Robot journalism is a growing factor in many newsrooms. One example of a news publisher that has used this strategy with great success is Bergens Tidende in Norway, which was awarded three INMA Global Media Awards silver places for its Home Sales Web site section, entirely populated with robot-generated content. 

Over the past year, Bergens Tidende has generated more than one million pageviews, and is now driving 5% of all BT’s article conversions of paying subscribers. How did they do it? By partnering with United Robots.

In a Webinar for INMA members, Cecilia Campbell, chief marketing officer at United Robots and blogger for INMA, explained that her company turns structured data into text, using AI and natural language generation, that is enriched with images, maps, graphics, metadata, and links.

“Then we send the content to the publisher, either into the CMS or directly to readers on [Web] sites, through apps, or push notifications,” Campbell said. “The vast majority of the articles we generate and send to publishers are automatically published with no tasks from the newsroom.”

In addition to Bergens Tidende, United Robots has also partnered with MittMedia, Aftonbladet, Schibsted, Gannett, and many others and produces automated content in nine languages.

“We know there’s concern in the newsroom that robo-journalism is technically difficult,” Campbell acknowledged. “That’s certainly true if you’re going to build your own robot or use a self-service, too. But we build the robot for you. We don’t sell tech; we build a service.”

Once the robot is built, the content begins being auto-generated and sent directly to the publisher.

Bergens Tidende has found great success with this robot-generated content, primarily in its real estate section. The core value of United Robots’ real estate offering is the single sale articles, Campbell said. This allows publishers to cover every house sale in every neighbourhood — content readers of hyper-local publishers love.

“The real estate content has a proven track record of driving subscription sales,” she said. Users purchase a subscription to view the real estate content and listings, which in turn drives up advertising sales and revenue.

In Scandinavia, real estate is the highest performing robo-content for digital subscription sales. Home sales verticals generate up to 20% higher CPMs due to premium inventory.

How to make automated content relevant to your readers

In short, the AI robot produces an article for each open real estate listing in the local and regional market — about 13,000 of them in the past year. This represents about one-quarter of all the BT content during that time.

These articles help BT reach an audience they would not have otherwise reached. The trick to this is to make the content relevant, Bergens Tidende Project Lead Jan Stian Vold told Webinar attendees.

“We treat it like any other journalism,” he said. “Don’t do it just because you can, or it will increase volume, but because it will bring real value to your readers.”

Volk said that in the beginning, robo-journalism was greeted with caution in the newsroom. They began slowly, automating just part of their content. Soon everyone embraced the model, from the top down.

“Real estate automation became the first product,” Volk told INMA members, launching in July 2020.

“It’s relatively easy to produce automated content,” he continued. “But the trick is to make it relevant. What we tried to do was reproduce, or imitate, all parts of the editorial process.”

One key is creating engaging headlines, paired with great photos. Volk called the real estate robo-content a “headline machine.” A set of rules is built into the robot to make sure the most interesting, engaging, high-quality, and relevant content is produced.

Another important aspect is to make sure the articles are presented as editorial and not confused with advertising.

“We take an editorial approach. We emphasise the journalistic value, which is important,” Volk said. “Buying or selling a house is one of the most important economic decisions you make. The importance of being informed in the real estate market benefits customers.”

He noted that automating this real estate content isn’t because journalists can’t do it, but because it is so time-consuming that it’s a very difficult area to cover well.

The result is extremely hyper-local content that area readers find of high value.

How to target readers

The next topic Volk discussed was how to make sure the right readers are able to find this robo-generated content. Geotargeting, to make sure readers get the local, relevant content on the home page of their community edition, is one way.

“This has improved the kick rate between 40 and 80%,” he said.

They also use the front page of the Web site to promote the best articles.

“Still, I think we can do more to make the service more user-friendly. A map solution is in the planning,” he said.

How to get the newsroom on board

BT’s automated real estate content has the support of the entire organisation, but Volk said getting there is an important part of the journey.

Some team members expressed concern that the content be obviously disclosed as robot-generated, which is the case. Each article is clearly designated that it has been automatically created by a robot.

Volk recounted a day that proved the value of the robo-journalism, when a billionaire sold a property in Bergen for one of the highest prices ever in the city. The robot picked up the story before anyone else.

“We have several other examples like this, where the robot is beating journalists to top stories,” he said. “This is of course a boost for the service.”

They have also recently started using United Robots to auto-generate annual reports.

The business model

“One important goal when we started out was to add a new revenue stream,” Volk said. “Like all media houses, Bergens Tidende is trying to become digitally sustainable.”

That’s going rather well, he reported. “We recently reached 50,000 digital subscribers, which has been a goal for us for many years.”

To secure further growth, however, they must keep adding new revenue streams. 

“As the digital subscribers grow, that ‘either/or’ dilemma between subscription revenues and ad revenues becomes less and less prominent. With more and more paying readers, we can aim for both sources and develop more strength. It’s a win-win situation.”

Volk shared some key numbers:

• Approximately 13,000 articles robo-generated since July 2020.

• 1,200 subscriptions sold.

• Real estate robot has produced one million clicks in one year.

• Annual report robot has generated 10,000 clicks per day since launching in late June 2021.

• Time spent and scroll depth is as good as with ordinary, non-robot articles.

As far as return on investment, the 1,000 digital subscriptions per year will generate approximately US$230,000, while advertising sales will bring in close to US$60,000.

“On top of that is the loyalty of readers already subscribing,” Volk added. “A service like this adds value.”

While the numbers aren’t “breathtaking,” Volk said, this is a new revenue stream is driving digital subscriptions and furthering Bergens Tidende on the path to digital sustainability.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .