web1 was ads and ecommerce — Google and Amazon.
web2 was SaaS — Salesforce and Netflix.
web3 is AI, tokenization, and crypto — but what’s the revenue model?
We already have the answer, it’s staring us in the face, we just have to adopt it.
AI Relies on Content as Raw Material
Content is the raw material that AI is built from. Content scrapers vacuum up every scrap of content they can get and use it to train AI.
Think about AI as being a factory full of machines. The models themselves are machines, while the product — the prompt responses, the stuff people actually want to get — is made from content ingested by the machinery.
The industry already spends $3b a year on content — raw material — expected to grow to $10b a year over the next decade. But so far, most content has been obtained “for free”, or at least, without the license of the owner or creator of the content.
Either the AI company scrapes data itself, or it obtains large data sets from companies that scrape data. There’s numerous marketplaces where sellers hawk large data sets that were questionably obtained.
Just last week, Reuters won a lawsuit against OpenAI for using its content to train without license. There’s at least 26 more lawsuits pending on the same theme.
If this continues, the precedent will be unavoidable — AI companies have to have a license use the content they want to train on. Which makes sense, because content is the raw material that AI requires in order for the machinery to produce something people want. In every other industry you have to buy your raw materials from a supplier to produce something, why not AI?
And AI doesn’t exist without content to train on, which puts AI companies in the position of any other factory — to operate legally, they’ll have to pay for the raw materials.
That means the $10b by 2030 market estimate for training content is probably wildly underestimated, since it was projected before it started to become clear that AI companies couldn’t just scrape content from every website on the planet and use it to train.
AI Companies Can’t Negotiate with Every Website
Most AI training data deals have been from one huge website, like reddit, Twitter, or Youtube, to a huge AI company, like OpenAI, Google, Anthropic, or Perplexity.
These deals are bespoke, human-negotiated, between groups of professional attorneys. That’s incredibly expensive and takes a long time.
But there’s 2,000,000 websites, thousands of newspapers, thousands of universities, thousands of fiction and non-fiction publishers, hundreds of scientific publishers, the list goes on and on.
A world-class AI model needs all of that content, or as much as it can possibly get, in order to be as knowledgable as possible, stay up to date, and be able to respond correctly to the largest number of prompts.
It’s just not possible for a large number of websites to negotiate with a large number of AI companies with human intermediaries.
This is where pricing and software automation come in to play.
Imagine how impossible a trip to the grocery store would be if we had to haggle with the grocer over every price for every item. Or how impossible the grocers’ job would be, to do the same thing with every single customer.
That’s why grocery stores set standardized pricing. Pay it or don’t, take it or leave it.
Now imagine before cash registers when a grocer had to manually track and tally every purchase. That’s where software automation comes in — these are the products, these are the prices, here’s what you pay. That’s a modern point-of-sale system.
Automating the purchase and sale of website content to AI companies is the future of AI access to training content.
But the internet isn’t set up that way — right now, there’s no “point of sale” system attached to websites.
Sure, some websites use paywalls or subscriptions, but that’s an “all you can eat” method that doesn’t work for AI.
And some websites have a shopping cart for physical goods, but the AI doesn’t want to own a pair of shoes, the AI wants to know what is the price of the shoes, who makes them, where are they from, what are they made of, and so on — data about the shoes, not the shoes themselves.
web3 Revenue is Monetizing AI Use of Content
Once we realize that AI relies on web content as raw material, that AI companies have to pay for what they use, and that neither the AI company nor the websites can afford to negotiate for access to every website individually, we start to see a new revenue method for web3.
An automated purchase-and-sale method that detects bots, puts a price on the content the bot requests, and takes payment. Only once payment is made does the website serve the content to the bot.
Just like ads and ecommerce dominated web1, and SaaS dominated web2, I believe that AI content licensing will be the dominant method for generating revenue for web3.
Let’s list a few assumptions.
- Your website serves 1m page in a given time period.
- 50% of traffic is bots.
- 1.5% of those bots are indexers and optimizers.
- AI overviews reduce your traffic by >80%.
- A page will make $0.00125 per ad per visitor.
- A page will have 5 ad placements, which rotate every 30 seconds.
- A visitor spends 5 minutes reading content.
- 5 ads x 10 placements x $0.00125 per ad per visitor = $0.0625 per pageview.
- You can convert 5% of bots to payors.
- You charge bots 5x as much as you get per pageview because they reduce traffic by 80%.
- That’s $7,695.31 in new revenue across that time period.
That’s only considering the up-front revenue. There’s too many variables to produce a specific number. The point is to say that these are not marginal figures.
These realizations are what led me to start building robots.nxt, so that we can help websites protect their content and make AI companies pay them for its use.
If you want to see how your site can make money by selling content to AI, sign up for a robots.nxt account and the onboarding flow will walk you through a revenue estimator.