Prorata’s chief business officer has set out four things the AI start-up needs to prove to publishers as it builds up to a wider launch of its products.
Prorata, which has received investment from Mail owner DMG Media and signed almost 100 publisher agreements, promises to share revenue split proportionally by the amount of each creator’s content used to create each AI-generated answer.
The start-up, which has a team of 80 people globally, has created an AI answer engine called Gist, just launched its first AI answer widget for a publisher, and is building an AI advertising tool to ensure it brings in the revenue it promises to split with creators.
Prorata will keep 50% of the revenue made from advertising alongside its AI answers and the remaining 50% will be shared proportionally among the publishers whose content is used to create the response.
Prorata’s 100 publisher agreements give it access to more than 500 titles to create what Jansen described as “arguably the largest database of licensed content used for gen AI answers”.
It uses that database to generate answers that can appear anywhere “within the trusted content world” of its publisher partners who integrate a Gist widget.
Chief business officer Annelies Jansen told Press Gazette one of the major principles Prorata needs to prove is that “AI engagement on a decentralised level is something that users want” and would use across publisher websites, adding that it’s “not a replacement of the search button in the top right corner”.
“This is about bringing new behaviour to existing users, but also using a network to drive users across a network of quality journalism, quality content.”
The second is that it can create that network effect benefiting all the publishers involved equally via its model of attributing where each part of an AI-generated answer came from.
“In, let’s say mid next year, I want to be able to have a series of use cases that say, on average, publishers who’ve integrated Gist.ai have been reporting plus double-digit growth of engagement on their own media channels. That’s my goal.”
The percentage of content from each publisher is displayed to users through an attribution bar, which Jansen hopes could become an industry standard because society is “owed that transparency”.
The third – which Jansen said “for some people would be number one” – is for Prorata’s AI ad unit to outperform existing ad units.
“Because our case study is plus 20% more engagement and plus 50% more CPC [cost per click], that is incremental to the industry.”
And finally, Jansen said, is the need to prove to publishers that Prorata is “very easy” to work with making it “the easiest onboarding journey they’ve ever experienced with a tech partner”.
She promised a partner portal “that will be a destination in its own right, where our partners will be able to see not just what is the value of my content, but how’s it being used? Where is it coming from?…
“One of our goals is that in the editorial room you’ll have a new line on your Chartbeat which says traffic from AI and the average revenue per answer is up week on week, because it needs to be embedded in the workflow, and it needs to be super easy.”
Other AI publisher deals ‘not a systematic approach to bringing fair value’
Prorata was founded by Bill Gross, who is widely credited for the “pay-per-click” model of search advertising. Gross, Jansen explained, was inspired by reading about The New York Times suing ChatGPT creator OpenAI for copyright infringement in December 2023.
“He thought: ‘This is weird. How is taking an LLM to court or trying to strike a deal going to build a sustainable business model for the whole industry, for all content creators?’” Jansen said.
She continued: “He thought: ‘Well, this is not going to help 99.9% of all people out there, because a deal actually requires you to bring in your legal team, and it’s not a systematic approach to how you bring fair value to the content used in a gen AI answer, and that’s where the idea for Prorata came from.
“The notion that you share revenue with the content creators is not new in the digital age. We see it in Spotify, we see it with Youtube or see it with Apple News, but in gen AI it’s just harder. It’s harder because you can’t count. And so we developed a system to count…
“We believe attribution is necessary to count contribution of original content in a gen AI answer. That’s why attribution is the foundation of our business, to be able to count it so you can share it. And I would say that, in that, we are different from deal makers.”
Publisher AI answer engines
Gist was built as an answer engine as a proof of concept to “prove to the market that attribution can work”, Jansen said, but a lot of the focus is now on building up the tech on publishers’ own sites.
Adweek launched its AI “companion” answer agent specifically for Cannes Lions in June which was able to draw on the publisher’s own news updates about the festival as well as lifestyle and travel content from the likes of Complex, The Skift and Atlas Obscura to help attendees make the most of their time in France.
Jansen said publishers can display the Gist widget in multiple ways, whether as an “ask me anything”, an article summarisation or related questions, and in various places on an article page.
Jansen said the idea is that people will stay on the sites longer with this interactive experience and that it can “keep them away from” the big tech platforms.
“If your experience in a media channel is not fulfilling, you often go to Google and now OpenAI or ChatGPT, Perplexity – can we, by providing a product to the industry to collaborate, not collaboration through negotiation, but collaboration through product solution, can we actually create an additional layer and network that lives between my own media channel and the big tech general environment?”
Traffic versus engagement question for publishers
But fewer people are going directly to publisher websites in the first place. The latest Reuters Institute Digital News Report found that less than a quarter (22%) of 18 to 24-year-olds globally said news websites and apps were their main source of news compared to 44% relying on social media and video networks. Among 25 to 34-year-olds the split was 26% newsbrands and 38% social media.
Jansen said in response that although Prorata is only working in text currently, they are building the tech to work in video and audio. The company has signed agreements with the likes of Sky and Universal Music Group.
Jansen also said the aim is the widgets will also bring incremental traffic gains for publishers, who could choose whether they want to prioritise traffic or engagement.
“For some publishers, they actually realise that there is so much incremental usage of their content in answers and the monetisation through the ad brings a significant revenue stream that they are less focused on the click through from the answers… for those publishers who have a subscription product, then that click through to their own channel is much more important.”
Asked about the potential disadvantage of users clicking sources in the answer widget, thereby leaving the publisher site, Jansen said the key will be for product people at both Prorata and the publisher to work together to find a “healthy medium”.
“What we see from our launch last week – related questions at the bottom of an answer, not surprisingly, are a really good way to keep people on site because they’ll spend more time on the site and every time they click on the related prompts, a new page opens up, a new opportunity to keep people inside. That’s also why it is so important that Gist.ai in terms of traffic count, lives within the publisher environment.”
Jansen said that AI answer engines are less satisfying for the user when it is based on only one publisher’s content because “in general, their questions are broader than what the publisher actually has to offer, even if you have an archive going back 150 years”.
Instead Prorata is trying to encourage publishers to be open to collaboration. Under the agreements they have been signing, a Gist answer widget on one publisher’s website can show content sourced from hundreds of other providers.
For this reason, publishers signing agreements with Prorata must meet its content quality standards so media companies do not unknowingly show untrustworthy information as a result of their mutual association.
“We are not afraid to publish an answer with contribution from other people in the ecosystem,” Jansen said. “As you can tell, I’m trying to avoid the word competitor because I think these publishers are not competing with each other anymore. They’re actually competing with big tech.”
She also said: “Every publisher on their own trying to solve for this will make it much tougher to turn the tide.”
Prorata had originally planned to charge subscriptions for its products which would have provided the revenue to be split between publishers.
But it changed strategy, Jansen said, upon thinking “why don’t we just go back to what publishers really understand, which is advertising”.
Prorata AI ad unit: Currently building supply and demand
Prorata is building an agentic AI ad unit that creates ads from material uploaded by advertisers and that also decides how and where the ads should show up alongside relevant content.
Jansen said this means the adverts are created “on the fly in real time, which is highly contextual, because it has the semantic understanding of both sides of the spectrum”.
These adverts will be able to appear on every Gist answer. They are undergoing a pilot in the US currently and will roll out in the US and UK in September. DMG Media, Time, Fortune, Lee Enterprises, Adweek and The Arena Group are also testing the ads on their own platforms.
In the meantime, Jansen said, they are working on building supply and demand – speaking to both publishers and advertisers at the recent Cannes Lions festival for example – so they can get over the “cold start problem” and create an ad business that “generates substantial money for publishers”.
Why Prorata says it signs agreements, not deals
Jansen opposes the use of the word “deals” in relation to Prorata’s agreements with publishers, saying “a deal’s a negotiation” whereas Prorata, which does not pay any money upfront but pins everything on the revenue share, has the same terms for all its partners.
Prorata was a partner at last month’s PPA Awards in London where it had a full-page advert in the programme stating: “They scrape and steal. We credit and compensate.”
Jansen said they have been deciding internally “how aggressive should we be in highlighting that your content is being stolen every day?… With the acceleration of the growth of users for LLMs, and no real solution yet, to stop that scraping without proper compensation we’ve actually become a little bit more outspoken, I would say, in our messaging.”
She added: “I think it is important to wake up or continue to let the industry know that if we don’t do anything, it’s not going to go away. This is not one where you think it’s going to be solved by the Government, or you think it’s going to be solved by the regulator, and I as a publisher could just continue to go on and do what I do.
“This needs to be solved by awareness and education on how you can actually participate in the incremental engagements. And so we make it easy to work with us, it’s for the benefit of the publishing ecosystem. We’re here to bring more money to the publishers.”
Despite her opposition to the deals, Jansen said they have actually helped foster an understanding on two key points.
She said they “create awareness for new user engagement tools, and also create realisation that the deals are only there for the happy few. The deals are not licensing deals. The deals are, get out of jail deals…”
Many publishers signing deals with the likes of OpenAI and Perplexity receive access to their AI tech to try out building their own products.
Jansen said building an AI website chatbot often appeared to be the easiest thing to do with those tech credits and claimed this showed publishers that using only their own body of content is “often not enough”.
“So we’ve seen examples of publishers with deals who have added content to their corpus, who have gone out and said ‘I want to license your content’ or simply… have added Wikipedia.”
Input versus output
Something else Jansen said is often misunderstood is that Prorata’s licensing and attribution system relates only to output – not input for AI training and scraping which is opposed, when done without permission, by most publishers.
“What is not helpful is that usually the two use cases, input and output, through journalists are being put into one problem to solve for. And I think that’s not helpful for the regulators. It’s not helpful for anybody that needs to make a decision in the industry.
“Because if you think it’s one problem, then you’re also looking for one solution. And if you’re able to decouple them and say, okay, there is one problem to solve for which is my content has been stolen and used for training, that’s an input problem, it’s stolen. How do I get value for whatever it’s stolen?”
Email pged@pressgazette.co.uk to point out mistakes, provide story tips or send in a letter for publication on our "Letters Page" blog
