Update 5 December 2024: More than a dozen new publishers in the UK, US, Spain, Japan and Latin America have signed up to Perplexity’s revenue-sharing programme, the AI company announced on 5 December.
The publishers include Adweek, The Independent, US local publishing owner Lee Enterprises, Los Angeles Times and World History Encyclopedia.
Also involved are Newspicks and Minkabu Infonoid, the first two Japanese news publishers to sign agreements with an AI company, Spanish-language publisher Prisa Media, Mexico News Daily, RTL Germany brands Stern and NTV, as well as independent brands Blavity, DPReview, Gear Patrol and Medialab.
Perplexity said the publishers will share in revenue generated from advertising when their content is referenced in AI-generated results.
They will also have access to its APIs and developer support to build features using its proprietary search technology, access to data and analytics to track trends and content performance, and receive free Perplexity Enterprise Pro for their staff for a year.
Perplexity also said it has appointed a head of publisher partnerships, Jessica Chan who previously built Linkedin’s content partner programmes, due to the level of interest it has received from news organisations.
Original story 30 July 2024: AI search chatbot Perplexity has revealed it will share revenue made from advertising in search results with the publishers whose content is referenced in them.
Time, Der Spiegel, Fortune, Entrepreneur, The Texas Tribune and WordPress owner Automattic are the initial partners of the scheme but Perplexity is urging other publishers to get in touch – even Forbes and Conde Nast, who have sent legal letters opposing the use of their copyrighted content in its system.
Chief business officer Dmitry Shevelenko told Press Gazette the programme was borne out of asking “what are the other necessary ingredients?” in the online ecosystem for Perplexity to be successful.
“And one of those, arguably the most important, is for there to be a thriving and vibrant set of business models for journalism.”
The Perplexity Publishers’ Program will allow publishers to earn a share of the revenue from interactions where their content is referenced after advertising is introduced to the “Keep exploring” feature, which offers suggested follow-up questions, in the next few months.
Publishers will also get:
- Access to analytics platform Scalepost.ai to see which of their articles show up frequently in Perplexity answers that get monetised
- Access to Perplexity’s APIs [application programming interfaces] so they can create their own custom answer engines for their websites
- Perplexity’s (normally paid-for) Enterprise Pro offering for all their employees for one year to try out the potential productivity gains.
Shevelenko said: “We don’t consume expressions of journalism, people don’t come to Perplexity to read the news. They come to ask questions and get answers to those questions. And what’s so powerful about journalism is it’s an incredibly rich source of facts.
“So we decided to embark on the revenue-share structure because, frankly, it’s something that Google never did. We’re a small, tiny start-up going up against these trillion-dollar companies and so the way we compete is not by just copying Google’s playbook – and I think a lot of publishers… wish they had had a different approach to Google from the outset.”
Time’s chief operating officer Mark Howard said they were “proud” to be a launch partner because Time is “committed to embracing innovative new technologies and platforms to advance our mission of providing trusted journalism for audiences around the world” and this will “continue to expand access to reliable information and engage audiences in new ways”. Time has similarly signed a multi-year “content deal and strategic partnership” with ChatGPT creator OpenAI.
How Perplexity works and why it’s different from ChatGPT
Perplexity, which was founded in 2022 and has raised more than $100m from investors including Amazon founder and Washington Post owner Jeff Bezos, works differently to, for example, ChatGPT which needs to train foundation models.
Perplexity has built its own web index with information about the content on various websites and doesn’t train its own foundation models. Shevelenko explained: “We only use web content in real time when a user has a question and we identify a website in our index that has information that’s responsive to that question, and we pull in that information or a summary of that information and we use that to generate an answer.”
According to Shevelenko, the average length of a question asked by a Perplexity user is ten words. “So when people come to Perplexity, they’re not asking general questions like ‘what’s the news today’? They’re asking very specific questions.”
He added that this also means they don’t see data indicating any “cannibalisation” of traffic that would otherwise go to news publishers. The idea that people would get all the answers they need without having to click through to publisher websites anymore is one of the big concerns about generative AI tools like Google AI Overviews.
Shevelenko added: “Whatever effect there may be, we’re more than mitigating it with the revenue share.”
Perplexity plan to introduce advertising
Perplexity’s “big business model”, as Shevelenko put it, is advertising despite it having a paid version. Most users only access the free offering.
The advertising that publishers will benefit from is not yet live but is currently intended to go live by the end of September.
It was first announced in April that Perplexity planned to add native advertising into related questions which feature links to sources, meaning brand-sponsored questions will appear alongside organic ones.
The advertising will be impressions-based, meaning Perplexity, and therefore participating publishers, will earn revenue every time a sponsored related question appears.
Other advertising formats, such as videos that play alongside a generated response, are also under consideration.
Shevelenko teased that publishers could in addition be “empowered” to sell Perplexity’s ad inventory so it is a “two-way relationship where they can actually monetise their experience working with brands…” The potential details of this are still being worked out.
They are also looking at co-bundling partnerships that could mean the paid Perplexity Pro comes with certain publisher subscriptions, and vice versa.
Shevelenko said of the intention behind these initiatives: “We need there to be a vibrant, thriving business model for journalism because there’s just no scenario where we win and journalism loses.
“Because if there isn’t this open publishing ecosystem and production of new facts, then the answer quality of Perplexity will start degrading and then nobody will use it.
“The incentives already are kind of aligned, and we’re just making that more explicit.”
Recent publisher criticism of Perplexity
The launch of the scheme comes after a flurry of critical news articles about Perplexity and its use of publisher content.
David Buttle, founder of media and technology consultancy DJB Strategies and former director of platform strategy and public affairs for the Financial Times, told Press Gazette that there is “limited goodwill for Perplexity left from media” and publishers have been “fighting back”.
Forbes and Conde Nast have both reportedly sent legal letters to the AI company alleging copyright infringement in how it uses their content.
Shevelenko told Press Gazette that Perplexity would still welcome both publishers to take part in the publisher programme.
“I think we strongly believe that what we do is within the boundaries of fair use, and there’s loads of legal precedent around it,” he said, in an echo of OpenAI’s core argument as it defends itself in a lawsuit from The New York Times, “but those same publishers – we’re happy to include them in the publisher programme and work with them.
“I think the fact that, in both cases, they leaked their supposed letters to the media before even sending them to us reveals that they’re playing not necessarily a game of trying to kind of collaborate and find solutions, but maybe other things… But yeah, we’re open to working with all publishers, including ones that have sent us letters.”
Intellectual property lawyer Bill Sigler of Fisch Sigler LLP based in Washington DC told Press Gazette: “There’s no magic legal pill currently available that addresses each side’s concerns. In the US, uncertainty around the creation and use of large language models will persist until our courts and elected leaders further define those rights.”
Forbes accused Perplexity of “ripping off” parts of its own stories and other publishers including Bloomberg and CNBC with “inadequate attribution”. Specifically, the issue was within new feature Perplexity Pages, which lets anyone create a “well-structured, beautifully formatted article” with the answers they receive from their queries.
Forbes editor and chief content officer Randall Lane wrote that the Pages version of Forbes reporting about former Google CEO Eric Schmidt’s “secretive drone project” featured “eerily similar wording, some entirely lifted fragments” and an illustration from a previous story about Schmidt. Perplexity also promoted the Page via a push notification and created an AI-generated podcast about it, the Youtube version of which ranked highly in Google.
“More egregiously, the post, which looked and read like a piece of journalism, didn’t mention Forbes at all, other than a line at the bottom of every few paragraphs that mentioned ‘sources,’ and a very small icon that looked to be the ‘F’ from the Forbes logo – if you squinted,” Lane added.
Shevelenko told Press Gazette that Pages is “an experimental product that, frankly, nobody really uses or looks at”. Regardless, he said, sources have now been made more prominent in line with the core answers product.
Wired, which is owned by Conde Nast, published an article in June headlined: “Perplexity Is A Bullshit Machine.” It alleged that in its own experimentation with the AI chatbot using prompts about the subjects of Wired reporting, Perplexity’s results at times summarised stories “inaccurately and with minimal attribution”, appearing “prone to bullshitting”.
Shevelenko told Press Gazette in response: “If you’re intentionally trying to manipulate an AI product that uses a large language model to get it to say things that don’t make sense, you can do that.
“You can write articles all day long about ‘look how stupid this AI is’ if you’re intentionally trying to extract those responses from it through putting prompts in that you know will do that… when people are using these things they’re using them because they’re trying to be more productive, spark their curiosity. They’re not trying to use them in an intentionally malicious way. And so they find a lot of utility and value in them.”
He added that hallucinations are “not a material problem” for Perplexity because “we don’t use large language models as a source of knowledge. We’re using them as the world’s best synthesiser and summariser of content”.
Wired also criticised Perplexity for apparently accessing websites that had opted out of allowing their sites from being crawled. The AI company has since blamed third-party crawlers that did not respect robots.txt coding and implemented “an important policy change”.
Shevelenko said they now “never… even if it’s through a third-party crawler” add “any partial or full text from a disallowed robots.txt site” to Perplexity’s index, although he added that this was not “in response” to Wired’s reporting but “just a natural evolution”.
“We still think there’s a balance there, where there are certain facts that live in the world that are useful to answering questions, and so there’s a summary of a piece of content that may be in our index, but we never look at the partial or full text content,” he added.
The criticism resulted in the announcement of the publisher programme being delayed. According to Shevelenko, it had been in the works since the start of the year and had been due to launch in June.
Perplexity to publishers: ‘The only real bar for us is high-quality content’
It was reported in April that Perplexity had reached 15 million monthly active users. Shevelenko declined to share a user figure but said: “We think of the world through the lens of questions that get asked on Perplexity and just in the last month, over 250 million questions were asked. To give you a sense of growth and scale, all of last year only 500 million questions were asked. So we’re definitely seeing some great growth in 2024.”
Perplexity will “keep adding” publishers to the programme, Shevelenko said: “This is just the start: ultimately anybody who produces high-quality content that gets used as a source input for answering questions by Perplexity, and if we’re monetising that through advertising, we think it’s fair that we share in that upside.”
He noted that both big and small publishers are welcome but that “obviously larger publishers tend to produce the types of content that is more authoritative and has more verified facts, so I think that they probably have more to gain in the sense that their sources would show up more frequently, and so they’d be eligible for more revenue share.
“But I think the only real bar for us is high-quality content. We obviously don’t want to encourage SEO spam and folks like that to be motivated.”
Asked for his takeaway message to publishers (who can contact Perplexity via publishers@perplexity.ai), Shevelenko said: “All this programme is doing is it’s letting us know your address so we can send you cheques, right? We’re in the business of sharing in the upside so they only have a lot to gain, and, from our perspective, nothing to lose.
“So we want to talk to them, and really partner deeply and help them use our technology in a profound way to improve their business and their products, and hopefully see a lot of value out of the revenue sharing coming from Perplexity.”
Email pged@pressgazette.co.uk to point out mistakes, provide story tips or send in a letter for publication on our "Letters Page" blog
