Ziff Davis v OpenAI: Implications for the Open Web w/ Marat Gaziev

Marat Gaziev (0:00)

I would say for the last 20 years, right, there's been this implicit relationship between search engines and search and media publishers or providers of information. That implicit relationship is like, we'll provide our content to you for free, essentially, content that we've spent a lot of resources, time and energy creating, and in return, you'll send us traffic, we'll monetize that traffic in ways that we can, which will then pay in turn for more content creation. So that implicit relationship has been there for the last 20 years.

Obviously, with AI and ChatGPT, that dynamic is changing. So that implicit relationship isn't as implicit anymore. So media publishers are looking for ways, what is the new economic model that's going to exist in the future that will essentially incentivize content creation?

Ross Hudgens (1:07)

Marat, great to have you. Yeah, we were at a dinner recently where we were talking search, and you're sharing some really interesting things about Ziff Davis, where you're at as VP of SEO and GEO.

Marat Gaziev (1:19)

That's right, yeah.

Ross Hudgens (1:22)

And IGN more specifically that you work on, that many people would know, maybe more people would know as a brand. And you're sharing a lot of interesting things. So I figured you're in Austin. It's a good enough reason to get you on, so I appreciate it.

Marat Gaziev (1:36)

Yeah, no, glad to be here. We're neighbors, so almost.

Ross Hudgens (1:40)

Yeah.

Marat Gaziev (1:41)

Same local area, but yeah, glad to be here. It is definitely an interesting time to be in search, especially at a large media company like Ziff Davis.

So there's a lot of moving parts and pieces and kind of this evolution of search that we're seeing. So yeah, happy to be here, happy to talk about any of the topics that you want to talk about.

Ross Hudgens (2:03)

So I mentioned IGN, could you quickly hit on maybe like three to four other properties Ziff Davis owns for Context People?

Marat Gaziev (2:10)

Yeah, I mean, Ziff Davis is a large media company, been around for almost 100 years, I would say. We operate about 45 different media brands, including PCMag, CNET, which is a recent acquisition, across different verticals, media verticals, I would say healthcare, gaming and entertainment, consumer tech and shopping.

We own RetailMeNot, offers.com, which are local Austin-based companies. I specifically work on or sit within the entertainment, the gaming and entertainment division, which consists of IGN, Humble Bundle, Eurogamer, and a few other properties.

Ross Hudgens (2:53)

Okay, nice. Yeah, that's good context.

One of the additional reasons I wanted to bring you on is that we're in a specific moment where your space, I mean, your space in particular has dynamics that I think are impacting. I mean, it's getting disrupted obviously, but there's also implications for what your space is doing for others in terms of mentions and citations and things like that, which we'll get into.

Marat Gaziev (3:19)

Yeah.

Ross Hudgens (3:20)

But for those unaware, one of the core things that we want to talk about is Ziff Davis sued OpenAI.

That was a pretty big moment. How long ago did that happen?

Marat Gaziev (3:37)

The lawsuit, we filed the official complaint a couple of years ago, I think two years ago or so. Yeah. We're currently in litigation.

Again, I'll just preface by saying I know as much about the lawsuit as anybody else, and you could certainly go online and read about our official complaint, look at the, I think Judge Stein had ruled, a district judge. He had allowed some of the complaints and before dismissed others. But yeah, so it was a very interesting moment. The lawsuit or the complaint, as I understand it, again, I'm not a lawyer, but from reading the complaint, the official documents, we're alleging that OpenAI had taken our content against DMCA and copyrighted content. And there's a couple of complaints there around trademark dilution, circumvention of technological measures, but essentially they took our content and they've used that content to train their GPT models. So more specifically, I believe it was GPT2. So of the data set that they used for training, I believe it was roughly around 19 million or so URLs that they used from the top 1,000 domains or so.

That data set, Ziff Davis owned and operated, properties consisted of about 0.8% of that training data set or about 196,000 web pages. From the data set that they made available, I think it was called WebText. Okay. So, and that content was taken, it was repurposed.

In some instances, it was distributed more broadly, it was CMI stripped out.

Ross Hudgens (5:37)