Media Briefing: Publishers’ AI task forces evolve into a more distributed model of experimentation

The task forces publishers created earlier this year to experiment and oversee initiatives around generative artificial intelligence technology have evolved over the course of this year to a more distributed model, with larger groups now involved in testing AI’s different applications across companies.

When OpenAI’s generative AI chatbot ChatGPT first launched in November 2022, publishers started forming task forces in the following months. Led by executives from different departments, their goal was to understand the impact of the generative AI tools and how they worked, and to find opportunities to incorporate the technology into their workflows. From there, they started to develop guidelines and positioning on their companies’ approaches to AI.

But while many task forces started out with just a handful of representatives from different teams — take those formed by Forbes and Trusted Media Brands in March, for example — they look quite different now. At Forbes, the original seven-person group has been disbanded and all employees are now encouraged to experiment with generative AI technology, according to the publisher’s chief content officer Randall Lane. And at TMB, the original “steering committee” is still intact, but five smaller subcommittees have formed to work on specific initiatives, according to Jacob Salamon, TMB’s vp of business development.

“We’re really trying to take more of a hub-and-spoke approach on the steering committee,” Salamon said. TMB’s original group helps teams build business cases for the AI tools they want to use and then get legal approval to use them, in addition to brainstorming new ideas and organizing company-wide initiatives related to AI. The group also sets up training sessions to teach teams how to use AI tools, he added.

Meanwhile, Business Insider created an editorial task force in July that is made up of “a couple dozen people” experimenting with generative AI technology, according to editor-in-chief Nicholas Carlson. 

Ingenio has gone a different route, and is planning to expand its full-time team dedicated to overseeing generative AI products, said Josh Jaffe, Ingenio’s president of media. The company separated its publishing operations from its AI team in April. Both teams report to Jaffe.

Task force changes

As generative AI tools have become more approachable this year, media companies have looked to democratize experimentation to determine how different teams are using the technology or how they can start doing so, publishing execs told Digiday.

“A year ago [when ChatGPT] had just come out, it was like this exotic thing. But now, there’s not a curious person in the media world who hasn’t used It. So the idea that only some people should use it and others should wait for them to tell them what’s going on — the world is moving too quickly,” Lane said. He said he believes 2024 is the year when “we figure out collectively how it can be a tool for everybody and what those best tools [for reporters] are.”

Forbes’ chief digital and information officer Vadim Supitskiy said there are three main areas that the company is encouraging employees to look into: how AI can help engage readers, how it can improve journalists’ workflow efficiency and how to build new products powered by the technology (such as Adelaide, a generative AI site search tool Forbes debuted in October). Supitskiy also said AI is regularly discussed in executive team meetings.

TMB’s subcommittees — which folks from the original group are also participating in — have different goals, such as creating editorial guidelines or developing an “AI roadmap” to determine where to invest more resources, Salamon said. Those subgroups tend to meet biweekly, the same cadence as the steering committee, he added.

TMB hosted an “AI hackathon” a few weeks ago to give employees a more “hands-on experience” with the technology, Salamon said. The hackathon had 75 participants from across the company organized into 10 teams who competed for prizes, Salamon said. One of the ideas that came from the hackathon was a prototype for an AI-powered meal planner that uses TMB’s catalog of thousands of recipes, tailored to specific requirements such as diet or number of ingredients. The idea for the event came from the steering committee, but the winning initiatives are being worked on by TMB’s smaller AI-focused groups.

For 2024, TMB’s AI-focused groups are helping to develop a “customer-facing search tool” (that will browse the company’s archives in a similar manner to Forbes’ Adelaide) and a content recommendation engine for their sites powered by AI algorithms that will be built in-house, Salamon said.

Business Insider created an editorial task force this summer, which includes representatives from its newsroom union and some folks from its product team, Carlson said. He said he wanted a big editorial group to be part of the decision-making on AI usage because other organizations’ “top-down” approach “seems to be like stepping on a rake.”

“In general, I think it’s good to circulate decision-making, socialize concepts and test it out with a group to see how people feel about it,” Carlson said.

Business Insider’s group meets about once a month, he noted. The task force is testing generative AI tools to create the first draft of the summary bullets at the top of the publisher’s articles and for image generation.

Ingenio plans to expand its AI-dedicated team of five early next year, Jaffe said. He declined to share how big he wants the team to be, noting that budget discussions were underway for 2024. 

When large generative AI tech companies unveil new tools, Jaffe wants Ingenio’s AI team to be able to figure out how to use those tools to create products for their readers. For example, OpenAI released an image recognition feature to ChatGPT in September and Ingenio plans to launch a product in January that lets people upload an image of their palm and a trained chatbot will provide a palm reading.

Ingenio’s AI group is behind the launch of 10 spiritual guide chatbots (the first one launched in June), Jaffe said. On average, users are spending more than 10 minutes with these chatbots. After three prompts, users are asked to register through their Google or Facebook accounts to continue the conversation. The conversion rate is 11.5%, according to Jaffe.

Changes to AI guidelines planned for 2024

As these publishers’ AI task forces evolve, more people are getting involved in brainstorming publishers’ guidelines regarding the use of AI. But so far, little has changed in that department, according to conversations with publishing execs.

TMB’s editorial subcommittee drafted an internal and external set of guidelines a few months ago, but they have not been published publicly, Salamon said. The internal, company-wide guidelines set specific tenets on AI usage (such as disclosing to managers when an employee uses AI tools to assist their work), and the other guidelines are for writers. There haven’t been any significant updates to either document, Salamon said.

Forbes’ original task force developed two “cardinal rules” early this year: Do your own writing and your own first-person reporting, Lane said. But there won’t be any additions to the publishers’ guidelines until sometime next year, he noted.

“Sometime in ‘24, we’re going to have enough advancement that we can start to build more fulsome guidelines. But right now, we’re still in the early days in the experimental phase,” Lane said. He said he wants to continue to encourage employees to experiment with generative AI tools, and, as more practical applications surface, more specific and “formal” guidelines will be necessary, he added.

Business Insider hasn’t created any specific guidelines beyond the ones outlined in Carlson’s memo sent to the company’s newsroom in April.

“Now’s the slow part,” Carlson said. “We’re experimenting with [generative AI], seeing what’s worth doing more of. … But it’s not like a snap of the fingers and all of a sudden a newsroom like ours is AI-wired. And part of that is by design, because you have to be careful. … We’re going through a deliberative process.”

What we’ve heard

“Early on, there was more fear from everyone in the publishing business — from executive team members all the way down to staff — [about what generative AI tools] are going to mean existentially for the company, for the industry, for our jobs. But the more and more we started to look at the tools … they took on a very different sort of flavor … that this is a moment in your career to learn a new tool to see how you can empower your work. … That was a paradigm shift [and] a tone shift in terms of what these tools represented [for our business].”

— Jacob Salamon, vp of business development at TMB

Eliminating bias in programmatic

While the past three years have put a spotlight on investing in Black-owned and minority-owned to create a more equitable advertising ecosystem, there is a misconception that racial bias has been eliminated or even significantly reduced from the programmatic advertising space as a result. 

Yes, many multicultural marketplaces were built to funnel dollars to these historically underfunded media companies and more brands and marketers have pledged to allocate a specific amount of money to reaching multicultural audiences, but by not updating buying practices or the technology that programmatic advertising was built on, those actions still run into a lot of roadblocks. 

“There is still bias that exists within programmatic today and the reason for that is due to legacy buying practices,” Kerel Cooper, vp of advertising at Group Black, said during an on-stage session at the Digiday Programmatic Marketing Summit, which took place in New Orleans this week.

Rudimentary brand safety and brand suitability tools, such as keyword blocklists and site classification, are regularly interfering with the scale marketers expect to achieve when buying programmatically. Not to mention the fact that the programmatic system at large was designed to disadvantage smaller publishers, which oftentimes include minority-owned media. 

“When you have a programmatic campaign running that has multiple deals in it, the machines are automatically going to optimize to where they are finding the most amount of inventory for the lowest price possible. And that usually harms smaller publishers,” said Cooper. 

This became a challenge when Group Black launched its programmatic network about a year ago, encompassing both the media company’s portfolio and a network of other Black-owned and minority-owned publications.

Instead of trying to case solve one-by-one, Cooper and his team reached out to verification firm DoubleVerify to uncover any systemic problems at the brand safety level of content verification.

Group Black and DoubleVerify removed outdated words from the keyword blocklists of 10 active campaigns that were not scaling properly and categorized sites at the page level (versus domain level), increasing available inventory by 15% and decreasing the amount of blocked sites by 98% across the media company’s network, Cooper said.

Cooper gave the example of the word “Obama,” which is still on many blocklists despite the fact that former President Barack Obama has not been president in almost seven years. Group Black and DoubleVerify found as much as 28% of the inventory being blocked was related to the keyword “Obama,” though the content was in the entertainment and lifestyle categories, not political or news coverage.

“I think we should do away with keyword block lists and we should focus more on categories specifically,” said Cooper. “There are always going to be categories that marketers are going to want not to be around whether that be breaking news or politics … but even in the work that we’ve done with DoubleVerify, we’ve shown that category blocking versus keyword blocking opens up more inventory.” — Kayleigh Barber

Numbers to know

1,500: The number of people Spotify is letting go in its third round of layoffs this year, representing about 17% of its workforce.

750: The number of Washington Post employees expected to participate in a 24-hour strike on Dec. 7, as the union protests staff cuts and contract negotiation delays.

10%: The amount that Bustle Digital Group’s overall revenue is expected to decline in 2023, bringing in a projected total of between $136 million and $144 million for the year.

180: The number of days BuzzFeed has to get its stock price up to the $1 per share minimum requirement to remain on the stock exchange, after it received a deadline extension from the Nasdaq.

What we’ve covered

News publishers hesitate to commit to investing more into Threads next year despite growing engagement

• News publishers are cautious to pour more resources into Threads, and don’t have plans to do so in the near future as limited available data makes it difficult to determine whether investing more into the social platform is worth it. For now, Threads remains a place for experimentation.
• Execs at The Boston Globe, CNN and The New York Times said they were seeing engagement grow on Threads since its launch. However, other news publishers like the BBC and the Guardian U.S. have stopped posting from their main accounts on the platform, and are monitoring whether it’s worth investing resources into. 

Read more about news publishers’ varying approaches to Threads in 2024 here.

Publishers’ programmatic revenue didn’t shake out the way they’d hoped, but it’s still a bright spot

• Publishers’ hopes for their programmatic ad revenue didn’t quite come to fruition this year, according to a Digiday+ Research survey of more than 300 publisher professionals. Nearly half (45%) said in Q1 2023 that building their programmatic business would be a large or very large focus in the next six months. 
• Fewer publishers are getting a large portion of their revenues from programmatic and more are getting a moderate portion from that source. Of those surveyed, 28% said in Q3 of this year that a large or very large portion of their revenues came from programmatic ads, down from 38% at the start of the year, and 35% of publishers said in Q3 that programmatic ads accounted for a moderate portion of their revenues, up from 21% in Q1.

Read more about publishers’ programmatic ad revenue here.

WTF is Google’s Protected Audience?

• Google’s Privacy Sandbox — its collection of ad targeting and measurement methods — renamed an update in April called FLEDGE to “Protected Audience.” The third-party cookie alternative operates by making ad auction decisions in the browser, rather than at the ad server level.
• While advertisers, publishers and ad tech firms may cringe at the idea of letting Google control more of the ad process inside its own browser, it’s just one of many approaches for replacing third-party cookies.

Learn more about Google’s Protected Audience by watching the video here.

What we’re reading

Dotdash Meredith advertising CRO Sara Badler heads to Morning Brew:

Badler is the new chief commercial officer of advertising at Morning Brew, according to Axios. Badler will be responsible for revenue growth, managing sales, strategy operations and the creative studio. She spent five years at Dotdash Meredith, most recently as CRO of enterprise advertising and partnerships. She’s also worked at The New York Times, Hearst and Forbes. 

Yahoo News hit by layoffs:

Yahoo News is the latest digital media publisher hit by layoffs as we approach the end of the year. While The Daily Beast didn’t say how many people were affected, the layoffs hit a number of verticals including the website’s “Originals” teams and the “In The Know” news vertical for Gen Z and millennial audiences, which is getting shut down.

Spotify cancels two podcasts:

Spotify is sunsetting two podcasts, “Heavyweight” and “Stolen,” according to The New York Times. The audio streaming platform has made significant cuts to its podcast business this year as it strives to remain profitable. Spotify will not renew the two podcasts’ contracts but the shows will have the option to find a home elsewhere. The shows are produced by Gimlet Media, the podcast studio Spotify acquired in 2019.

Readers want publishers to disclose AI-generated content, but trust them less when they do:

An overwhelming majority (over 80%) of readers want news publishers to disclose when generative AI has contributed to the coverage they’re consuming, but new research from the University of Minnesota’s Benjamin Toff and Oxford Internet Institute’s Felix M. Simon found that audiences perceive news labeled accordingly — and the organizations behind them — as less trustworthy (though not less accurate or more biased), according to Nieman Lab. However, when sources are provided alongside the text, labels disclosing the use of AI do not reduce trust in news, the study found.

NPR has a new podcast exec:

NPR announced Collin Campbell, Gimlet’s former head of new show development, will oversee the company’s podcast strategy. NPR let go of nearly 10% of its workforce earlier this year and canceled four podcasts. Campbell will replace head of programming Anya Grundmann, who announced that she will leave NPR at the end of this year after three decades at the network.