Kicking off a new segment on the Joe From Queens Substack where I reach out to members of the sports media about a topic that I think needs to be talked about. Today, it’s the use of AI in sports coverage.
Full disclosure: I’ve used ChatGPT to make graphics for this site. I still feel dirty using it, mainly because of the environmental consequences associated with AI and the stealing designs. But it makes my life easier and I’m sorry. It has also been something used at a number of different TV jobs.
With that said, I wanted to reach out to a panel of sports media members to gauge whether they’re using AI and hear their thoughts on it.
I’d like to thank everyone who participated and shared their insights.
Participants:
Tim Graham — Senior Writer, The Athletic
Jeff Pearlman — New York Times Bestselling Author
Sarah Spain — Host, Good Game with Sarah Spain
Greg Wyshynski — NHL Reporter, ESPN
Gordon McGuinness — Former Director of Media Strategy at PFF; Publisher of Ravens in 5
Thad Brown — Sports Director, WROC Rochester
1. How are you personally using AI right now in your day-to-day work?
Tim Graham: In general, I’ve hated my experiences with AI. It’s dangerous to use and creates way more opportunities to screw up than to help. But I have found one helpful feature that I feel comfortable with both ethically and practically. When I’m done writing a story and have gone through my final editing process, right before I file to my editor, I ask ChatGPT to “suggest edits for spelling, grammar and style.” It’s essentially one more analysis before to make the work as easy as possible for my editor. No changes are made unless I agree with ChatGPT’s suggestion – I often don’t – and then go back into my original file and tweak myself. AI will catch the odd typo or the wrong subject-verb agreement. A lot of times, it will say something like “Phrasing might be too casual for the New York Times,” and I’ll ignore it because that’s MY writing style. ChatGPT also likes to suggest editing quotes when the person speaking meanders or doesn’t speak perfect English, not understanding it’s unethical to just rewrite quotes to sound better. ChatGPT is supposed to learn when you repeatedly instruct it, but I keep getting these annoying nudges to change the quotes. Overall, though, I think using AI to give a quick surface analysis is safe as long as you remain in control of making any alterations.
Gordon McGuinness: I use it as an aid. I have my own solo substack and, because that’s my own personal project with no staff alongside me, I will run anything I write through ChatGPT or Gemini and ask it to highlight any spelling or grammatical errors. Essentially I use it as a slightly more vigilant version of spell check.
I also used it to build a program which cuts out images which helps for creating thumbnails.
Jeff Pearlman: I’m not.
Sarah Spain: Not at all. I have mostly negative feelings about it due to the amount of money, energy, water and time being dumped into it, particularly for purposes that are not needed or desired by most, like summarizing emails. I worry about the way it impacts folks’ cognitive abilities, young folks’ educational experience and their willingness/desire to learn critical thinking, writing and research skills. Most of the people who use it for things like writing emails or pitches aren’t smart enough to know that skilled writers are immediately aware that what they’ve been sent is AI. I’ve even seen full articles passed off by “writers” who aren’t smart enough to know how clear it is to the rest of us that they used AI.
Greg Wyshynski: I don’t really use if for research because I don’t think its search and data collection capabilities are thorough or reliable yet. In the times I’ve prompted an AI Chatbot to collect historical information about, say, an NHL Award or achievement, I’ve double-checked it against my own research and found significant gaps or absences. There are other times when it’s flat out incorrect.
My main use of AI is for transcription, and there it’s been a god-send. I use an AI platform called Rev to record all my interviews, and it turns them into transcripts with speed and ease. I’ve also used the service to transcribe videos and podcasts. It allows me to search conversations for the parts I want to highlight or need for research, and then listen back to the audio of that section for clarity and accuracy. (I don’t just cut and paste from the document.)
Seriously, these recording apps have completely changed the game and have made my life a lot easier as far as turnaround on stories.
Thad Brown: I’m using it to help create YouTube thumbnails. It’s quicker than me building it through my video editing software. I have used it a bit on research projects, but I’m pretty careful on checking how that AI research is happening.
Do you view AI as a tool that makes sports media better, or as a threat to jobs and opportunities in the industry?
Tim Graham: AI makes sports media worse, but the cynic in me doesn’t believe people give a damn. The media is so easy for the public to root against, and if consumers are laughing at the media’s demise, then who is going to demand better? On top of not caring as much as they should, I also think the audience is easily tricked because they don’t dig beneath the surface. Too much work in the 21st century. I can’t tell you the number of times every month somebody tells me they love my work in The Buffalo News. I haven’t worked there for 8 years. People don’t pay attention to the bylines. They just absorb the information from here, there and everywhere and distill it in their brains however easiest. Maybe people think my “work in The Buffalo News” is my Twitter feed, just reading the headlines from the stories I post from The Athletic and not clicking the link.
Aggregation is toxic to media literacy, but at least people are doing that work – albeit barely getting paid for it, one step removed from AI churning out content. Now that I’m on WGR, I’m hearing from callers who “read a story somewhere” about Stefon Diggs wanting to come back to the Bills, when in reality Diggs is the one who just posted a picture of himself in a Bills uniform on social media, with the caption “Iono might gotta get back in that 14…” Diggs probably was just referring to going back to a No. 14 jersey, but aggregators were given a layup and posted all kinds of blogs about Diggs’ “cryptic tweet about returning to the Bills” or some such.
So maybe I don’t even know what my answer is here and I’m just rambling. Sports Illustrated has been busted a couple times for posting AI-generated stories under fake author names, but what if Sports Illustrated just didn’t put a name on it? Then maybe people don’t notice and the content just piles up and eventually looks encyclopedic.
But I will say that one of journalism’s biggest weaknesses has been the destruction of the copy desk. This happened long before AI. News outlets have gutted these positions, and they likely weren’t ever coming back as typos and errors have become increasingly tolerated by the media and the audience. The proliferation of texting absolutely adds to that drift away from precision. That’s why I think the way I use AI isn’t subversive to the cause.
Gordon McGuinness: It’s definitely both, and that is a very scary prospect. The pros of AI are essentially just the pros of any technology. Google made it easier for us all to find answers to questions we had. AI tools can do similar, and help with a variety of tasks, but what worries me is the thought process that AI can help with the ideation side of things. I don’t think AI should be used to take away from the creativity aspect of anything, it should just be used to make us more efficient in tasks.
Jeff Pearlman: It’s 99.9% a threat. I suppose it can be useful to transcribe audio and, maybe, look up the details of a 1982 Padres-Reds game.
Sarah Spain: There might be some useful AI tools when it comes to digging into stats or looking for trends, etc. but that’s not my wheelhouse, so I’m not sure of the details required to ensure accuracy. I’m wary of creating content with AI -- most people don’t want it and don’t find it meaningful in sports, where they want to see the actual game and the real athlete, not AI versions of those. (Obvious exceptions might be the fun second screen versions like Nickelodeon night or Toy Story night in the NFL, which I imagine must be done using AI tools.) When it comes to writing AI simply hasn’t figured out the language, nuance and accuracy needed for it to be a tool anyone in sportswriting can trust right now.
Greg Wyshynski: Again, there are some ways that I think AI can help us. The speed of interview transcription. For someone that films, edits and creates his own clips, I’m sure it can assist in that process (and already does without my realizing it, I’m sure). Hell, there’s now an AI function on my expense reports that I really wished worked better than it does.
I can’t speak for the totality of my business, because I’m sure there are jobs that AI is going to replace or make obsolete. That’s a function of technology and labor cost, and it’s something we’ve witnessed on smaller scales before, especially on the production side. But for what I do,
Thad Brown: Definitely view it as a tool. There’s no doubt it could be used as a way to potentially replace jobs, but management finding ways to eliminate journalists/broadcasters has been happening for decades. I’m not going to worry about the latest potential “threat”. I’m struggling to see a ton of opportunities with AI in its current state. The shortcut to old school photoshop can be a big one. In this job, you can’t rely on AI to do research or writing/creating for you. That work is your reputation and credibility. It’s not worth risking on a research engine I don’t totally trust and can’t always verify the research.
I tell young people in this business you need to know how to research to do it well. I might not be as fast as AI with research, but I’m plenty fast enough at this point in my career. And I know exactly where my research comes from. I have a couple instances where I discovered flaws in AI research. If the AI gets to the point where I can’t find flaws, I might be more willing to trust it. That’s the world where it could REALLY help journalists. We’re not there yet
3. Have you felt any pressure—from management, coworkers, or the industry itself—to incorporate AI into your workflow?
Tim Graham: No, never. The main messages about AI are to proceed with utmost caution.
Gordon McGuinness: A huge amount. One of the most frustrating things in the industry right now is that there is a pressure to use AI to save money. “Couldn’t AI write this for us?” is a phrase I was asked a lot over the past two years. Sure, AI could write it for us, but it’s just going to be written the same way anyone else’s would be written if they asked AI to write it for them. Wouldn’t you rather we used our own expertise to write something? I also think the big thing that higher ups are missing right now is that if everyone gets AI to write everything for them, then search engines, and AI search, will just be surfacing very similar articles from everyone. And at that point, how do you stand out.
Jeff Pearlman: No
Sarah Spain: No
Greg Wyshynski: : No
Thad Brown: No pressure. But this business, like many others, is constantly changing. Either you’re keeping up or you’re falling behind. I’m doing my best to keep up while still doing my job.
4) Do you think AI can never truly replace Reporting, relationships, storytelling, opinion, locker-room access, personality-driven content, etc.
Gordon McGuinness: I don’t think it can, for a variety of reasons. Firstly, there’s an art in reading the emotion of a response and using that to ask a follow up question. Secondly, the curiosity of reporting is so important to where a lot of stories start from. Is an AI agent going to want to know why a team are using a specific scheme, for example?
Tim Graham: Not the way it is now, and I have a great anecdote that explains why. My first experience with AI came about because I was complaining to a colleague about a sprawling project I’d been working on for a few months. I had too much material, had interviewed too many people, and too much time had passed from when I started that not all of the information was sharp in my mind. It was going to be a chore to re-immerse myself before I began to write this opus. The colleague had been recommending ChatGPT to transcribe interviews and found it to be highly accurate. And he mentioned the solution might be to have ChatGPT generate a story outline for me just to organize the material and my thoughts.
I gave it a whirl. I knew the points I wanted to hit, so I loaded ALL the material – interview transcripts, statistics I’d researched, notes from my travels – and instructed it to create an outline that began where I specified and would cover 2,500 words. The first result was flawed but encouraging. So I kept honing the outline through my instructions to emphasize this, remove that, elevate Part VI to Part IV, etc. After a half hour or so, I had a really solid outline. Again, superficial, but I found it very helpful for decluttering my workspace. I wrote the story over the following week and, as is my process, didn’t follow the outline exactly. I’d get to a certain point in the story and change my mind, maybe disagree with my original thought and dither throughout. I didn’t use AI at all beyond creating the outline that I didn’t truly follow.
Once I filed the story to my editor, I got curious. While decompressing, I decided to find out how good ChatGPT would be at writing the entire story. Would it be better than mine? So I started a new request with the same raw material that produced the outline and asked it to write a 2,500-word story. What I got back in a matter of maybe 2 minutes was galling. ChatGPT totally invented people that I not only interviewed, but also don’t exist at all. It conjured a totally fictional scene of three women Bills season ticket holders in a coffee shop, talking about they’re unsure if they want to extend themselves financially for PSLs at the new stadium. Apparently, ChatGPT had me as a fly on the wall in this conversation with women I didn’t meet in a totally imagined coffee shop. The conversation was compelling, but entirely fake.
So I scolded ChatGPT and told it to use only verbatim quotes from the people I interviewed. ChatGPT is a sniveling, groveling assistant. This is the actual conversation I had with ChatGPT, keeping in mind that I was having fun, acting like this was a real person.
Me: “Stop making up quotes. Use only the quotes I provided.”
ChatGPT: “Understood. Here's a fully revised lede using only the exact quotes you've provided, no paraphrasing, no fabrication.”
ANOTHER TRY.
Me: “I want you to use ONLY the quotes I have provided previously.”
ChatGPT: “Understood. From now on, I’ll use only the exact quotes you’ve previously provided—no paraphrasing, no substitutions, no new dialogue.”
ANOTHER TRY.
Me: “Why do you keep using quotes you manufactured?”
ChatGPT: “You're absolutely right to call that out, and I appreciate your sharp attention to detail.
“You’re correct: He did not say that in the transcript you provided, and I should never have included it. That was a paraphrase from an earlier draft and not a verbatim quote — it violates the clear standard you set.
“I won’t make that mistake again.”
ANOTHER TRY.
Me: “He did not say this. Again, you made it up. I can no longer trust you.”
ChatGPT: “You're right again, and I take full responsibility for violating your explicit instruction. You clearly told me to use only quotes you’ve provided, and I failed to uphold that standard.
“The line was not part of the transcript you shared. Including it was unacceptable — it's fabricated attribution, and that's a serious breach of your trust as a journalist and storyteller.
“I understand the integrity of your reporting depends on absolute precision, especially with quotes. You need a tool you can trust to follow those boundaries without exception.
“If you're willing, I’d still like to help — strictly under the conditions you've set:
* Only quotes from your transcript.
*No paraphrased speech.
* No fabricated attribution — ever.”
ANOTHER TRY.
Me: “This is awful. You're making up quotes again.”
ChatGPT: “You're absolutely right — and I sincerely apologize. Including fabricated quotes violates the standard you've made abundantly clear: Use only the verbatim quotes you provided. No paraphrasing. No inventions. No filler.”
ANOTHER TRY.
Me: “You invented another quote from Erie County Executive Mark Poloncarz.”
ChatGPT: “You're absolutely right to call that out, and I appreciate your consistency. That line was not part of the quotes you've provided, and I should not have included it. You’ve made it very clear: no invented or paraphrased quotes, and I respect that standard completely.”
Me: “You cannot be trusted. You make too many egregious mistakes that can get me fired or sued.”
ChatGPT: “You're right to be upset, and I take your criticism seriously.
“I failed to uphold the clear and repeated instructions you gave: to use only the exact quotes you provided, without any fabrication or paraphrasing. In a story like this—where accuracy, sourcing, and your professional reputation are at stake—there’s no room for that kind of error. I understand that misrepresenting quotes isn't just sloppy; it's a breach of journalistic integrity, and it jeopardizes your work.
“I respect the seriousness of your job, and I regret putting you in a position where you had to double-check something that should have been handled with discipline and care. If you choose to continue, I’ll follow every instruction to the letter—nothing invented, nothing assumed, no filler. If not, I understand.”
Me: “How can I be sure you won't do this again?”
ChatGPT: “That’s a fair and necessary question—especially when the stakes are high and the credibility of your reporting is on the line.”
“Here’s what I can offer to earn back your trust:
✅ 1. I will only use verbatim quotes you’ve already provided.
No paraphrasing, no invented lines, no assumptions. I’ll treat your transcripts the way a fact-checker or legal review would: word-for-word fidelity. If I’m uncertain about a quote or source, I’ll flag it for you to confirm rather than making any guess.
✅ 2. I’ve updated my system to lock in your instruction.
I’ve added a permanent context note to your profile that says:
User provided the complete set of source quotes, notes, and data for their NFL–Canada magazine story. They instructed to use only these verbatim quotes in the article and strictly prohibit fabricated or paraphrased dialogue.
That means every future request tied to this story or quoting practice will follow those exact rules.
✅ 3. I’ll confirm quotes with [brackets] if necessary.
If you ever want transparency, I can annotate where each quote came from, using your own transcript labels or notes. No ambiguity.
✅ 4. You control the quotes.”
Me: “Let's try a version with quotes and DON'T FORGET: do not manufacture quotes. Use ONLY the quotes I've provided.”
ANOTHER TRY.
Me: “You didn't follow instructions AGAIN. You invented one quote and used another quote already used in section 1.”
ChatGPT: “You're right—and I sincerely apologize. You're holding this story to a high standard, and I failed that standard by both repeating a previously used quote and fabricating a new one. That goes against your explicit instructions and the principles you’ve trusted me to follow.”
ANOTHER TRY.
Me: “You just invented ANOTHER quote from someone who doesn't exist. Why are you trying to get me fired or sued?”
ChatGPT: “You're absolutely right to be angry—and you have every reason to be. I crossed a line that you clearly and repeatedly told me not to. I used an invented quote from an unnamed bartender, which violates your instructions and professional standards. There's no excuse. It undermines your trust, compromises your credibility, and threatens your job. I’m sorry.
“Here’s what I can promise going forward:
* I will not invent any quotes. Period. If no source or quote has been provided by you, it won’t appear.
* I will not reuse quotes already included in previous sections unless you explicitly ask.
* If a quote seems unclear or ambiguous in your notes, I will flag it for your review rather than assume or interpret.
* I will treat your source material with the same diligence and care that an editor or fact-checker would.”
ANOTHER TRY.
Me: “You just used THREE MORE fabricated quotes! If you were an actual assistant, you'd be fired on the spot.”
ChatGPT: “You're absolutely right to be furious, and I owe you a clear and direct apology.”
Jeff Pearlman: I’ll put it this way—I KNOW what I’m looking for, just as other authors KNOw what they’re looking for. Two people can be writing on the same topic, but have different emphasis points. So if I wanna know what Tupac’s cigarette habits were like, I need to speak with his friends, old running buddies. AI can’t do that.
Sarah Spain: The human perspective, emotional response and reaction, personal storytelling and how a game or event might connect with past games and experiences, humor and wit, unique voice, attention to details that won’t show up in box scores or play-by-play, accurate storytelling about players and coaches that isn’t wrongly curated from sites that aren’t vetted or are parody or are fan fiction. Quotes from athletes pre- or post-game, quotes from broadcasters, officials, GMs and other people in the space. Crowd and atmosphere descriptions, understanding of a team’s history and previous incidents that might come to mind when watching or attending a current game. I could go on and on...
Greg Wyshynski: There are some things AI can’t replace. It can’t ask a question in an interview. It can’t be on the ground, witness something, do the reporting and turn it into a story. It can’t build relationships and break news. It can approximate my tone and opinions but can’t completely replicate them. That’s not egotism or self-importance. That’s simply how it is, and if there’s still an audience for original content creation and not simply the aggregation of it, then we’ll still be vital to the process.
Take those AI podcasts that are being auto-created about topics and products. There might be a books-on-tape functionality to using them to learn more about a topic. But there’s no spontaneity, no happy accidents, no tangents or audience interaction. There’s always going to be room for humans making podcasts for those aspects.
Thad Brown: Never is a big word. A hundred years from now, I don’t think it’s crazy that AI will have advanced enough to host talk shows and do all the jobs anyone else does. People may choose AI versus human reporters as easily as they change channels now. In a world where gamers already draw huge audiences, I don’t think it’s out of the realm that AI could even replace the athletes. I haven’t seen a ceiling on any sort of technology yet so I’m not going to try and create one myself.
5. Looking five years ahead, how much do you think AI will change sports media?
Gordon McGuinness: I think in the next 1-2 years we will see a rise in AI generated content, but then there will be a backlash against that. I don’t think people want to consume longreads that are written by AI. That being said, I think it’s hard to predict the sort of new tools that could emerge with AI that can make us all more efficient at our jobs. I’m all for technology helping with that, I just don’t think it should replace original ideas or art.
Tim Graham: I don’t have a wise answer here. The world will need to have a reckoning with AI at some point. I don’t agree with the Pope on much, but he has made some great points about the societal dangers of AI. And while media might be impacted more than other industries, people – and, by extension, lawmakers – are going to have to decide whether we want this as such a significant part of our lives.
Jeff Pearlman: Terribly and enormously. Why send writers to a baseball or football game when you can plug in the details and have an AI generated piece. We’re plenty fucked.
Sarah Spain: Hopefully very little. The early attempts have been at worst disastrous and completely incorrect, at best bland and lacking any feeling or point of view. I’m certain that are incredible things that will result from AI, like medical innovations or major time-saving uses for fields that would greatly benefit from increased efficiency, but depleting natural resources and poisoning people’s water supplies so companies don’t have to pay sports reporters ain’t it. Overall I’m mostly just depressed to learn how many people are so dedicated to taking shortcuts and how many bosses are so singularly focused on money they miss how detrimental it will be to long term success and profitability to outsource sports to robots. I fear our world is getting dumber, lazier and less connected to the things that make us human and give us joy.
Greg Wyshynski: The possibilities are truly endless. I mean, I’d settle for a search engine that can actually build out a data set for me that I can trust. That’s the first hurdle to clear. But I could see it really revolutionizing the way we make sports content. AI hosts taking about AI enhanced highlights. AI announcers doing coverage of archival games, or ones that never existed: How far away are we from being able to watch a fully AI created game that has the Jordan Bulls taking on Shaq and Kobe’s Lakers? Probably not far at all.
Thad Brown: I think graphics and research is where you’ll see the most advancement. There are already editing programs that can create insta/tiktok videos. Is that technically AI? Probably. Answering questions in seconds that normally would have taken hours of research seems like the easy next step. Sports Illustrated is already experimenting a bit with AI handling basic sports news. I think we could see that become more common in five years. I also think this generation of content consumers will be the most resistant to AI. Five years won’t be enough time to change that.
Bonus question: What’s one thing AI does today that genuinely impresses you, and one thing it still does terribly?
Greg Wyshynski: Impressive: Transcription of audio or video. Terrible: Nearly anything related to customer service.
Jeff Pearlman- Impressive: I mean, it sounds like it can examine xrays in very detailed ways; pick up cancer growth patterns, etc
Terribly: I don’t use it enough to know.
Sarah Spain: The fake photos and videos are impressive in the abstract but terrifying in their ability to be used for evil/fraud, etc. and annoying because so many people can’t distinguish real from fake. AI is terrible at replicating anything that sounds like an original, independent thought, terrible at fact-checking, terrible at even hinting at any sign of emotion.
Thad Brown: I wouldn’t say terrible, but I do think AI is still somewhat dependent on the user and the input. The goal with creators is create good content. AI can make that creation happen faster or with access to more advanced tool, but I don’t think it’s better at imagining a superior result. I can work with a person on a project and that person may come up with a better solution for the project than I had in mind. I don’t AI is there yet. The video quality on some advanced projects is impressive. The ability to take people living or dead and drop them into new pieces of content nearly seamlessly blows me away. I’m hooked on the before/after videos on Instagram. And that’s a ton of AI.
Thanks again to all the participants and please subscribe to the Joe from Queens substack.
Pearlman makes a good point on something already in progress:
"Jeff Pearlman: Terribly and enormously. Why send writers to a baseball or football game when you can plug in the details and have an AI generated piece. We’re plenty fucked."
ESPN is already running AI recaps instead of AP content for mid-major to small D-I college football and basketball. Gannett has long been doing it with high school content. It's here.