Bill Gross wants to save media ... with AI

By Fred Vogelstein

I remember when Bill Gross launched GoTo.com back in early 1998. The idea seemed nuts. Search was broken. Spammers had polluted results. In many cases they’d become just a list of who’d paid the most for placement. Google wasn’t even a company until the end of that year. But GoTo’s solution - to just be up front about spammers’ deception - seemed completely wrong. “Pay for search? That’s making it worse, not better,” most in Silicon Valley said. A chunk of the audience actually hissed at Gross when he was done unveiling GoTo at the 1998 Ted Conference in Monterey, he said..

Today we know what happened instead: The idea behind GoTo changed the world. GoTo was the first meaningful pay-for-performance search ad company - and half of what became Google Adwords four years later. When Google married pay-for-performance ads with its superior search results, it revolutionized the modern internet. No one thought you could make a dime with traditional search until Google did this. Google today is worth more than $2 trillion because it did.

Gross should have made a zillion dollars from this, but as he’ll tell you himself, he screwed it up. The idea seemed so obvious to him that he didn’t immediately think to patent it. And by the time he’d realized his mistake, during his first conversations with investment bankers wanting to take GoTo public, pay-per-click, as it came to be known, was unpatentable. 

I thought I’d heard most of the stories about the early days of search. But listening to Gross explain this one made my eyes bulge. 

“The bankers said, ‘Where's all your IP?’ We said, ‘We don't have any IP.’ They go, ‘Well, didn't you patent this thing?’ I said, ‘No …  I thought it was obvious (and unpatentable). And they said, ‘Well, you gotta patent it.’ Well, it had been a year and seven days since I had shown it on stage. You can only file for things that have been shown publicly within one year. So I missed the deadline to file for a pay-per-click by seven days.” 

Gross and GoTo still successfully sued Google for patent infringement over five background pay-per-click patents, which Google settled for about $360 million right before it went public in 2004. He and his shareholders also sold GoTo, which had changed its name to Overture, to Yahoo for nearly $2 billion in 2003.  “Of course,” he says “the main thing (pay-per-click) would have been worth trillions. But who knew?”

I'm telling you all this because, after two decades launching and investing in more than 150 companies with his incubator Idealab, Gross is back with another either crazy or visionary search idea called ProRata. It’ll work just like a conventional AI search bot like ChatGPT or Perplexity. It’ll use Meta’s open source LLM.

But it will do something none of the others do: Pay content providers for being the sources of those answers. He’s got written commitments from nearly two dozen top publishers to access their entire archives plus enough verbal - soon to be written - agreements to more than double that number. Meta’s LLM will parse each question. And ProRata will then use its access to this giant archive of publishers to generate answers. He’s launching it in Beta to 10,000 users some time in the next month.

This is a huge idea, even if Gross himself can’t make it work. The internet revolution has radically altered the balance of power between the creator class and the people who distribute their work. A generation ago they controlled the whole widget, as Steve Jobs used to say. Content, production, and distribution all lived under one roof. The internet disaggregated all that. Super rich and powerful Silicon Valley companies like Google and Facebook became distributor kings while creators and publishers, despite their protests, became their supplicants. But if ProRata or something like it succeeds, it will shift that balance of power back in the direction of publishers and creators for the first time in a generation. 

It’s super relevant now because we are at an inflection point in how news, information and entertainment will be consumed online. Traditional search is dying. It’s being replaced by AI search bots. Google will deny that. But once you use something like ChatGPT 4.0, it’s obvious. Ask Google a question, and you still get ten blue links. Ask ChatGPT, and you increasingly get the correct answer. It’s what many of us have wanted from search engines all along. 

This means billions of dollars in content licensing are at stake. The AI companies believe, as technologists always have, that the law allows them to build their databases without paying for the content or paying very little. Publishers, meanwhile, see this as an existential battle. A generation ago they didn’t take Google seriously enough early enough when they could have demanded compensation for every search. They are determined not to let that happen again. The New York Times has sued OpenAI and Microsoft because of it. And there are dozens of other lawsuits like it pending.

As you might imagine, Gross is fast becoming publishers’ new best friend. “They (publishers) are seeing this movie again (with companies like OpenAI sucking their content into their AI databases without paying for it)  and they're saying, ‘Here’s someone who's been there who's on our side,’ ” Gross said. “So when they talk to me and I say, ‘I’ve got the development team, and I'm willing to share that code with you, and you can collaborate with us to make it work the way you want,’ it’s golden to them.”.

Where’s the money to pay publishers coming from? Gross’s plan is to run ads running next to his AI chatbot answers and/or convince users to pay ProRata for an ad free subscription. Publishers will get half that revenue. He said he hasn’t settled on what he’s going to name the chatbot or what he’s going to charge for a subscription. But he’s raised $25 million from investors to enable publishers to get paid while he grows that revenue stream.

It sounds straightforward, but it’s actually super complicated practically and technically.  ProRata’s answers won’t just cite the sources it’s used to generate each response, it’ll highlight the specific passages in those stories it used. Then it will decide how those sources should share credit for the answer. 

He showed me one demo answer showing CNN contributing 22 percent, the Atlantic contributing 20 percent and the Financial Times the rest. “So we found these three articles that were the best articles from our corpus to answer the question, we summarized it. Then these percentages show you the relative amount of their content we used in the answer. So they (our publishing partners) would get a report back that would show them that. Meanwhile … the user can see where the content came from. They can click through to the article. But (our sources) get paid even before the user clicks through. Meaning for the mere showing of that excerpt they earn money, which is really great. They really love that,  of course” 

The demo felt cleverly similar to the way Google search works, just with two enormous improvements: It answers your question instead of supplying a list of search results. And it shows you its work. So you see the specific passages in the sources ProRata used as well ProRata’s analysis of how much each source contributed. We know generally how Google’s search algorithm works. But it never lets anyone see exactly how it decides the order of its search results. That’s part of Google’s secret sauce.

What’s actually driving ProRata’s tech is essentially a derivative of anti plagiarism software, which more and more teachers, students and scholars are using to either catch cheaters or to bullet proof their work against charges of copying even unintentionally. “It's like you're trying to take an article … and say,  ‘Where is this plagiarized from and in what percentage?’ ” Gross said.  The big difference is that Gross has set up ProRata to do so-called fuzzy matches - a conceptual match - instead of exact matches the way plagiarism software works. 

None of this was possible technologically a generation ago. But more importantly, Gross says, it wouldn’t have helped much because the traditional search economy is a constant battle between companies like Google and an entire industry of search engine optimizers trying to game the system for their own benefit. There’s no place for a middleman like ProRata there.    “But when you search with an LLM, there is no Pagerank. There is no gaming. There is no URL. There's no trick. There's nothing you can do. The LLM is just deciding what's the best answer to this question. So this is sort of the Holy Grail of search. It really is. And as far as I can tell, it's un-gameable. There's nothing you can do because it's not relying on signals that you can play with.”  

Gross said he got the idea for ProRata after the New York Times sued OpenAI at the end of 2023.  “I got angry that (what OpenAI is doing) isn't fair,” he said.  Since then he’s hired nearly two dozen engineers and AI experts, in addition to raising $25 million. 

Perhaps most importantly, most of the big news and magazine publishers are lining up to partner. They include the Atlantic, Fortune, Time, and the Financial Times. But they also include authoritative topic specific publications like Science magazine, Nature, New Scientist, and Scientific American. Universal Music Group has agreed to make all 200 of their artists’ lyrics and cover art searchable.  More than a dozen top selling authors like Walter Isaacson and Tony Robbins have signed on too.  “FT CEO John RIding called me from some Greek island while on vacation,” Gross said. “This is just so top of mind for all of them.”

Gross’ ultimate goal is to become the middleman between publishers and the AI companies the same way that Spotify has become a middleman between listeners and artists/publishers, or YouTube has become a middleman between viewers and artists/publishers. “With Spotify the measurement is streams. With YouTube the measurement is views. For ProRata the measurement is concepts appearing in chat bot searches.” He obviously needs a crisper term for this.

Gross says he doesn’t care whether ProRata is doing all the work itself or whether it’s licensing its technology to other LLMs. “So we came up with a way to do the attribution. We're doing it for text, movies, images and music …. all LLM output. My goal is to license this attribution technology to everyone - it could be Dall-e, ChatGPT or whatever -  and let them do it (so that they can) fairly compensate (publishers too.)” 

How does he think that’s going to happen? “If all of a sudden the publishers who are blocking other LLMs from crawling their content,  now open their floodgates to me because I'm sharing revenue …”  I now have the leverage to get the LLMs to listen to my pitch. He said that since he plans license fees of only one to two percent of LLM revenues from his technology, that will be attractive to them. “I don't want this to be only for me. I want the whole industry to do this. I just want to be the third party attribution technology that you all use to figure out how to share the revenue.”

In fact, he’s trying to have the prospect of licensing ProRata’s technology drive settlements of the publisher lawsuits.  He told me, for example, that  NYT executives are working this angle hard. He says they’ve been asking OpenAI about attribution and payment technology like his all year. “OpenAI claims it can't be done. So now the New York Times can say, ‘Well, what about this guy right here? He's proof that it can be done,’ ” Gross said.

Sure, some publishers are cutting deals with OpenAI and others. No one knows how this battle is going to shake out, so publishers need to pursue revenue wherever it exists. But most of those deals just allow them non-exclusive access to train their LLMs. And none of them do what publishers really want long term: to get paid for their content based on some measurement of usage.

What if the AI companies just build their own versions of ProRata? Gross doesn’t think that’ll work well. “Publishers don't want the very company who stole from them to be the arbiter of the percentages (how much publishers receive). So there's a certain credibility that OpenAI (for example) gets by choosing an independent third party (like us) to do the percentages versus them doing it themselves.”

There would be some delicious symmetry if ProRata turned out to be as important to our futures as GoTo’s pay-per-click was a generation ago. But even Gross knows it’s way too soon to predict that. He knows as well as anyone that the entrepreneur with a great new idea is often not the entrepreneur who turns that idea into a sustainable enterprise. During the Internet Bubble 25 years ago Gross seemed to be starting a new company every week. But the meltdown hit him hard, and many of those companies didn’t make it. He also knows that at 66 years old he’s probably not the right person to run ProRata for the long term, even though when you talk to him he seems just as bursting with ideas as he did when I met him more than two decades ago. 

And, of course, his AI chatbot needs to be popular, which is hard to evaluate today. It won’t be out for a few more weeks. That alone is a big hill. Many have already developed ChatGPT or Perplexity habits. Gross knows he’ll also have to convince some of the LLMs to license his tech and, like ProRata, split the revenue with publishers 50/50. “But to convince them to split the revenue is not going to be trivial because they don't want to split the revenue right now,” Gross said. “So I need to show with what I launch in a few weeks that by doing this (my way) I get better content and thus deliver better answers.” 

And none of what Gross is doing will matter much if the various lawsuits against OpenAI and other AI companies get decided in their favor. It’s hard for me to see how a judge rules for them. But I’ve seen compelling arguments for that path. It’s a reminder of how early we are in figuring out what this new world is going to look like. 

It’s easy to dismiss any entrepreneur who tells you that he’s more interested in using his company to start a conversation than he is in using it to make money. But to me, anyway, it’s credible coming from Gross. He’s a lot closer to the end of a remarkable career than the beginning. “I'm hoping that by paving the way, taking some arrows in our back, we'll show the way that this can be done.  I would like our system to be the preeminent version of it. But some form of this has to happen for the sake of creativity and democracy. It has to happen.”

And if ProRata does catch on and becomes a big deal, rest assured that there is one mistake Gross isn’t going to make again. “I started ProRata on December 31 last year. I met with the patent attorney on January 2 (this year). And we filed (our patents) on January 4.”