Applications of generative AI for publishers
Artificial intelligence has evolved quickly in the last year. ChatGPT has shown how easy it is to create well written content on any topic. Is this an opportunity for publishers and media businesses to streamline content production or audience analysis? Or does it pose a risk to readers’ ability to judge reliable and authoritative content? This is an evolving topic, and we may be at the top of the Gartner hype cycle, but I have tried to summarise how media businesses are using AI right now.
How many publishers are using AI?
RISJ’s survey of news publishers shows that almost a third are using AI regularly for recommendations in content production:
Charlie Beckett, Director of Polis at the LSE, predicts that we are at a tipping point for AI-enabled or “synthetic” media. He believes the proportion of AI-created content in news organisations will rise from under 10% to a third. And where news media go, niche publishers will follow.
Risks of generative AI for publishers
Inaccuracy and hallucination
Chat GPT is good at creating fluent copy but is reliant on its training data (only up to 2021) for the factual content. A major concern about ChatGPT is including inaccurate facts or even making facts up (known as “hallucination”). There are examples of it making up plausible but non-existent academic papers, see this thread.
The training data for tools like ChatGPT are drawn from the internet, so there may be a bias to more sensational content or unreliable sources. And there have been some unsettling stories of AI reflecting current biases, such as suggesting girls study humanities rather than STEM, or that employers can pay their black staff lower wages.
As more content is created by generative AI, search engines may find it hard to distinguish between authoritative content and AI-created content, where facts may not be easy to check.
Some are concerned about AI making journalists redundant, but it is more likely to be replacing tasks than entire jobs. AI can be used to speed up repetitive tasks and help journalists be more productive. Charlie Beckett’s view is that AI will push journalists to higher standards, to distinguish their work from AI-created content.
Applications of generative AI for publishers
Publishers are already using AI to create content more quickly and in more formats.
David Caswell shared some practical examples from the BBC in a recent PPA webinar. He’s worth following on twitter @structstories
- Automated copy editing eg fix the typos
- Automated summarisation (of one or more articles)
- Create a list of factual claims for checking in an article (which may be written by ChatGPT)
- Cleaning a data set by sorting into consistent headings
- Writing a step by step explainer of an event (and describe a suitable image)
- Convert a written article into a script, then use a synthetic voice (eg ElevenLabs Voice) to create audio, and a video package (eg Audiogram) to turn this into a graphical video – or even a simulated presenter using Synthesia
Chris Stone of New Statesman is sharing lots of examples on twitter @ChrisStoneTV
- Editing text with a tool called Merlin
- A tool based on GPT-4 that reviews financial statements as PDFs and can answer questions
Di5rupt’s recent report on AI lists some useful tools for journalism and content creation
- Transcription: turn audio into text with otter.ai, Sembly or Airgram
- Translation: Google, DeepL, Smartling, Phrase
- Research: ChatGPT is a starting point, although as noted above its sources may be fictitious
- Creating images: some publishers have used Midjourney to create images for articles. Stable Diffusion can create images based on text input
- Writing code for website templates or immersive stories
Audiences and analytics
AI can help publishers be more efficient when managing audiences and online marketing.
Sarah Marshall of Conde Nast, in PPA event:
- Generate a first draft of search queries for SEO
- Auto-translations of syndicated articles
- Automatic tagging (to power related articles)
- Predict which stories will get greatest audience
Di5rupt’s AI report:
- Analysing data: finding facts and trends in CSVs to find facts or make predictions
- Audience segmentation: Piano can segment your users into groups based on likelihood to subscribe
- Chatbots to automate customer service
- Email marketing personalisation and automation, also drafting copy and subject lines (eg Phrasee. )
Examples of publishers using AI
Many large media organisations are already generating sports results, weather reports and news articles with AI.
- Wolfgang Zehrt, AI content expert based in Germany, supplies content for 40 publications. This German news website demonstrates automated content created by GPT 4.
- UK news publisher Reach is testing AI-created articles in local news, and BuzzFeed is using AI to generate quizzes.
- Italian newspaper Il Foglio has challenged its readers to correctly identify AI texts in its daily edition, according to this article from RISJ.
- Danish news media company Zetland has developed a speech to text transcription service for journalists designed for smaller languages (like Danish) called Good Tape.
- Canadian newspaper Globe and Mail has developed a tool called Sophi to automate its web homepages, allowing editors to be more productive.
With a third of news media already actively using AI, it’s not a time to sit on your hands. Test it out, ask your team to experiment and look for opportunities to swap tips with your peers.
I moderate a community of publishing and media leaders – the Speciall Media Group – on Guild and will ask the 240 members to share their experiences on AI as well as other best practice.
Here’s the link to request to join.
About the author
Carolyn Morgan has acquired, launched, built, and sold specialist media businesses in print, digital and events. She now advises niche consumer and B2B publishers on developing new products and digital revenue streams as a consultant and NED.