Ways to test news chatbots while preserving trust
Hi there,
Last quick reminder that we’re inviting a new set of readers this week. If you’ve been enjoying Business Side, please forward this to a couple colleagues and point them to the sign-up link.
Thanks for being an early reader!
Generative AI, by its nature, isn’t designed for accuracy. So publishers have been hesitant to use AI chatbots to engage readers (understandably so).
An app called Curio just released a news chatbot that offers some inspiration for how publishers might balance experimentation and trust.
A little background on Curio: The company, which launched in 2016, licenses content from publishers and offers audio versions of articles, read by professional voice actors. Curio’s partners include over 30 publishers such as The Wall Street Journal, The Guardian, and The Atlantic.
Here’s how Curio’s AI chatbot works: You type a question into the chat box, and then Curio returns a “playlist” with snippets from different articles. An AI voice introduces each piece and offers a conclusion.
Curio’s theory is that combining clips from fact-checked stories helps avoid inaccuracies. (They didn’t comment on Curio’s AI-generated introductions and conclusions – which may have the accuracy risks that come along with generative AI.) “We are trying to create, from a technical perspective, an AI that doesn’t hallucinate,” said Chief Marketing Officer Gastón Tourn.
Curio’s general philosophy points towards several ways for publishers to experiment with AI chatbots. The most conservative approach would be simply recommending articles based on a request. Another conservative approach would be stringing together text verbatim from different articles.
Skift, the B2B travel publisher, recently launched an AI chatbot that takes a more ambitious (and riskier) approach, using generative AI to compose original responses based on Skift’s archive. Skift acknowledged that the product isn’t fully ready, but it launched anyway to “help us learn all its use cases, make it more accurate, give us ideas on new features to build in, and frankly, figure out how to stay ahead of everyone else in the industry.”
Whether taking a conservative or risky path, being transparent with readers will go a long way.
Here’s a case study on how NPR’s Audience Editor for Latino Audiences structures his role:
Pablo Valdivia told Nieman Lab that his work falls into three major buckets.
Acting as connective tissue. “As the connector, I pull out stories that I believe Latinos would value most and ensure that they don’t slip through the cracks. That can look like platforming stories on our social pages, making sure certain podcast episodes have a digital presence, or connecting a reporter to a video producer.”
Writing and pitching stories. “I have my ear to the community and try to make sure our newsroom does too. I flag potential stories and topics for coverage and pick stories up here and there when needed.”
Running a social media brand for Latinos. NPR Oye “is dedicated to spotlighting stories and videos from within NPR that are by and for Latinos.”
And here’s the latest news in digital media:
The NYT launched an audio app. Paying subscribers can use New York Times Audio to access the Times’ free podcasts, as well as new exclusive shows. Stephanie Preiss, who leads the audio business at the Times, said “we believe that—kind of similar to what we’ve done in text journalism, if you will—we can start to move our most engaged users into our own apps and platforms.”
The BBC will share its fact-checking process more publicly. A 60-person team, called BBC Verify, will work across platforms to verify reporting & share evidence with audiences. According to BBC News chief executive Deborah Turness, “News consumers have told us that the more they know about the work our journalists do, the more they will know they can trust our journalism.”
OpenAI blocked Fiscal Note from using its tech to support lobbying. OpenAI objected to Fiscal Note using its platform to help lobbyists send targeted messages to elected officials. OpenAI’s policies prohibit using its model for “political campaigning or lobbying,” including “generating high volumes of campaign materials” and “generating campaign materials personalized to or targeted at specific demographics.”
More on AI:
Semafor is using AI to proofread stories.
The AP is using AI to analyze earnings reports and extract info from documents.
Amazon plans to bring ChatGPT-style AI to Alexa devices.
More on publishers:
WaPo subscribers get $5 back when they pay with Venmo.
Global media leaders are calling for the release of Apple Daily publisher Jimmy Lai, who is detained in Hong Kong.
A former iHeartMedia exec is launching a podcast investment firm.
Thanks for being a part of Business Side’s private beta. Have a great day!