ESPN and Tennis Channel are under fire for AI moves, even if they haven't created Skynet yet. ESPN and Tennis Channel are under fire for AI moves, even if they haven’t created Skynet yet.

“Something is better than nothing” has long been a truism.

But in the sports media landscape, that’s not always the case.

The latest examples revolve around separate ESPN and Tennis Channel artificial intelligence experiments that have taken a lot of blowback. First, there was ESPN’s announcement Thursday on plans to use generative AI recaps for “underserved” sports, including NWSL and PLL games:

Amazingly, the initial tweet there (since deleted) included an error in the example:

But well beyond that, this model of using AI to generate stories rather than just hiring freelance writers for these recaps received a lot of criticism.

Here’s some of that (language warning):

Another questionable use of AI announced Thursday came from Tennis Channel’s parent company, Sinclair. There, they talked up how they would use AI to provide Spanish-language translations of their coverage of the upcoming WTA Guadalajara Open Akron (in Zapopan, Mexico) for viewers in Spain:

That one got less noticed and commented on than the ESPN announcement, but it still drew some blowback:

Both cases fit a more significant trend of sports media companies using technology to replace human work, often producing problematic results. Generative AI frequently remains very poor at getting the details right, which notably happened with Gannett’s muchcriticized forays into this space last year and continues to happen with X/Twitter’s “Grok” AI-generated recaps of trending topics.

And yes, ESPN’s announcement here claimed, “each AI-generated recap will be reviewed by a human editor to ensure quality and accuracy,” but that also hasn’t worked in the past. If an editor was actually given sufficient time to review this content properly, they probably could write it and do a better job. However, the scale level here (the only reason anyone is even trying these approaches) only works with limited “review” time and effort for any individual story.

Similarly, the Tennis Channel AI translation “beta test” feels like it could quickly go wrong.

AI-powered translation can help give an overall sense of something in a different language when an actual translator isn’t available. Still, translation is an art, especially as language often comes with not explicitly literal sayings. (This is a big part of why interpreters and translators can be so important in sports interviews, with nuances of questions and answers often missed otherwise.)

Human agency is vital for conveying meaning and knowing what to avoid. And taking an AI translation to a broadcast comes with its risks. This is being presented as a finished product worthy of being on-air for viewers, so if it goes wrong, that’s problematic content to have aired. (And it’s problematic content to have aired with no human guardrail to stand in the way; it may be a beta test, and Sinclair says they will mark this as AI content following their company policies, but it’s still going to air.)

Both of these companies are trying to emphasize in their releases that this is new content they would not have otherwise had rather than a replacement for jobs. But while that can technically be true in terms of them not firing anyone to bring in AI, writing game recaps and offering commentary in a different language are both jobs traditionally held by humans.

Doing those jobs for less-covered leagues and events can be an essential experience and stepping stone for many pursuing higher-profile sports media jobs. Saying, “We won’t offer these jobs, but we’ll use AI to do them,” is a choice, and it’s being done more and more, including with the IBM/The Masters generative highlights AI commentary this year and with AI-generated articles on Sports Illustrated‘s site and G/O Media sites last year.

It should also be noted that AI is not free.

With any program like this, there’s investment in building the AI, the computer resources to run it, the power to run those computers, and the human executives who oversee it. And it’s debatable whether that’s cheaper than just paying freelancers to write game recaps or Spanish-language announcers to provide their tournament calls.

Interestingly, two of the three leagues covered here (the NWSL and WTA) are women’s sports leagues. Companies love to talk up their investment in women’s sports, especially around the growing ratings for those leagues, but that doesn’t always extend to hiring people. And these sports may be “underserved,” as ESPN puts it in their release.

Still, they could very easily solve that themselves with different staffing decisions and priorities rather than AI-generated recaps.

AI will actually be cheaper for companies if and when they can massively scale it. And many journalists and announcers should be concerned about these moves. It’s easy enough to start with “underserved” sports your company wasn’t providing recaps for or tournaments you weren’t offering a Spanish-language call for. But where that actually makes financial sense is when this extends to things that are more in demand, where more expensive writers and announcers are currently creating content, and where one day they may also be replaced with this approach, with companies then pointing to “Well, we did it in this other sport, and it worked.”

What’s interesting about the ESPN move is that this was anticipated and pushed back on almost a decade ago by an unlikely source: then-Dallas Mavericks majority owner Mark Cuban. Then, he pulled credentials for ESPN.com’s Marc Stein and Tim McMahon (and all ESPN.com writers) in a complaint about “computer-generated generic coverage.”

Of course, Cuban was wrong on many of the details there. He was complaining about game coverage written by humans but with The Associated Press rather than ESPN. And no high-profile company has yet tried to extend this to a high-profile league like the NBA; the AP did use automation for some minor league baseball and low-level college game stories at that point and has since expanded that a bit, but not to the NBA.

However, Cuban’s point looks better in retrospect with the continued expansion of AI-generated content. That hasn’t really come for the NBA yet, but the recent moves suggest it might not be too far off. The real takeaway from Cuban’s point then and from these current moves is that “some coverage” is not always better than “no coverage.”

With the ESPN move in particular, it’s not that no one was creating written game recaps of NWSL and PLL games—people, including independent reporters, bloggers, and staffers who work for team websites. And if ESPN truly cared about these sports, they could either hire some of those people (even just on freelance per-piece contracts) or pay those originating sites to license their pieces (as they do with the AP and other wire services).

Instead, ESPN is spending money on an AI solution to create recaps. Judging by all evidence to date, those recaps will likely have errors (and some won’t be caught by the editors here), and at the very least, will be missing authorial intent and tone. Those recaps will steal search traffic from sites that are trying and, simultaneously, will give ESPN a license to claim, “We’re covering these leagues!” with the implication that they’re doing much more than they are.

This is somewhat similar to a debate that’s playing out in California. There, a funding announcement for “journalism” actually wound up as a bill to send tens of millions of dollars to AI initiatives, which seem likely to replace many journalists. That’s received strong and deserved pushback from many, including the Media Guild of the West union, but it shows how eager many are to turn to technological “solutions” that don’t actually do what they claim rather than just paying reporters. And ESPN’s effort here seems to fit in along some of those lines.

There are arguments for some sports uses of AI, particularly for offering things that cannot actually be done by humans in any reasonable way. A recent example that saw some praise was NBCUniversal’s option of an AI version of Al Michaels to call Olympic highlights. Michaels signed off on this and got paid, so this wasn’t a lost job (and this was a truly-added feature for those seeking it out; it wasn’t the standard highlights calls used on NBC’s various broadcasts, which did come from live announcers), and this allowed for more particular calls in his voice than he was going to provide.

And AI Michaels isn’t necessarily all that different from long-existing things. Those include sports video games stitching together pre-recorded clips from prominent announcers to correspond to what happens (you can’t have Rece Davis calling each College Football 25 game, but you can pay him to put in 250 hours of work that then gets used for millions of games over the years) and various maps applications paying celebrities to use their voices.

For prominent figures like Michaels and Cris Collinsworth, it’s understandable why they see some potential benefits in AI; that lets them get paid for less work and works (at least to some extent; we’ll see how much attention this gets when it’s less of a novelty) because people are interested in their particular voice, even if it’s not being provided live. But licensed use of a celebrity’s voice is a long way from AI-generated coverage replacing less-famous human writers or announcers.

Similarly, it’s not hard to envision ESPN and/or Tennis Channel coming up with some at least somewhat reasonable uses of AI down the road. ESPN chairman Jimmy Pitaro has frequently talked about the importance of AI, including in the above release on this project. And something he’s often cited is the idea of a personalized SportsCenter (yes, it’s weird how much people at ESPN still talk about SportsCenter, even in a post-Norby era) for users of their eventual Flagship (full linear ESPN and more) direct-to-consumer app, featuring highlights personalized for them (including from less prominent things, with him citing “highlights from Ivy League lacrosse”).

There are ways some of that could maybe work. Automated clipping is still not significant at this point, but perhaps it will improve to where it can reliably get highlights from the thousands of ESPN+-only events no human is clipping. AI could then conceivably generate this personalized SportsCenter based on particular team selections. That would be a legitimately new service rather than a job replacement, and the high level of personalization planned means this wouldn’t likely be feasible to offer differently. But generative AI recaps are a much different story, with these being an attempt to compete with an existing product that humans can and are creating.

That goes to the more significant point here. The counterpoint to “Something is better than nothing” is “If you can’t say anything nice, don’t say anything at all.” For a long time, the argument in less-prominent sports (especially women’s) has been that they should be grateful for scraps. However, the massive and growing interest in women’s sports, including the NWSL and WTA, suggests that at least part of the past problem has been a lack of high-quality coverage and platform.

With these recaps in particular, it might have been better for ESPN to just avoid them entirely rather than put out low-quality AI-generated content that competes with actual coverage elsewhere. It certainly would have been better from a PR standpoint. They weren’t getting much flak for not offering NWSL recaps, but they’re certainly getting flak for commissioning computers rather than writers to produce those.

[ESPN Front Row Tennis Channel]

About Andrew Bucholtz

Andrew Bucholtz has been covering sports media for Awful Announcing since 2012. He is also a staff writer for The Comeback. His previous work includes time at Yahoo! Sports Canada and Black Press.