AI.

The fear of hallucinations is still very real, and even discrete attempts to try it out would likely be considered controversial. Human in the loop is the default position. 

How about trying it in secret? Well, one interesting aside occurred when a representative at a table transpired to be from a publisher who had been publicly outed for accidentally sending unedited AI content to audiences — "It was before my time!" — and how everyone at the table was aware of the controversy.

The takeaway? Media is an industry of talk, and there is no such thing as a secret!