USA Today Updates Every AI-Generated Sports Article to Correct “Errors”

After pressing pause on its messy foray into AI-generated content, publishing giant Gannett seems to be making use of the AI effort's downtime.

Gannett, which owns USA Today in addition to hundreds of local publications including The Arizona Republic, The Detroit Free Press, and The Tennessean, entered the internet's crosshairs last week when it was discovered that the publisher was quietly rolling out AI-generated high school sports stories in its regional newspapers.

Criticism abounded. The AI-spun sports blurbs, generated by a company dubbed Lede AI and designed to sum up the scores and basic events of any given high school match-up in varying Gannett-served regions, were terrible, repetitive, and sometimes borderline illegible, clearly written by something that didn't witness the event in question. In our reporting, we also found that some of the AI-generated nonsense was being published in USA Today proper, and not just on the websites of hyperlocal Gannett-owned newspapers.

In its attempt to save face, the publisher chose to "pause" its AI machine. But Gannett seems reluctant to completely pull the plug. Case in point: rather than simply scrubbing its many pages of Lede AI-generated content clean, the publisher appears to now be updating the automated sports blurbs to "correct errors in coding, programming or style."

This rewrite effort has included updating every AI-cheffed article that's appeared in USA Today. Take, for example, this article, which was first published by The Tennessean and later circulated in USA Today. As archived, the content originally read:

The Hardin County Tigers defeated the Memphis Business Execs 48-12 in a Tennessee high school football game on Friday. Hardin County scored early and often to roll over Memphis Business 48-12 in a Tennessee high school football matchup.

As it surely goes without saying, this silly AI-produced blurb was oddly repetitive. Now, however, the human-updated content reads simply:

The Hardin County Tigers defeated the Memphis Business Execs 48-12 in a Tennessee high school football game on Friday.

Still not terribly thrilling, though at least it's more coherent.

It's not just USA Today that's been making edits. Gannett appears to be rewriting automated content across the board, an effort that's included a rewrite to the original AI-generated article that launched the internet firestorm in the first place. The nail-biting AI write-up, titled "Westerville North escapes Westerville Central in thin win in Ohio high school football action" and published in the Ohio-based Columbus Dispatch on August 18, drew wide criticism for its poorly constructed, repetitive format and use of the hilariously empty description of the local football game as a "close encounter of the athletic kind."

That original article, which you can still see for yourself in this Twitter-formerly-X screenshot, read:

The Westerville North Warriors defeated the Westerville Central Warhawks 21-12 in an Ohio high school football game on Friday.

Westerville North edged Westerville Central 21-12 in a close encounter of the athletic kind at Westerville North High on Aug. 18 in Ohio football action. Westerville North opened with a 7-0 advantage over Westerville Central through the first quarter.

The Warhawks trimmed the margin to make it 7-6 at halftime.

Westerville North jumped to a 21-6 lead heading into the final quarter.

The Warriors chalked up this decision in spite of the Warhawks' spirited fourth-quarter performance.

Very bad! Now, though, the article is noticeably trimmed up around the edges, with the notorious "close encounter of the athletic kind" line nowhere to be found:

The Westerville North Warriors defeated the Westerville Central Warhawks 21-12 in an Ohio high school football game on Friday.

Westerville North opened with a 7-0 advantage over Westerville Central through the first quarter.

The Warhawks trimmed the margin to make it 7-6 at halftime.

Westerville North jumped to a 21-6 lead heading into the final quarter.

The Warriors chalked up this decision in spite of the Warhawks' spirited fourth-quarter performance.

Still bad, but not as bad. It's like sprinkling sugar on lemon — definitely not lemonade, and you'll still make a face when you taste it, but it's slightly more palatable nonetheless.

Similar updates have been made at other Gannett publications (although, for whatever reason, newspapers have chosen to hang onto this bizarre "Rip Van Winkle imitation" line that's appeared in many of its newspapers.)

To be clear, mistakes often happen in journalism, and when they do, they should always be corrected. But the rush to patch up these many dozens of AI-generated also seems to be a clear sign of a more concerning detail: that no human eyes seriously reviewed any of this content before it was published, a frustrating feature of the messy ordeal that appears to fly in the face of USA Today's own AI ethics guidelines.

"AI-generated content must be verified for accuracy and actuality," reads a small bullet point in the network's AI policy, "before being used in reporting."

And though it may well be true that none of the articles in question were expressly wrong about local sports scores, this footnote surely implies that human editors have reviewed all AI-generated material that winds up in the USA Today network's content before it's published.

Based on Gannett's many mistakes, and the subsequent effort to retroactively patch up the AI synopses, it's unlikely that any such human review was taking place.

Asked to comment on whether any of its AI-generated material had been vetted by a human before it was published, Gannett declined, instead offering the same statement it did when we interviewed one of its human sports writers, who called the AI content "embarrassing."

"In addition to adding hundreds of reporting jobs across the country, we have been experimenting with automation and AI to build tools for our journalists and add content for our readers," the company said. "We have paused the high school sports LedeAI experiment in all local markets where they were published and will continue to evaluate vendors as we refine processes to ensure all the news and information we provide meets the highest journalistic standards."

Sure, though we might suggest that the next time a major publisher is hoping to adhere to the "highest" of journalistic standards, they might avoid pressing auto-publish — and skipping any semblance of an editing process — on generative AI material in the first place. Just a thought.

More on AI and local sports journalism: USA Today and Many Other Newspapers Are Churning Out Terrible AI-Generated Sports Stories