This new AI answer engine plans to pay media companies for their content

When ChatGPT launched two years ago, it blew open the doors to a new way to search for information online. Instead of sifting through a search engine’s list of links, users can now query a chatbot and get an actual answer. Today, AI titans like Google and OpenAI and upstarts like Perplexity are all working on their versions of AI overviews and answer engines—new ways to deliver facts in a post-search engine world. As those efforts evolve, publishers and copyright holders from news sites to music companies are demanding a share of the money AI search generates based on their content.

This summer, following accusations of plagiarism from Fortune, Time, and others, Perplexity announced that it planned to share ad revenue with the publishers of articles cited by the company’s answer engine once it begins adding ads to its platform (it’s targeting a Q4 launch). It’s a big step from a company that makes a product that, if adopted more widely, has the potential to reduce traffic to publishers’ websites by generating answers based on their content. Google, meanwhile, is rolling out ads to its AI overviews, but has said that it won’t share revenue with publishers that contribute to results.

One thing is clear: Determining how much to pay publishers for their contributions to “answers” is a complex calculation. That’s something Bill Gross—a serial entrepreneur who’s credited with developing the pay-per-click advertising tech that still undergirds traditional web search—wants to remedy with AI company ProRata. The company will debut its answer engine this fall, with the goal of making revenue splitting for publishers more widespread and effective.

ProRata recently closed a $25 million Series A funding round, with investors that include AI-focused venture capital firm Mayfield and Gross’s own tech incubator IdeaLab. ProRata’s answer engine will launch with content from 10 authors and more than 20 content partners, including include The Guardian, Universal Music Group, The Atlantic, and Business Insider and Politico parent company Axel Springer.

“There are other sites that crawl the web and give answers without attribution or sharing revenue,” says Gross, who serves as ProRata’s CEO. “I want to show that you get better quality when you pay the partners. I want to show the whole system working.”

Pro Rata’s approach to revenue-sharing

Similar to Perplexity, ProRata’s answer engine will cite its sources, but takes a slightly different approach, says Gross. Built on Meta’s Llama as its foundational large language model, ProRata’s search will only perform retrieval-augmented generation on content that it has licensed. ProRata’s answer engine also uses proprietary attribution algorithms designed to calculate how much any given publisher’s content contributed to an answer.

ProRata has a decidedly smaller knowledge pool to draw on than competing AI search engines that scan the entire Internet for content to inform an answer. “We won’t have the whole web,” Gross says. “But I contend that will be better because we won’t have every piece of random garbage you can crawl—garbage in equals garbage out.” The company says its answer engine will launch with more than 25 million articles, and a spokesperson claims that less than “one tenth of 1% of questions might not be answered.”

ProRata plans to generate revenue through ads and subscriptions. It will keep 50% of the revenue and split the remaining 50% among publishers based on the extent to which they contributed to the answer, as determined by the algorithm.

Publishers see promise

As AI-generated search takes off, publishers are increasingly wary of its potential to undermine their traffic and, in turn, ad revenue tied to site metrics. That’s why Nicholas Thompson, CEO of The Atlantic, says he’s taking a proactive approach to AI. 

The company inked a partnership with OpenAI in May to allow it to train its LLMs on Atlantic content. It will also be paid for its contribution to answers that appear in results from SearchGPT, OpenAI’s version of an answer engine, which is currently in alpha testing. (When The Atlantic announced the partnership, its senior editor Damon Beres called it “a devil’s bargain.”)

Thompson says that the OpenAI partnership is “primarily about bringing new readers to the Atlantic through AI search.” Working with ProRata, other the other hand, is “much more about encouraging a technological system that can reward copyright holders and media companies for content that is used in AI.”

Although generative AI isn’t a significant factor in the industry-wide traffic decline that publishers are currently experiencing, it’s likely only a matter of time. Thompson says he isn’t waiting to explore how publishers can partner with AI companies. “We don’t know who’s going to build it, but we do know that AI search is going to be an important part of the future internet,” he says. “We want to get fair value exchange and be able to partner with whoever’s going to lead that.”

Music companies on board

For the music industry, the AI boom and the rise of generative AI might feel like deja vu from the Napster era. But unlike Napster, which just shared files, AI large language models can take in lyrics, images, and other copyrighted material, then incorporate it into generative content.

“Early on, we discovered our images were used to train [AI image generator] Stable Diffusion,” says Chris Horton, SVP of strategic technology at Universal Music Group. He adds that evidence of copyrighted lyrics and songs have also appeared in generative AI tools. UMG has sued Anthropic for training its models on lyrics from UMG artists, and launched similar suits against music-generating tools Suno and Udio.

At launch, ProRata’s answer engine will be able to surface lyrics, images, and information about UMG artists, with plans to add music in future updates. Horton says that arriving at such a partnership was challenge, as labels and artists “need to know there’s something it for them—they’re getting paid. It’s going to take some kind of attribution in order for a payment mechanism to work.”

Horton says ProRata was one of the first companies to talk about precise attribution and revenue-sharing for answer engine source material. “In the early days, a lot of companies were saying ‘attribution is really hard, we can’t do it, it’s impossible,'” Horton says. “[ProRata’s] attribution means the money can be split up in a way that’s a lot more meaningful to the people contributing [their intellection property].”

Horton says as UMG explores AI tools for both fans and artists, it’s “spreading the bets around” on AI companies that will use its massive catalog ethically. He points to ProRata’s strategy and Gross’s track record as reasons UMG has thrown its weight behind the new company. “

We’re very focused on making sure that we’re positioning AI as an opportunity, not just a risk,” he says. “Bill Gross was one of the first people to figure how to monetize the internet—so maybe he’s got another shot with AI.”