Everyone is wrong about AI Slop.
Right there, I said it. AI-generated content has become a mainstream topic (as evidenced by this John Oliver video about it), but the public discourse surrounding it is a bit detached from reality. In the next few months, I plan to write a series of posts that comment and add nuance to some of the underlying claims informing public discourse around AI-generated content. As I do so, I hope to shed light on the whole AI content creation ecosystem, which I have “delved deep into” in the last few weeks.
Today, let me take a stab at:
“AI-generated content is cheap to create.”
This is demonstrably false for video, which currently costs a significant amount (as of June 2025). For example, consider the very amusing cat olympics video:

Enable 3rd party cookies or use another browser
This has 34 seconds. As of the time of writing, access to the model (likely) used here (Veo 3) can be obtained either through a monthly subscription costing $250 or via the official API. The subscription provides 12,500 credits, and generating a single 8-second video costs 150 credits. This means that a video sequence costs three bucks. (Using the API costs twice as much as 0.75 cents per second of video).
The video above has six different scenes, each probably created independently. Meaning it costs $18 to make it. But hey, it is implausible that everything worked perfectly the first time. Let us conservatively assume that for each take included in the final cut, the creator made four videos, meaning that three videos did not make the final version. This means this video costs $72.
Let us assume a relatively high “Revenue Per Mille,” i.e., how much money TikTok pays you for 1,000 views. Specifically, let's assume someone can earn 70 cents per 1,000 views. To offset costs, they need their videos to reach a little over 100,000 people. Which, frankly, is relatively high. I don’t have the exact figures, but I’m pretty sure that the vast majority of videos achieve fewer than 100,000 views. I’d be surprised if the median person using Veo3 isn’t losing money.
Running on VC Money
This begs the question. If video is so expensive, why are we seeing so much AI-generated video everywhere? And I’d argue that what is going on here is that:
A bunch of people are indeed losing money doing AI-generated content.
Few people are making a significant amount of money from it.
The cost of producing AI-generated content is reduced by the fact that numerous startups are burning money to enable people to create content.
Remember when Uber just started, and it was incredibly cheap to Uber anywhere? I believe this is what is happening here. I’d argue that few of the AI content creation people actually use the top tools offered by Google or OpenAI. Indeed, if you click on one of the thousands of how-to generate AI tutorial videos, you will probably hear of a bunch of different AI generation platforms: Vadoo, Mirage, Leonardo AI.
Most of these platforms do not have an official API and operate on a “You get k credits a month” basis. These are probably AI startups operating at a loss. E.g., Leonardo AI offers Google’s Veo3 cheaper than Google:
Leonardo.Ai offers a lower cost of entry to access Veo 3, from just $10 USD* per month. Video generation with Veo 3 is more affordable on the Leonardo platform, with a lower equivalent dollar cost of total tokens per generation ($0.75 on Google’s platform vs. approximately $0.30 on Leonardo.Ai).
A Market Perspective
Okay, but let’s take a step back. What does it mean for something to be cheap? The cat Olympics video has 3 million views, meaning it might have earned someone $2,000 (assuming it is monetized). And frankly, video-generated content will only become cheaper, so it will probably cost only $0.70 to make one of those in a year or two. So why am I still saying “AI-generated content is not cheap”? Because creating (some types of) content on TikTok is already pretty cheap.
Enable 3rd party cookies or use another browser
Let’s contrast AI-generated video with another dominant format on TikTok: reacts. Videos with stitched reactions can be made in minutes on a phone, for free. These require zero API tokens, no model prompts, and no rendering time. And if you’re recycling existing media (which many viral TikToks do), your marginal production cost is effectively zero. Yet, you can still get “Internet famous” from doing reacts—get sponsorships, brand deals, make merch, etc. Khaby Lame, the world’s most popular TikTok, kinda did it.
Watching content on TikTok is akin to shopping in a supermarket. People choose to “buy” different things with their attention. Some of it is inexpensive to make and intended for quick consumption, such as a bag of chips. Others are more niche, complex, and costly to produce, such as a fancy artisanal cheese that takes weeks to make. There is a demand for random, mildly amusing content on TikTok. But will AI completely change people’s information needs? Probably not.
So what?
I don’t mean to say that AI-generated content won’t have an impact on the Web. However, I think that how they will change our information ecosystem is nuanced, and frankly, hard to predict. It does not suffice that AI-generated content is inexpensive relative to high-production-value content. The big question here is: if more “meaningless” AI-generated content is around, will people reduce the amount of “meaningful” content they consume? It is worth noting that meaningful and meaningless here are very hard to define. If you consider that only “hard news” type of information is meaningful, probably most of TikTok is pointless already.
To end things, I think it is also not reasonable to equate AI-generated content with meaningless, regardless of what you choose to define as such. Algorithmic feeds reward “the best creators,” and I think the reason that better AI-generated content does not yet exist is that it is very hard to produce more complex content with it (and very, very expensive). This may also change in the near future, which will be interesting to watch.