How do journalists really want to use generative AI?

Credit: Created using Magic Media, a generative AI app by Canva

A new report by the Associated Press, University of Amsterdam and Northwestern University considers the untapped potential of AI in the newsrooms.

The Generative AI in Journalism: The Evolution of Newswork and Ethics in a Generative Information Ecosystem is based on survey responses from 300 news professionals, mostly from the US. It explores how they currently use, and would like to use, generative AI - automated technology that has never been easier to use or afford.

Most journalists say they understand how generative AI works (81 per cent) and most newsrooms use it in some capacity (74 per cent).

There are some clear news cases so far, most notably generating text (70 per cent), especially to read and summarise large volumes of text. Working with data, however, is a more interesting problem.

• Read more: How NRK uses AI-generated summaries to boost younger readers’ engagement

Four in five journalists want generative AI to help them analyse data and information, and nobody said they did not want AI's help for this task.

It is easy to see why: journalists find analysing data and information the third most tedious typical journalistic task. Just seven per cent of journalists offload the task to the tech but double that number (14 per cent) would be open to doing this if it can "guarantee quality results."

So quality seems to be the rub. Co-author of the report Hannes Cools, assistant professor at the University of Amsterdam, says this is compounded by considerable ethical challenges.

He has seen journalists acquiring datasets that are not publicly available and contain lots of GDPR-sensitive information. Even if these datasets contain disclaimers, generative AI tools will often not account for it when crudely pasting the data in.

So what can news publishers do? At a time of lawsuits between news publishers and generative AI companies over the use of news content, news organisations can create their own tools trained on their content.

Colombian independent news outlet Cuestión Pública provides one example through its Odin Project. The tool is trained on its investigative journalism, which contextualises breaking news in terms of the relationships and connections among the elites of Colombian society, such as who is related to whom, and who sits on whose board. This makes for striking and deeply referenced social media content.

Sobrinos y primos del presidente del Senado Iván Name se han movido entre cargos a través de asesorías en entidades públicas, en los mandatos de Juan Manuel Santos, Iván Duque y hoy en el de Gustavo Petro. Te contamos más de esta red familiar 👇 https://t.co/kq76mTaROE

— Cuestión Pública (@cuestion_p) April 12, 2024

The deeper benefit is that this preserves the knowledge that is often locked away in the brains of experienced editors and is lost when they move on or retire, says David Caswell, founder of innovation consultancy StoryFlow and a former executive product manager at BBC News Labs, in a role focused on AI-based initiatives.

Caswell was involved in last year's AI in Journalism Challenge, which sought to understand what digitally native, mission-focused, low-staffed newsrooms can do if given the tools, support and freedom to run with the generative AI.

The Cuestión Pública received an honourable mention, but it was Rappler in The Philippines that won for developing its TLDR model. This AI converts news stories into a range of multimedia content for Gen Z and millennial news consumers.

Reformatting or creating multimedia content is what 20 per cent of the AP survey respondents are doing with generative AI. But the smart move here is leveraging it for a new brand.

Another example comes from the academic news website The Conversation, which also participated in the challenge. It is looking to generative AI to reversion content for niche microsites. It set up a site, YTTA.ai, to help young Indonesian voters make informed decisions ahead of their national election back in February.

Caswell noted on the Journalism.co.uk podcast, that only two in 12 of the projects involved improving efficiencies, like search engine optimisation. The AP survey finds something similar - few news organisations are using it for coding (five per cent) or metadata (less than one per cent).

Publishers of all sizes need to be cautious in their approach to generative AI. But despite thinner resources, smaller publishers appear to be moving faster.

"The radical democratisation [of generative AI] is disproportionately advantaging smaller, more agile, more imaginative news organisations with less to lose," says Caswell.

Larger, legacy news outlets feel more responsibility on their shoulders. They often have brands several centuries old and societal impact, which slows down the execution of ideas. He recommended developing more nimble, more autonomous AI teams to begin experimenting.

And many are experimenting and accelerating, to their credit. Some Scandinavian news brands were dabbling with AI before the boom of generative AI, and it is little wonder that Danish news organisation Ekstra Bladet or Swedish newsgroup Schibsted are now developing their own proprietary tools for journalists.

According to Cools, this is in recognition that generative AI tools are black boxes that lack transparency and leave journalists unable to know whether they can trust its results. Worryingly, one in ten journalists rely on their "gut feeling" or "editorial judgement" to decide on this.

It is one of the reasons why so many in the survey call for clearer newsroom guidelines, and why more newsroom roles dedicated to AI strategy and training are starting to appear.

As the report notes, there is also an unmet need for newsrooms to develop their own tools that have bespoke interfaces for journalists. As in, tools which also provide a playbook of recommended and approved prompts. However, developing these tools either requires new expertise or partnering with other companies.

• Read more: How a Greenlandic publisher uses its own AI translator to boost subscriptions

And it is not like small newsrooms will have the know-how or pocket change to do this. That said, this could present new opportunities for collaboration between bigger and smaller newsrooms.

"AI will change journalism more as a process than a product," concludes Cools.

Continue the conversation with us at the digital journalism conference Newsrewired on 22 May 2024. Check out the full agenda and grab your ticket now

Free daily newsletter

If you like our news and feature articles, you can sign up to receive our free daily (Mon-Fri) email newsletter (mobile friendly).