The Globe’s newsroom leaders are making decisions using AI—and they’re not the only ones
Illustration by Drew Shannon
Note: As of late March 2023, Gabe Gonda no longer works at Sophi.
It’s autumn 2017 and The Globe and Mail is undergoing yet another redesign/relaunch, a ritual undertaken every few years to maintain currency and improve cost efficiency. At the time, Gabe Gonda is the paper’s managing director, corporate development. He oversees several editorial sections or groups and looks at the printout in front of him as colleagues from the analytics division present a slideshow of the document’s highlights. Gonda is sitting across the table with the Globe’s data science, revenue, and marketing teams. They’re gathered around a long conference table in a boardroom near the top of the 17-storey building. If they want a break from the data, they’ll look south from the Globe’s headquarters to see the East Bayfront, a Toronto neighbourhood transitioning from its industrial past to a mixed-use residential and commercial community. Gonda says the Globe was consolidating a few sections at the time, but it just so happens that it was also incubating technology that would force its leaders to make a different move.
Change in the newsroom is familiar to Gonda, who started working for the Globe in 2010 before moving to its independent subsidiary, Sophi. He has held a variety of management roles during his tenure, including arts editor, which he described in an article for Sophi as being a “happy dream.” Even so, the reality for everyone is that newspaper advertising revenues have been in decline for some time, and still are.
The year 2012 did look like the end of the world for most print news publishers. That same year, the Newspaper Association of America (now renamed News/Media Alliance) released a report on American newspaper revenue and found that print advertising revenues—$18.9 billion in 2012—were at their lowest annual level since the organization started tracking industry data in 1950. The ad crunch caught up to the Globe in 2012 as well. Executives flailed at the problem by introducing a metered paywall for the newspaper’s website and announcing layoffs.
Gonda, who was the editor of the arts section at that time, had to restructure his team and the section itself. “There was to be no more standalone Arts section during the week,” he wrote in a 2021 blog post for Sophi, adding “veteran journalists I had admired as a reader, and come to value deeply as colleagues, made a grim path to my office to review their exit packages.”
Despite these cuts, the Globe also hired its first data scientists to begin “taking back control of its destiny with the help of technology,” according to Gonda’s post. Veteran data scientist Mike O’Neill, who joined the Globe in 2012 as manager of digital analytics and data science, was one of the new hires. O’Neill had been senior manager of data science at Research In Motion (now BlackBerry) and had held management positions at TD Bank and Rogers Communications. Gonda says that O’Neill’s task was to put together a team to solve what is commonly known as “promotion bias.” That is, the imbalance between a story’s value and how a website like the Globe’s chooses to promote it.
O’Neill was already providing the Globe editorial team with audience analytics on a regular basis. But with time and the help of other data scientists like O’Neill, Gonda says that the quality of the analytics only improved. Sweeping away the concern of how well the stories were read in print compared to online, O’Neill asked questions like, “What’s the value of each piece of content, not simply based on popularity, but based on balanced metrics?”
“The ones that are just getting measured clicks—that’s not going to tell you anything other than what’s popular,” says Gonda. “And what’s popular is not what’s good for the business.” Knowing this, O’Neill and his team started developing early iterations of Sophi, described on its website as “a suite of AI-powered optimization, prediction and automation solutions.”
Within a few years, Sophi would not only be enhancing decisions made at the Globe, but it would also start making obvious choices for the paper, form the basis of a new business model, and point the way to growth.
The Light Bulb Moment
Boardroom get-togethers have been a regular occurrence for Gonda since joining the paper, but one 2017 meeting is “how the stars aligned.” Among the list of objectives was to review data and analytics generated by Sophi. “One of the great things that Sophi can tell you,” Gonda says, “is what you’re producing too much of—or what you’re producing too little of.” Between the numbers and graphs spread out on the table—and projected on the screen above their heads—what stood out for the Globe’s leadership team was that the publication was not producing nearly enough opinion pieces to match demand.
In effect, Sophi’s analytics helped justify the creation of a new section in the print edition of the weekend Globe. In other words, this time AI-generated information had told them to evolve the business model—and, according to Gonda, contributed to investment in the newsroom. “There was a light bulb moment,” Gonda says. “We were sitting around a table discussing all this, and as a group we decided we would invest in creating a new product—the opinion section of The Globe and Mail—based on that gap between demand and supply.”
Gonda and colleagues then added senior editors and “chase producer-style journalists,” doubling the size of the weekend opinion section team. “If you had been someone in my shoes in the previous half-decade or so, that was a pretty unusual moment,” he says, “to create a new product that seems sustainable based on audience insight that was still core to the business and allowed us to hire new journalists and spend some freelance money on big, ambitious ideas.”
Swedish tech company United Robots sells an AI product that automates one of the core components of journalism: writing
Nicole Blanchett, an associate professor in the School of Journalism at Toronto Metropolitan University (TMU), studied how metrics and analytics shape newsroom practice. Her 2021 research explored how audience input, interpreted through data, influences the creation of newsroom routines and norms. Blanchett argues that in newsrooms, journalists can lack the training required to apply audience data in an effective way. Instead, they rely on more “easily understood metrics like clicks,” which “often take priority over complex measures, such as engagement.” She says, “Some people may think that the only use of metrics and analytics, for example, is this kind of traffic-chasing mechanism and that journalistic instinct is a better way to know what’s important.”
With Sophi’s data stream, the Globe has found a way around that problem. In fact, it wanted to build on the success that the newsroom was beginning to see, so it gave Sophi an additional task: become an automated content curator based on the analytics it was providing. For every newsroom with an online presence, editors typically have deliberate plans for each article’s placement and visuals on the website’s homepage. But now Sophi could automate that decision-making, meaning that as demand for stories changes, the program continuously updates the publication’s homepage. In a 2020 case study of Sophi and the Globe, editor-in-chief David Walmsley stated that “[t]he newsroom of the future is one where journalists can focus on finding and telling great stories—something that machines can’t do.” He added, “[T]his is why we asked our data scientists to automate the web pages, slowly and carefully testing the results before gradually implementing it across practically the entire site.”
Though Sophi’s origins at the Globe are based in analytics, Gonda says that the tech company “doesn’t really sell that commercially any longer.” Sophi’s site and print layout automation tool is one of two main solutions that the firm offers. The second is Sophi for Paywalls, offered in two forms. The first decides which articles should go behind a paywall, based on predictive modelling. The second is a dynamic, real-time, one-to-one personalized paywall, based on both content and user behaviour.
Over the ensuing years, the Globe claims that it has radically shifted its subscription base and advertisement revenue ratio, from 30:70 to 65:35. As highlighted on Sophi’s website, the Globe claims that it saw a 10 percent increase in subscription acquisitions because of AI. “One of the things that has been evident since we’ve adopted Sophi,” Gonda says, “is that our ability to invest in journalism has been improved and it’s really strengthened the core of what we do.” Robyn Doolittle’s 2017
“Unfounded” series, the culmination of a 20-month investigation into the fundamentally flawed police processes in sexual assault cases, is one example of how AI reaffirms the value of the journalist. “In the old world, that kind of thing would have been perceived as absolutely critical to the brand, but we wouldn’t have had insight into how that shapes the business,” Gonda says. “But the Sophi score for that journalism was just massive.”
Never Forget the ROB
Because the Globe is privately owned by Woodbridge, the Thomson family’s investment company, and not a publicly traded company, there are no figures available with which to analyze the Globe’s financial position. (When asked, Gonda declined to disclose details of the Globe’s overall revenue before and after utilizing Sophi, but said that the paper is profitable.)
However, Chris Waddell, professor and Carty Chair in Business and Financial Journalism at Carleton University, and a long-time observer of the newspaper industry, says there are a few factors to consider when analyzing the Globe’s status. He says no one should forget that the core of its business is the Report on Business. Ostensibly, the ROB may be in competition with the Financial Post, the National Post’s business section, but Waddell says, in reality, “The Financial Post, like everything else at Postmedia, is a constantly shrinking publication that doesn’t have the breadth or staff to match the ROB.” In essence, the ROB has no competition in Canada and is essential reading for the business community. This affords the Globe an impressive foundation of subscribers who are much less likely to cancel than non-business subscribers. As well, many subscribers are companies that buy subscriptions for their employees—the cost of which can then be written off as a business expense. Waddell asks, “How many Globe subscriptions do you think the big five banks buy, for instance?”
This committed subscriber base puts the Globe in the same enviable position as major business media properties such as New York’s Wall Street Journal and London’s Financial Times—papers that have been successful in converting subscribers to digital—except for the fact that the Globe does not have the same worldwide reach these other papers enjoy. Still, that core is important. Waddell says, “People subscribe because they think the information they get helps them make decisions that make them money, and they are willing to pay money to make money.”
Using predictive modelling, Sophi for Paywalls tells publishers which articles they should allow only subscribers to read
The other factor to consider when looking at subscriber revenue versus advertising revenue is how many subscribers are print subscribers and how many are digital. The more digital subscriptions the Globe can sign up, the lower the cost of production. Digital production costs virtually nothing, whereas printing means outlays—for paper, ink, distribution, and printing. “None of those costs apply to digital subscriptions,” says Waddell, “so the profit from digital subscriptions is much higher than print subscriptions. In fact, you, as a digital subscriber, pay the Globe’s distribution cost through your monthly internet bill.”
Reducing expenses is crucial to a business for which the well of advertising revenue is drying up. The more the Globe can convert its readers to digital, says Waddell, the more it can reduce its costs and increase its chances of profitability. Still, this is a tightrope walk for legacy papers. Dedicated older readers generally still prefer print to digital, but this is not a problem unique to the Globe, and it’s one only time can solve. The key issue is how much time legacy papers like the Globe have.
Money and the AI Shift
The financial means of any business plays a substantial role in its deployment of new technologies. Legacy media outlets and well-funded corporations such as the Globe have the potential to achieve a successful synthesis between editorial and data personnel. Of course, it doesn’t always work out that way. Patrick White, a professor at Université du Québec à Montréal’s (UQAM) journalism school, says media in general have “a tough time” approaching the future. During the 1990s internet revolution and the social network boom in 2005 to 2010, he says news organizations had a “hard time taking a step back to interpret the future.” Like everyone else, he witnessed the rise in popularity of email in the late ’90s. White, who worked for CTV National News from 1990 to 1997, says his newsroom had the means to adopt this new method of communication—though it was a bit late, since most university students had been using the internet since 1991. “It was a slow start,” he says. “CTV is a cooperative of TV stations across the country on several networks, including the affiliated stations across Canada, so it was very disorganized and not very centralized.”
White also saw that a lot of news outlets were unable to take advantage of the internet. “If you’re not part of a large media group in Canada,” he says, “it’s very difficult to find money to tackle this technological digital shift.”
In 2021, White worked with UQAM communication master’s student Nicolas St-Germain to look at how AI-related tools were being employed in Canadian newsrooms. They sent out a 12-question survey, along with a comment section, to 13 major Canadian media outlets or ones operating in Canada, to determine the state of AI deployment in the industry. The resulting paper reported “a disparity in the use of AI-related tools,” and that deployment of this technology varied across survey participants. For example, while the Globe has fully integrated AI into its newsroom, Montreal-based daily newspaper Métro doesn’t use it at all, indicating “that financial means and a media outlet’s reach play a role in whether it has integrated the use of this technology.”
While adoption of AI by smaller newsrooms, in White’s view, “does not look great,” he does think media companies with small teams and even smaller financial means can be a part of the digital transformation. His advice is blunt: “You need to stop working in silos. The key is cooperation with smaller regional groups to have a financially bigger force to proceed with technological change.”
Other AI-Related Initiatives
Having deep pockets isn’t the only path to exploring the applications of AI in journalism. Universities like TMU are experimenting with the technology, and federal money can be accessed through programs like the Pan-Canadian AI Strategy. Launched by the federal government in 2017, the program allocated $125 million for financing Canadian-based AI research and development.
A private path is Google’s News Initiative, which provides financial and technical support to publishers and journalists to accelerate innovation in the news ecosystem. Asmaa Malik and Gavin Adamson, two TMU journalism professors, received funding from the Google News Initiative’s North American Innovation Challenge to develop an AI-enabled platform that quantifies the number and kind of sources quoted within an article. The project is called the Journalism Representation Index and was established in partnership with the Winnipeg Free Press. Through machine learning, JeRI weighs sources by factors like in-story placement and frequency of attribution to deliver an index score representing their institutional power. According to JeRI’s website, the feedback it provides “could help establish a framework that would introduce some level of accountability to newsrooms and news-gathering.”
Similarly, news outlets such as Reuters, AFP, Nikkei, and La Nación have jointly produced the AIJO Project, a program that uses AI to detect binary gender bias in their reporting.
Across the Atlantic, Swedish tech company United Robots has created an AI product that automates one of the core components of journalism—the writing. Using data science, AI, and natural language generation, or NLG, UR offers what it calls “content-as-a-service.” News publications that subscribe to UR’s products are fed large volumes of journalistic content written by a trained machine.
Henning Johannesson, who initially worked with Mittmedia, originally created a text robot in 2015 to cover more sports matches based on a sports database from Everysport. Cecilia Campbell, chief marketing officer at UR, says a second robot was built based on Mittmedia’s analysis, which found that there was huge interest in real estate and sold houses—a demand the company couldn’t match. In response, Mittmedia built a robot to collect real estate data from local communities and was able to produce about 480 pieces a week. “There was a big business incentive behind it,” says Campbell, “as well as being able to provide this kind of interesting information to local communities.”
A journalist’s work requires direct communication with people, in-depth research, and nuanced perspectives. Campbell says that journalists are overqualified to write short stories like sports recaps and individual house sales. By delegating this work to robots, journalists can invest their time in stories that offer more expertise, voices, and context, providing readers with news they want to read. Campbell said all senior management at UR have journalism backgrounds and share the goal to make “local publishing viable again.”
Campbell has more than 20 years of experience reporting on the news publishing industry internationally. When she joined UR in 2018, her first couple of years were spent educating people about automated content, not on UR’s necessarily, but the concept of articles written by robots more generally. She claims that since the market is more mature, people are starting to understand the value of content-as-service, noting that UR’s first set of clients were trailblazers. In 2022, Brunswick News debuted its use of UR’s technology to cover junior ice hockey games in New Brunswick.
More recently, AI engines are being trained to not only analyze data, but also to generate novel content from them. This is called generative AI, and, due to the rise of AI chatbots like ChatGPT, it has generated a lot of hype for its ability to answer a wide variety of questions with mostly clear, conversational, and accurate responses.
Give Me Your Future
Gonda says a “bold and unusual business decision” led to the creation of Sophi. Instead of replacing journalists, he argues that these automation tools allow for the reallocation of newsroom staff to more “journalistic processes,” such as assigning stories. Sophi now places more than 99 percent of the content on the Globe’s digital pages, which means less production work lands on editorial staff. Gonda argues that this is good news: “People can be redeployed or resources can be redeployed to creating more podcast producers or covering interesting topic areas.”
As for what happens next, this new methodology could be a bright new way to do journalism for many, or yet another disruption for the industry. Despite what anyone thinks, the future is now, and soon the answer will be clear.