(Welcome to the Entertainment Strategy Guy, a newsletter on the entertainment industry and business strategy. I write a weekly Streaming Ratings Report and a bi-weekly strategy column, along with occasional deep dives into other topics, like today’s article. Please subscribe.)
Often, reading entertainment industry newsletter after entertainment industry newsletter, I feel whiplashed between two things:
- The claim that Hollywood uses more data more than ever.
- And reporters/analysts getting things completely wrong.
Just checking the news one morning a couple of weeks ago, the first paragraph of an article about the creator economy got four things wrong, including two things that I debunked in this very newsletter. Then I listened to a podcast where a reporter completely discounted Paramount+’s recent run of streaming hits. (Clearly, they’re not a reader!) I won’t even mention how much Formula 1 misinformation I heard last summer after Apple Studio’s F1 film came out…
Worse, everyone is more and more confident about everything, while being just as wrong or misinformed as everyone always has been. (I’m always hesitant to say that things are worse than ever…that’s a tough thing to prove.)
Sure, back in the day, between box office and Nielsen ratings, everyone knew a lot more about what movies and TV shows were working, but today, the news consumer has access to way, way, way more Hollywood news coverage than ever before. Weirdly, I wonder if access to more information has led to more misinformation overall. Or if social media leads to more herding of opinions. Or, as Kevin Drum used to write, the internet makes smart people smarter and everyone else dumber.
Maybe the real difference between data wonks and non-data wonks is that the data wonks realize how much they don’t know; everyone else finds tidbits, one-off examples or anecdotes that confirm their preconceived notions and stop there.
All of which is to say, it’s time for another edition of “What I Got Right, What I Got Wrong”. This go around, I really tried to focus on what I got wrong, which is often more useful than touting what I got right, even though it’s way more fulfilling.
But first, a new feature: feedback!
Feedback: More Taylor Swift Data Fallacy Examples!
After I introduced my concept of the “Taylor Swift Data Fallacy”, I got three excellent examples from three different fellow pundits/writers/analysts:
- Barbie (Popular toy goes to theaters), suggested by Emily Horgan. This isn’t to say that other toy properties won’t do well in theaters—they could!—but Barbie dolls have been incredibly popular for generations, arguably the number one female-coded toy since the 1950s, with no real slowdown at any point, including today. Let’s be honest: are you going to be at all surprised if other toy-based films don’t do nearly as well as Barbie?
- Pokémon Go (Popular TV show/video game turned into an augmented reality game), suggested by Doug Creutz. This was another excellent example. Frankly, I’ve always wondered why augmented reality games haven’t done better. This game might have been the perfect marriage of IP with AR and gameplay that no one else could replicate.
- KPop Demon Hunters (Popular straight-to-streaming film heads to theaters for its second window), suggested by Matthew Frank of The Ankler. I don’t know how I didn’t think of this example, but it’s perfect. I’m sure other straight-to-streaming films might head to theaters at some point in the future—and I’ll take what I can get to help the box office—but I doubt any will be as successful as KPop Demon Hunters, arguably the most successful Netflix film of all time. The only caveat here? I actually think Netflix could have pushed KPop even higher if they had worked with all the theaters and reported their box office. (Also, KPop is headed back to theaters for Halloween.)
I actually like feedback like this, where people are “Yes and-ing” what I wrote and expanding on the idea. That’s the internet at its absolutely best.
Keep the ideas flowing! You can find my email here, reach out on Notes, LinkedIn, Twitter and Bluesky.
Even More Feedback: Definitions
Since I released my dictionary, I’ve gotten suggestions for more definitions to add to my list, including defining “residuals” versus “royalties” versus “profit participation”, an excellent suggestion from Allan Jamieson, a clarification on ATL/BTL and consumer products from Emily Horgan, someone recommended adding “slop”, and a list of terms like “ARPU” from Taresh Manuel Mullick.
I’ve already started writing at least thirty more entries to write up, so hopefully I can get that updated soon.
WRONG – VOD Does Mean PVOD and TVOD
After the introduction to the Streaming Ratings Report, where I ranted about “VOD” meaning “POVD” or “TVOD”, friends of the newsletter, Sonny Bunch and Ben Dreyfuss, argued that, in terms of how customers use it, “VOD” means renting or buying a film.
I guess I should relent on this point, though I might point out a distinction between the way industry professionals use terminology and the average person, but I’m going to update that entry. (Though in my report, I’m still going to use “TVOD”.)
WRONG: Sports prices are STILL booming!
Earlier this year, as a number of sports media rights deals failed to close quickly, I wondered if we’d hit the bubble for sports valuations. After all, I’d predicted that very scenario in an article for The Ankler years ago.
Well, thanks to Paramount’s gigantic UFC deal, it’s safe to say that mega sports deals are alive and well. Plus, ESPN went on a buying spree. That said, two leagues still await their fate: MLB hasn’t closed its new deals—though rumors of an NBC deal abound—and neither has Formula 1—though Apple rumors also abound. If that latter deal goes through, it would show F1 is more interested in cash, not visibility.
I would still repeat what I wrote earlier this year, “In this world, where sports rights keep getting expensive, but customers have more options than ever, I’d expect smaller revenue and profits related to sports than before.”
RIGHT: Soundstages were a bubble! So Was Peak TV!
Revisiting my Ankler article on bubbles, wow, I made some very, very prescient calls, warning that soundstages were being overbuilt in LA. And that’s coming true!
Obviously, I totally nailed my call on “Peak TV” being a bubble…a year before the WGA/SAG-AFTRA strikes.
(As for music catalogue rights and celebrity production companies, I’ll mark both of those issues as TBD. I write a lot about music catalogue rights in these updates, and that area has slowed, but not much, and it might still be a bubble, so let’s check in next time.)
WRONG: The Disney Bundle is Dying?
Well, I wouldn’t say the Disney bundle is dead. After all, ESPN has a standalone app that costs, what, $40 a month? But Hulu and Disney+ will soon merge apps, a decision I get, but don’t fully love. So mark this as wrong.
WRONG…then Very, Very RIGHT: I’m Still Unimpressed With LLMs, and ChatGPT5 was a Huge Dud
After I published my last update on how my team and I are using LLMs/AI, I realized, a week later, that I wasn’t using the thinking model for my program. Yikes!!! And sure enough, when I started using the thinking model, I saw immediate improvements in my data collection, leading to the sinking feeling you get when you make a big, public mistake. I was ready to issue a huge mea culpa…
But those “improvements” didn’t last a day. After a few more prompts, all the gains disappeared.
So mea culpa retracted!
That was in July. Then in August, ChatGPT 5 came out and, if anything, it’s worse, since my LLM freezes constantly now. For almost every ongoing chat I have, if I ask a question, I have to refresh my browser. And I still can’t get LLMs to do incredibly basic tasks, like format links or find IMDb links or even count lists. My LLM’s web search abilities have seemingly eroded, at least if I provide a time frame for a specific subject.
Here’s a brand new, super fresh, very specific example (the sorts of details I often don’t get anywhere else): I started a new thread on my LLM, with a very specific prompt, specifying three times to only search the web for news stories on a specific topic within a specific time frame.
The first result was outside of the time frame. With a note that the link was outside of the time frame. And then typing my response, my browser already slowed down.
Am I the only person dealing with this? I can’t square the circle between the LLM/AI hype I read in the Hollywood press and how I’m unable to get LLMs to work well for me. I just read an article on ChatGPT 5 that argued everyone in town is using it now, saying, “Run it through GPT 5”. How can ChatGPT synthesize a full season’s worth of scripts when I can’t get my LLM to return links within a specific time frame?
But that article also doesn’t mention the critical reception of ChatGPT 5. It was a huge bust! Just read Gary Marcus, Cal Newport, Ed Zitron (of course), Émile P. Torres, and others on how it flopped. Jen Topping had a good roundup.
And it’s not just me. Evan Shapiro wrote an absolutely wonderful article about his absolutely awful experience with ChatGPT, which might be my favorite article of his of all time. Here’s another article on LLM’s mistakes. And I just wrote about how ChatGPT 5 screwed up a data analysis project for me.
To be clear, I’m not giving up on LLMs or AI. I’m going to keep experimenting and trying to improve the functionality. I’m going to switch browsers to see if I can get it to stop crashing. I’m probably going to switch models, to see if I can find something better. I’m going to work on developing more detailed prompts and starting new conversations.
I do wonder, at some point, though, if I’m just wasting my own time. On a few tasks, LLMs provide small incremental time gains, but I wonder if I stopped using it tomorrow, if I would lose anything. The only real benefit of LLMs is as a search engine, but as I wrote last time, that says more about Google and monopolies than anything else.
I’m very worried for many writers/pundits/analysts, the full-blown LLM converts, who say they’re “using it in all their workflows”. Like using LLMs to synthesize a page full of research. How their work isn’t being larded up with mistakes, errors and hallucinations, I don’t know.
Wrong: The LLM-Internet Doom Loop is Worse Than I Feared
Last time I wrote about LLMs, I wrote this:
“In other words, with the advent of summaries by AI, traffic to the websites hosting information has fallen. This has obvious implications for the quality of the information on the web writ large. If creators cannot be compensated for their work, they’ll stop creating! And then the “summaries” online will decrease in quality, in a reverse flywheel of doom.”
Turns out, I didn’t predict what would take the place of those summaries: articles written by LLMs. Just check out this image, which went viral:
(H/T Jen Topping and Julia Alexander)
Since LLMs “hallucinate” (or just make stuff up, often at a 20% rate), that error rate is now compounding. Already, we’re seeing LLMs now citing hallucinated lies in articles that were written by LLMs, as Kurzgesagt so excellently showcased in their most recent video:
So that’s the actual reverse flywheel of doom.
Quick Hits and Corrections
- RIGHT: Paramount/Skydance is on a buying spree. Earlier this summer, after the Paramount/Skydance merger closed, I recommended that the new company should, strategically, go on a buying spree, but I didn’t know it would actually happen! (Not every company takes my advice!) Well, between getting UFC and The Free Press, signing the Duffer Bros. and James Mangold to overall deals, distributing Legendary’s films, hiring Cindy Holland and Jeff Shell, and more, it’s safe to say that they’re on said buying spree!
- WRONG: Horror Chart Data. In my article on horror films last July, I initially messed up the numbers on horror films. The chart was accurate, but the labeled numbers were wrong.
- RIGHT: Mulaney’s talk show wasn’t a hit. Last year, I was skeptical that Netflix was going to “re-invent” late night with Everybody’s Live with John Mulaney. Sure enough, it didn’t, averaging 300K to 500K viewers an episode globally. (Like sitcoms, I have no idea how/when this genre transitions to streaming.)
- RIGHT: Amazon is shutting down Wondery. Man, what timing! Hulu’s Only Murders in the Building just parodied this very company this season! I’ve long been skeptical of Amazon’s approach to podcasting—I described their Wondery acquisition as “fine” and noted that they’re competing with themselves—so I’m not surprised at this.
- RIGHT: The Winner’s Curse in action. I just defined this word in my dictionary, and The Conversation published a terrific example of it in action
- WRONG: Are Broadway musicals down? I’ve been touting Broadway’s big year season all year, but Walt Hickey wrote this recently, citing a Bloomberg article: “Broadway musicals are not doing great, based on reports that of the 18 musicals that opened last year, none have yet actually turned a profit. The absence of any original musicals has been a nagging concern…” It’s worth tracking this in the year ahead.
- WRONG: Rick & Morty premiere. I mentioned that Rick & Morty didn’t appear on the charts, forgetting that the new seasons have a multi-month hold-back from streaming.
