The Washington Post, Aftonbladet find success with new chat products

By Sonali Verma

INMA

Toronto, Ontario, Canada

Connect      

The team at Sweden’s Aftonbladet took on an interesting challenge this year: What if they picked a topic their audience typically does not engage with and built a chat product around it to drive engagement?

The result was Valkompisen, a GenAI “election buddy” chatbot that beat all expectations. Since its launch in May, it has answered more than 150,000 questions on the EU elections held in June. It could talk about the structure of the EU, the history of the EU, previous election results, current topics in the EU debate, votes in the EU Parliament, party groups, as well as the opinions of the Swedish parties on various issues and which candidates were running.

Valkompisen, Aftonbladet’s chatbot that answered questions on the EU elections.
Valkompisen, Aftonbladet’s chatbot that answered questions on the EU elections.

It used OpenAI’s GPT-4 and was trained on data from more than 100 sources collected by Aftonbladet’s journalists and developers. “It is based on official, reliable sources (and on answers to questions we put to the parties), and it is instructed not to make things up or fill in, which sometimes happens with the AI-powered chatbots,” said Martin Schori, associate publisher of Aftonbladet.

Reader education was part of the launch. The news brand helped readers by providing tips on how to write prompts and encouraged them to maintain a dialogue with the bot rather than giving up after one attempt to get an answer, while also telling them about the flaws inherent in GenAI bots.

It was not the obvious choice to invest “so much money and resources in a completely new product, based on a completely new behaviour and where we let go of some of the rigid control” over the quality of output, Schori said.

But Aftonbladet gained invaluable insight into what its readers were wondering about and is now looking at how else to use this technology. 

The Washington Post has also just launched a new chat product, this one focused on climate-related topics, ranging from climate change to carbon offsetting to recycling. 

Climate Answers will only give answers that are drawn from the Post’s own journalism and is aimed at bridging the gap between how the publisher writes about climate and the information its audience is seeking.

The Washington Post’s new climate chat product.
The Washington Post’s new climate chat product.

It is worth noting the news company is being very transparent with its readers about the objectives of the bot.

“The Washington Post has been publishing coverage of the climate and environment for more than two decades. We know that readers may have questions that aren’t always easy to answer,” the newspaper said in a message to readers. 

“This tool searches the past eight years of our articles and ranks the results based on relevancy. We then use a large language model to write a response to match the query you have posed.

“Our intent is not to replace the critical role our journalists play but to offer readers new ways to engage with the work we have already published. How can AI allow us to surface answers from our deep trove of reporting? What types of reader journeys do we make possible via a novel search interface?”

If users spot an error, they can provide feedback to the Post via a feedback form. “This is an experiment we are releasing so we can learn from you how to improve the experience. We expect it may not always function exactly as we hope — which is why we are asking you to confirm with the published articles.”

Climate Answers was launched after the newspaper found that new formats for its content work well to engage younger readers — for example, they prefer story summaries over headlines in determining whether to read further. The Post’s senior editor for AI strategy and innovation, Phoebe Connelly, used to head its Next Generation initiative, which was aimed at attracting younger readers.  

Date for the calendar: Please join us on Wednesday, August 7, at 10 a.m. ET/1600 CET to hear two success stories of GenAI innovation — one on audio playlists and one on internal workflow tools that were widely adopted. 

GenAI adoption: It’s complicated

GenAI has been around for more than a year and a half now. We are now at the point where most news brands have actively experimented with GenAI and built various products. But, based on my conversations with INMA members, one challenge now popping up is that the users for whom they were built — generally, newsrooms — are often slow to embrace them.

Indeed, we saw similar results in a poll conducted during a GenAI Webinar on July 10, attended live by about 70 INMA members:

Survey results from attendees at my July 10 GenAI Webinar.
Survey results from attendees at my July 10 GenAI Webinar.

Similarly, the Online News Association polled 35 journalists and found that only about half used AI tools often or always:

Results of a recent Online News Association poll of journalist use of AI tools.
Results of a recent Online News Association poll of journalist use of AI tools.

This challenge is not specific to the media industry or to a particular part of the world. A survey conducted by SAS and Coleman Parkes of 1,600 decision-makers in key global markets looked at where different regions ranked in fully using and implementing generative AI into their organisation’s processes:

  • North America: 20%

  • APAC: 10%

  • LATAM: 8%

  • Northern Europe: 7%

  • South West and Eastern Europe: 7%

“With any new technology, organisations must navigate a discovery phase, separating hype from reality, to understand the complexity of real-world implementations in the enterprise. We have reached this moment with generative AI,” said SAS CTO Bryan Harris. “As we exit the hype cycle, it is now about purposefully implementing and delivering repeatable and trusted business results from GenAI.”

Harris’ sentiment is echoed by media leaders: “This is revolutionary technology, but it has not produced revolutionary changes,” despite being integrated into the CMS, where journalists should find it easy to use, one executive told me. 

One reason is that staff members are so busy that learning how to use a new tool appears to be another chore they need to set aside time for. Many news organisations have extensive training programmes, which include Slack channels, training videos, as well as human-to-human workshops — but dragging busy employees away from their day jobs to pay attention to these is a challenge.

Another is the fear of losing jobs. “We actually had staff using the tools and then pretending not to use them because they were so effective and they were worried about losing their jobs,” a technology leader at a news publisher told me.

Again, this is not a problem that is specific to the media industry.

“Lack of trust remains a major barrier to large-scale generative AI adoption and deployment. Two key aspects of trust we observed are: trust in the quality and reliability of generative AI’s output, and trust from workers that the technology will make their jobs easier without replacing them,” Deloitte’s Q2 State of Generative AI report found.

Gartner found a different barrier to adoption topped the list: the ROI calculation.

“The primary obstacle to AI adoption, as reported by 49% of survey participants, is the difficulty in estimating and demonstrating the value of AI projects. This issue surpasses other barriers such as talent shortages, technical difficulties, data-related problems, lack of business alignment and trust in AI,” Gartner wrote in May after surveying 644 respondents in the U.S., Germany, and the U.K. 

“Business value continues to be a challenge for organisations when it comes to AI. As organisations scale AI, they need to consider the total cost of ownership of their projects, as well as the wide spectrum of benefits beyond productivity improvement.”

Infographic from Gartner’s May 2024 survey.
Infographic from Gartner’s May 2024 survey.

The survey also found that, on average, only 48% of AI projects make it into production.

Where is GenAI genuinely not useful? What have you tried that did not work? Please take 30 seconds to anonymously tell us. We are collating responses for research (and will not name people or organisations).

Worthwhile links

  • GenAI and traffic: Prepare for a future where we write for smaller audiences — build your business model around them.
  • GenAI usage in journalism: Looks like editors are not checking over the bot’s work much.
  • GenAI for data analysis: Some handy tips here, here, and here
  • GenAI: Getting started: Some good points in here for anyone still wondering how to get going. 
  • GenAI and data poaching: Looks like bots are still crawling sites for training data.
  • GenAI frenemies: The New York Times is using OpenAI technology to write headlines even as the newspaper continues legal action against the tech company.
  • GenAI and facts: Journalists, it’s still getting them wrong.
  • GenAI and jargon: A researcher used GenAI to demystify complex research to help reporters.
  • GenAI and revenue: You’ve read about the crazy costs. Now, a look at the revenue hole.
  • GenAI to raise the dead: Long-dead actors will narrate books for ElevenLabs.
  • GenAI to raise the dead II: Reunite with your dead loved ones.
  • GenAI and music: Is this fair, transformative use, or copyright infringement?
  • GenAI and climate: Google’s greenhouse gas emissions rise almost 50% due to AI.
  • GenAI 3-D printing: Type in words describing what you want and get a printed product, courtesy of Meta. 

An AI diversion

It’s road trip season in the northern hemisphere, with school holidays and warm weather. Take a look at the highway of the future, which will talk to the cars of the future.

About this newsletter

Today’s newsletter is written by Sonali Verma, based in Toronto, and lead for the INMA Generative AI Initiative. Sonali will share research, case studies, and thought leadership on the topic of generative AI and how it relates to all areas of news media.

This newsletter is a public face of the Generative AI Initiative by INMA, outlined here. E-mail Sonali at sonali.verma@inma.org or connect with her on INMA’s Slack channel with thoughts, suggestions, and questions.

About Sonali Verma

By continuing to browse or by clicking “ACCEPT,” you agree to the storing of cookies on your device to enhance your site experience. To learn more about how we use cookies, please see our privacy policy.
x

I ACCEPT