Nieman Foundation at Harvard
PressPad, an attempt to bring some class diversity to posh British journalism, is shutting down
ABOUT                    SUBSCRIBE
May 6, 2019, 10:21 a.m.

Why Richland Source built a system for automating high school sports articles (and stopped selling apparel)

“We spent six to eight weeks where a human being sat and edited every single piece before we felt comfortable that we could run this.”

In the robots-are-coming-for-journalism world, it’s usually big news organizations like The Washington Post or The Associated Press experimenting with automating their articles. Now Richland Source, the six-year-old digital news outlet in Mansfield, Ohio, is stepping up to the plate.

With Abundat, an artificial intelligence company that shares its coworking space (that’s how they met), Richland Source is now automating articles from the results of Ohio high schools’ sporting events. The details are exactly what the parents and fans want, managing editor Larry Phillips said.

“This covers the nuts and bolts of what the reader expects from their local news outlet. This frees people up to do more involved journalism, more of things they got into the business to do,” he said.

Richland Source built Lede AI over the past year with Abundat’s help, though the news outlet fully owns the product and used Abundat as a contractor. The process wasn’t pretty — early on, all five of Richland Source’s full-time writers and some helpers spent a Friday night translating Scorestream game data from Lede AI’s early-days textfile output to articles in the site’s CMS, and Phillips’ staff wrote hundreds of scenarios — but the team got to a point where it’s now comfortable allowing Lede AI to publish late-night games while they sleep.

“We spent six to eight weeks where a human being sat and edited every single piece before we felt comfortable that we could run this,” Jay Allred, Richland Source’s president, said.

Now, after completing a beta phase with seven other news organizations (which Richland Source declined to name) and over 20,000 articles published with zero inaccuracies, the team is trying to get other newsrooms onboard.

What do these articles actually look like? Often, just a headline, “Sports Desk” byline, a sentence, and a bunch of ads. (There’s no mention of the software or robo-writing on the articles themselves, but Allred and Phillips pointed to a featured article Richland Source published last week explaining Lede Ai.) Here are some examples, with screenshots of the shorter ones:

  1. Here’s one highlighted in Lede Ai’s whitepaper; it was the longest one I saw:

    Massillon Washington could use an Emily Post tutorial in manners, but its competitive spirit was fine-tuned while punishing Wadsworth 41-19 in Division II Ohio high school football action on Friday night.

    The Tigers opened with a 7-0 advantage over the Grizzlies through the first quarter. Massillon Washington’s offense darted to a 24-10 lead over Wadsworth at halftime. The Tigers carried a 27-12 lead into the fourth quarter.

    This marked the Grizzlies first loss of the season, as they completed a 12-1 campaign. Massillon sports a 13-0 mark heading to the state semifinals.

    The OHSAA releases the state semifinal pairings and locations on Sunday.

  2. Here’s a baseball game (or I assume it is, based on the baseball tag at the bottom):

  3. A “fast pace” and a quick read:

  4. And lastly:

The Washington Post has already followed a similar path with automating high school football results using its AI Heliograf (as well as smaller election results and Olympics alerts), and the Associated Press has automatically-generated articles about minor league baseball results and MLB previews.

But Richland Source is trying to turn Lede AI into its own mini-version of the Post’s Arc Publishing, building out this software as a revenue source. That’s in addition to its ads and revamped membership program. (It sold off the local-pride Made in the 419 apparel business — “We liked where we lived and the community vibe that had but we discovered pretty quickly we weren’t effective retailers,” Allred said — and has put the popcorn bag partnerships on hold for now.)

Allred said the team has considered automating other kinds of stories, like real estate transfers or marriage and death certificates, but reverted to sports because of the existing relationship with Scorestream. Sports stories are more likely to get search traffic, too — last night’s high school football game probably has a higher chance of being Googled than, well, that person from church you haven’t seen in a while.

“Let’s make something that creates opportunity for newsrooms and small local companies to dominate a market,” Allred said.

And from the editorial perspective: “It gives us the superpower element that we can cover the entire state. If a reader is looking for a football score, Richland Source is going to be there first and accurately,” Phillips said.

The team might try those other topic areas at some point. First, though, Allred and Abundat’s founder Evan Ryan are working on a way for Lede AI to flag trends across the high schools that one of Phillip’s sportswriters could then pick up and write an analysis on. (“We don’t have an analytics department. You’re talking to the analytics department,” Allred said.)

In the years since we last wrote about it, Richland Source has expanded to two more counties, and is trying to boost its membership bases to cover the two new reporters. The reader revenue quest has reshaped the for-profit site’s approach, with Hearken-driven stories and a smoother process to actually give the site your money. (Richland Source also recently participated in Facebook’s membership accelerator for local news sites.) The organization also introduced human-curated-and-written newsletters and events for members. So far, membership has doubled in the past 90 days and is on track to double again in the coming 90, Allred said.

Richland Source isn’t yet profitable, but Lede AI should nudge it closer. The whitepaper on the software’s site includes a breakdown of the ad sales from its own articles (Allred is selling Lede AI at $200/month and $4/article if the article gets 10 or more pageviews in the first 30 days since publication):

During the 2018 HS football season, Richland Source published:

  1. 1,866 total articles
  2. 148,700 total page views.
  3. 1,166 had greater than 10 views and were invoiced
  4. 700 had less than 10 views and were free
  5. They served three programmatic display positions on each article @ 2.00 CPM

Here’s how that broke down.

  1. Programmatic ad revenues: $892.20
  2. Expenses: $4,664 ($4.00/article * 1166 articles)
  3. Income: -$3,771.80 (and we covered the whole state)

Note: we did not do the following:

  1. Cut any freelance costs
  2. Get any outside sponsorships
  3. Do any article promotion, paid or even on our social media for free.

“We know we built a tool we needed,” Allred said. “Once we figured out the tool worked, we thought maybe another local newsroom might like this tool, too.”

POSTED     May 6, 2019, 10:21 a.m.
Show tags
Join the 60,000 who get the freshest future-of-journalism news in our daily email.
PressPad, an attempt to bring some class diversity to posh British journalism, is shutting down
“While there is even more need for this intervention than when we began the project, the initiative needs more resources than the current team can provide.”
Is the Texas Tribune an example or an exception? A conversation with Evan Smith about earned income
“I think risk aversion is the thing that’s killing our business right now.”
The California Journalism Preservation Act would do more harm than good. Here’s how the state might better help news
“If there are resources to be put to work, we must ask where those resources should come from, who should receive them, and on what basis they should be distributed.”