The language around AI “killing” search is beyond hyperbolic. Every week, articles declare that LLMs will replace Google, website traffic will all but disappear, and the open internet as we know it will collapse. But when you actually run the numbers, the entire narrative falls apart.
That’s not to say that we are not in choppy waters right now and that the open web is realigning to a new normal. But the dust hasn’t settled. There is still more to be developed and explored with LLMs before we can declare the “death” of this internet or that mode of information gathering.
Why do I believe this? I did some math to give myself comfort and remind myself that there are some durable models of the internet that I believe should persist even in this turmoil.
The Licensing Problem Nobody’s Talking About
OpenAI has signed a number of content deals with publishers like Axios, News Corp., Condé Nast, Hearst and Dotdash Meredith. But is that enough to support the model? Can licensing really support the data that LLMs need? I don’t think so.
According to the Common Crawl data set, roughly 33 million websites were scraped to train models like GPT. Conservative estimates suggest about 60% of domains — approximately 8 million websites — are actually used in ChatGPT’s training data (source). Additionally, OpenAI signed a deal with Reddit reportedly worth $60 million per year.
So, let’s build the model based on the above data. Let’s be extremely generous and assume that 99% of the internet wasn’t usable by OpenAI’s models and that they need to license 2% of the internet moving forward.
Even with that charitable assumption, we’re looking at 160,000 domains that OpenAI would want to use and license long-term.
Now let’s say they want to license those at 1% the cost of the Reddit deal or $600,000. That means that OpenAI and other models would need to invest $96,800,000,000 in licensing deals to achieve just 1% of what they achieved with Reddit. That doesn’t feel sustainable, even for the largest players in AI.
Even if they had the capital to execute something like this, how would they actually do it? Would they have a licensing team developing partnerships with 160,000 websites? Would they build a self-service portal that would evaluate the licensing value of each website? Both options seem fraught with scale issues and misaligned incentives.
The Quid Pro Quo Must Survive
This brings us to a fundamental truth about how the internet functions: it’s built on a quid pro quo. Content creators provide valuable information, and in exchange, aggregators (like search engines) provide traffic for the content creators to (hopefully) monetize in some way. This symbiotic relationship is the economic foundation of the open web of which LLMs depend on to succeed. The hope for monetization drives a large amount of the content on the internet.
LLMs have upset this model by scraping content and generating responses without sending traffic back to the sources. But here’s the thing: they can’t sustain that disruption indefinitely. Even Google’s lawyers accidentally acknowledged this reality in a recent court filing, stating that “the open internet is dying.”
If that’s the case, where will they get new model data? If the open internet dies, LLM progress dies with it. They need fresh, diverse, high-quality content to remain useful, and they can’t afford to license all of it. Which means some version of the traditional traffic-for-content exchange has to continue.
I don’t know exactly how far down the bar will be lowered. But it’s simply not possible for it to be zero and have LLMs survive.
What Actually Happens Next
The underlying infrastructure that LLMs use to gather and process information is fundamentally the same as what powers traditional search. So what we’re seeing isn’t the death of search, it’s simply evolution.
And evolution requires adaptation, not panic, and certainly not mourning.
The internet will find a new equilibrium. LLMs will need to drive more traffic to publishers in order to maintain their content pipeline. Publishers will learn to optimize for AI platforms alongside traditional search.
The quid pro quo will take on new forms. But it cannot disappear entirely. So when someone tells you search is dead, ask them to show you the math. Because the numbers tell a very different story than the headlines.
After all, would OpenAI be hiring a content strategist if the internet was going to die?
