Perplexity is here to answer your questions.
Answering questions requires good information, though, and providing answers is easier if the people with the information actually want to give it to you. That’s a super-simple way of describing why Perplexity — the “answer engine” AI startup that pairs generative summaries with search — this week launched the Perplexity Publishers’ Program. By sharing advertising revenue with partners, Perplexity hopes to create better incentives for them to allow access to their content.
To unpack the new program and what it means to the media business, I spoke to Dmitry Shevelenko, the Chief Business Officer of Perplexity. If you're a cynic, you might think the move is purely defensive — that Perplexity is doing this so it doesn’t get sued. My takeaway, though, after talking to Shevelenko, is that Perplexity recognizes it needs good facts and good journalism to help fuel its so-called "answer engine, and that it wants to find a way to keep those coming.
But yes, he and I do talk about those notorious accusations about plagiarism from Forbes and Wired, what's the right way to attribute original reporting in an AI summary, and yes — that OTHER AI search engine that was just announced by a competitor you might know called OpenAI.
I learned a lot from our conversation, and I hope you do too. Don't forget to listen to the end where we talk about how publishers can create their own "mini versions" of Perplexity on their site, and why the hell it's called Perplexity in the first place. As ever, if you enjoyed this conversation, it would be great if you could follow the show on Substack, Apple Podcasts, Spotify, or any other podcast app, really. Also, I’d appreciate it if you’d leave a rating or review — it really does help the show. And if you’re on YouTube, please like the video and subscribe to the channel. Much obliged.
Transcript
Pete Pachal: Dmitry, thank you for stopping by The Media Copilot.
Dmitry Shevelenko: Awesome. Excited to be here, looking forward to chatting.
So I know you guys have some news. And you recently announced your publisher program, I want to get into all that. And I want to get to that pretty quickly. But first, AI, and even the companies in it are all still pretty new. And we're all still wrapping our heads around things. And I'd really like to level set with you just a little bit on Perplexity itself, if you don't mind. And, you know, just sort of bring everyone up to speed on what we're even talking about here. So, number one, how would you describe a perplexity as well, as well as its purpose? What is Perplexity is purpose?
Yeah. So the purpose of perplexity is to answer people's questions. Perplexity, I like to talk about it as an AI answer engine. And our founding insight. And the thing that defined perplexity from the get go, was not to use large language models or AI as a source of knowledge. But instead to use the real time internet and all the great high quality sources on it as the input, you know, where the information exists, and to use large language models and AI is a really good summarizer and synthesizer of that knowledge. And so it's really this marriage of a search engine and AI to do that synthesis that makes Perplexity different and unique from the outset and something that we're focused on. But, you know, you should come to perplexity whenever you have a question that, you know, you know, the more complicated, the better. And will, you know, the site will try to answer it for you.
And it seems to me that broadly, when someone wants a piece of information or some set of information, and the current way we do that, obviously, is with search online most of the time, and perplexity is sort of finding ways to remove friction within that process generalized.
I think it's, there's definitely some removal of friction. But But honestly, what's really driving the behavior is, is growing the curiosity that people bring to the internet, because you're getting answers to questions that wouldn't have made sense to ask Google in the first place, right? Like the one way to think about this is the median length of a query on perplexity is 10 words. When's the last time you put 10 words in? into a Google search box. And yeah, from what we've seen the median query length on Google is two or three words, right? So people are using. I mean, we're all about the perplexity versus Google competition. And it's fun kind of, you know, social media jousting. But the reality is actually a bit more nuanced and interesting, which is worship growing the pie, like people are asking questions that they never asked before on the internet, because you can now use natural language questions and natural language answers. And that's, that's kind of opening up the aperture for curiosity.
That's, that's really interesting. I want to get into habits and what search has done and what maybe what you're alluding to that perplexity is either undoing or maybe augmenting. But I do want to get your news. But before I even get to that, I have one more question, which is arguably just as important. What is the business model? At its most basic? Again, we'll get to the news in a minute, but how does perplexity make money or even plan to make money? You know, I realized there's runway here because you're a fairly new company. But I really want to understand the framework. What is the vision?
So the first monetization product we had is our perplexity pro subscription, which has access to a more advanced version of our search capability called Pro search, lets you attach files, lets you do things like voice to voice ask questions with your voice, get an answer with your voice. And, and so that is a $20 month subscription for consumers. And that started growing nicely. But But really, we launched that not because we thought that was going to be our long term model. But it was a nice validation that people found perplexity useful enough that they would be willing to pay for it, right? That's kind of a classic VC litmus test of like, Hey, is your product good enough that somebody would want to pay for it? And turns out, the answer was, yes. We've also in April launch perplexity enterprise Pro, which is a version of perplexity Pro that has the security and control features that companies want. So they can make it available to their employees, because perplexity one of its superpowers is is driving productivity for folks. And a lot of people were already using perplexity as Shadow IT, where they're using it for work in an unsanctioned way. But the long term business model for perplexity is advertising. And, you know, the vast majority of our users are free users. And and we think that the thing that is powerful though, is when you start with two revenue, streams, subscriptions and advertising, it doesn't force a race to the bottom on the type of advertising that you have to do, right. And so we don't, we don't want to be in a position where the advertising experience in perplexity, like the tracks material value from the user. And by having that diversification, we're able to kind of balance the those priorities and kind of, you know, it lets us build higher quality advertising. And so, you know, part of the announcement is like, yeah, we are turning on advertising and perplexity in this coming quarter. And that's how we'll be revenue sharing with publishers.
Yeah, that's what I wanted to get to next. So. So let's get to the news. Since that was a you know, you did the segue for me. So, you know, you're officially launching your perplexity publishers program, I understand that you're starting with a number of the publishers that we all know time fortune, their Spiegel, Texas Tribune and a few others. And you're going to be both licensing the contents from those sites, I guess, or sharing in the revenue or, like, what is exactly happening here in terms of the business agreement?
So it's a really simple structure. You know, if we're you are ourselves monetizing a page view, where a publishers content was used to help generate the answer. We share in that in that revenue with that publisher, right? And so regardless of how we got the content to be inputted as a source for an answer, whether it's a direct API integration, or we call their site, it's ultimately if we're using content, and we're monetizing it. And you know, we think the right long term incentive alignment is to share revenue in all those scenarios. Right. And that's kind of the crux of this program is a long term, multi year revenue sharing agreement where if we're using content we're sharing in that upside.
Okay, so Will I? So you're starting with this handful of publishers? And I guess what I'm curious about. And it has to do with sort of one of the I think there was a quote in the announcement from your CEO, Aravind Srinivas. And he said that it, the goal is to create a framework that's scalable and sustainable. So it seems to me that sort of doing individual deals with publishers isn't doesn't strike me as scalable? Is there some goal here to be more self serve about this, like, this might be a little bit selfish of me, because I'm a very tiny publisher that probably doesn't appear on anyone's radar, but there's a lot of small publishers that are in between me and time. How do they become a part of this partner partnership program? Well,
I think the so absolutely. I mean, that's why we call it a program not, you know, kind of a, you know, just one off partnerships, right, we're not doing and I think that's kind of where some other AI companies have, unfortunately, gone sideways, is they've prioritize just the biggest publishers. And ultimately, for us, we don't care whether it's, you know, one of the most famous media brands, or it's an independent journalists, you know, with their own WordPress site, and actually one of the launch partners is automatic for their wordpress.com publishers, right. So, and we this will all be self serve, you know, in a way almost kind of inspired by what what X has done with their Creator payouts, right. So it's like, we want a way to send checks to publishers and creators that when you know, their content is being used to help answer people's questions. And that's being monetized with advertising. You know, we want to share in the upside, if Ik, you know, the only real overtime selection criteria will be is that a high quality source? Right? We don't want to incentivize, you know, I mean, our search engine should be filtering that out anyways. But we want to make sure, you know, these are not folks that are spreading misinformation or things like that. But But otherwise, yeah, we'd love to have you in the publisher program. And, you know, we, we needed to just start with, you know, five and six, just to get it going. And you know, you know, get the mechanics good. But, you know, we have a email for publishers@perplexity.ai, we want to get in touch with everyone. And we'll make this onboarding really smooth for folks.
All right, well, I hope the overlords at substack are listening and take down that email. So the others word he used was sustainable. And I guess that'll lead alludes to, maybe what the rev share is, can you talk about what sustainable means? And talk a little bit about maybe even the numbers here in terms of like, when you talk about sharing revenue with the monetization?
Yeah, so we're not getting this specific numbers, I can share it's double digit percentages, revenue share. And and we, you know, sustainable means these are not like one time, you know, lump sums, these are multi year agreements. And we're making a open ended commitment to this model, right. And frankly, our investors don't love that we're doing this. They would rather we try to keep a margin profile. That's more like Google's, but we believe that part of how we compete against trillion dollar companies isn't by copying everything they do, right. And so we think this is an important differentiator. The the sustainable part is, it really is that alignment incentive with with revenue sharing, right, so if we're not using content, you know, there's no revenue being shared, if we are using it. There's a lot being shared. And it's also tied to kind of the success of advertising on perplexity, which for us, that's our big bet. So, I think that sustainability, I think the other dimension of sustainability is, you know, we started this program, you know, way back in January is when we conceived of it before there was any of the recent kind of criticism of, of perplexity, you know, in the media. And because we're just doing exercises imagining, you know, what does the world look like what also must be true of perplexity is successful? And one of those things was, arguably one of the most important ones is there must be a vibrant business model for the for journalism and the production of new facts about the world. Right, because that's we don't use the expression of journalism and perplexity but we certainly the facts that that journalists uncover the truth around the of the world that is very useful in answering people's questions and if there isn't a healthy, diverse, vibrant publishing ecosystem. That's not good for the our ability to answer questions, right? So the scenario where we're successful, and there is not a vibrant journalists ecosystem, that world can't exist, right? Either we fail, you know, for a variety of reasons. But you know, we can't succeed without journalism succeeding as well. And so this is, this is where we want it to kind of put put a stake in the ground saying that this is important. And, you know, let's kind of from first principles do the thing that Google never did, which is revenue sharing and sharing the upside.
Alright, it seems like they have no intention of doing that either, even with this new era, but I'll guess I'll talk to them about that. So something else I was thinking about, I'm just gonna put my publisher hat on for a second here, and get a little into the weeds on revenue, which is to say advertising is one part of Publisher revenue models today. But another significant part over the past several years has been commerce or affiliate linking. And often people, you know, one of the things they often typing to search engines is what is the best product for something I'm trying to shop for, for example, and the whole economy has emerged in the publishing world around this with affiliate links. Is this in some way taking into account in perplexities, answers? Can they pass on affiliate links, in addition to the advertising? Like, how does that work? Yeah, so
we actually built a structure that if we do any kind of affiliate linking ourselves, that that would be eligible for the revenue share. So we currently don't do that in perplexity. But if we ever were to that, that's kind of factored in, and I, you know, something sort of, certainly on our list of things to look into.
Yeah, and I don't wanna get too into the weeds of it, obviously, you know, I've worked in publishing for a while. So I think about that I've thought about this a lot. And in terms of like, the, what the reader is looking for, you want sort of the answer to your question of what's the best thing for me. And if I'm getting that in the perplexity UI, maybe my instinct, instead of clicking on the link, if it's not there is to sort of just open up another tab and go to the recommended thing. You know, the convenience of the link is great. But again, like I obviously the rev share gets a little complicated. I don't think it's hard to figure out. But it's more like passing on the link and making sure that that link is either the publishers or I don't know, maybe it's a new kind of link. That's a hybrid. Maybe someones thought through all this. But for me, it gets like a little complicated, because then you have like all these parties that got perplexity, the publisher, the affiliate link, and the retailer, all, all sort of grabbing some share of that.
Yeah, I think the I'll put our without revealing too much of our product roadmap, but we think there's a big opportunity to just like streamline, how when people are researching things, how we can make it easier for them to buy them. And so I think there's definitely kind of a lot we can do there. And we'd love to kind of bring publishers who are helping generate kind of the, the source input on on the recommendations and the facts around products into the folder as well.
Nice. So you mentioned you started working on this in January. And yes, that's definitely before all the kerfuffle around perplexity recently, you know, let's get into that in a few minutes. But January was after the famous New York Times lawsuit that was, I think, hit in the week between Christmas and New Year's. And there's obviously been a lot made of that. In the many months since we've seen a lot of publishers make deals with Alexa, open AI and others, obviously yourselves with this new announcement. And it just seems to me, there's kind of been a long and winding road sort of to get to this place, where publishers are having more direct relationships with AI companies. You know, there was a lot of, obviously, web scraping has been around since the web has been around, it's usually been in the realm of search engines, and to some extent, research. Everyone seemed to be okay with that. And now that that same data has been used to train MLMs, things seem to have gotten really great, really fast. And, you know, the, the whole sort of de facto of agreement seem to be you know, you're gonna give us traffic for indexing. Now, that's sort of no longer true. Just sort of big picture. You know, how do you how do you guys sort of see the whole sort of information ecosystem here and that sort of de facto agreement that was always sort of there between web crawling, and, and traffic?
Yeah, I mean, I think our vision For like the future, this is this revenue sharing structure which, which again, Google never offered. And so this is, you know, we think actually aligns incentives on a much stronger basis than then kind of traffic links, which which can ebb and flow. And yeah, as publishers experience, one point, Facebook was a lot of traffic, and then it died, you know, and so they've kind of had those ebbs and flows with Google as well. And I think it's also worth just on some of the context setting you provided your perplexity does not train foundation models. And so we, we use third party lens, or we use open source models that we fine tune, but we were not in the business of scraping the web, and using that to train an AI. In fact, we don't use any web content to improve our AI models. We use prompt data, but not the actual web data. And again, we're not in the business of creating foundation models whatsoever. So you know, the, the open AI in New York Times thing that really had no bearing on, you know, our decision to launch a publisher program. And as I said, we're only 18 months old as a company operating in the space. And so from from January today is a meaningful chunk of our lifetime. So we're moving very quickly. But it's not really in reaction to you know, what others are doing. It's us kind of quickly iterating and improving on the product and, you know, taking feedback from the market.
Right. I'm glad you're here to clarify things, because I think this sort of segues into the sort of, I guess, I call it a kerfuffle, between perplexity and some publishers recently, so obviously, I'm referring to when Forbes accused perplexity of essentially plagiarizing one of its articles wired seemed to jump in there and have some criticisms too. But one of the things sort of within all this, and I don't want to go over the whole thing again, it's more that it was but what it speaks to sort of what you were just describing a little bit, and the nuances involved, which is to say there was an accusation leveled that the robots dot txt file. Now again, it's getting into the weeds of websites that put the basically publishers use this as a signal to search engines and AI engines on like, what you can do with their content, what's what, what is allowed, so to speak, you can block training or search or at least tell the thing that's going to do that, that it shouldn't do that. And there was an accusation level that's like, oh, is perplexity, ignoring this, because some article ended up being referenced or summarized in the engine, when that site had its site configured to to tell it to not do that? So can you clarify like, what was that about? And like, like the, you know, that is, is that was that a mistake was there like, I know, our robots dot txt, as I alluded to just a second ago, is kind of the honor system on the web, in some ways. But since you guys don't train models, but you are a search engine, how do you interact with it? And then what is your stance on? I guess, respecting it?
Yeah, great tee up there. So I think in terms of the I mean, I'll first make a general comment, which is, if I, if a journalist or a commentator on on X, Twitter, wants to generate a screenshot that gets an AI or an LLM, to say something stupid. You can do that all day long. Just like you can trick a person, you can trick an AI or an LM, I don't think that type of journalism is particularly useful for the public, because that's not how people use these products, like people use them, not to stress test them, they use them to try to answer questions from their everyday life, their work life. And they work very well in those situations. And if they did it, people wouldn't be using them for that. So I think there's like, you know, there's kind of the stress testing and red teaming of hypothetical scenarios. And there's actual usage, right. And it's important to distinguish between them, it kind of brings us to the wired example. So what wire did is they in, in the question itself, put in a URL of a website. And when you were in that, and that URL was disallowed on robots dot txt. Now, when you when you did that before in perplexity, we would on behalf of the user, go and look at the content of that URL to understand their question, but that's not crawling, right? Like Robots. txt is a set of instructions for crawling. This is a user initiated action where it's like, hey, you know, equivalent to that user themselves. Just going to that website, copying and pasting the contents and putting that into the All right. And so we, we saved them the stuff. So I think it's a disingenuous criticism to say that robots dot txt is irrelevant there.
I see what you mean, I see the distinction.
That said, because nobody actually uses perplexity that way we actually, you know, if it's a, if you tried to do that now, it would just say we can't go to that URL, right, because we don't want there to be confusion on the topic. And so I think that piece is is. So there's one part there. I think a second part is, you know, we believe that in democracies and open societies, there is no monopoly on facts. And and so, you know, and this is, I think, something journalists believe deeply, too. You know, if Fox News says, Hey, MSNBC, I don't want you using clips from Fox News in your reporting. MSNBC would not adhere to that, because there's essential information, facts that are in that reporting that MSNBC wants MSNBC wants to comment on. And so you know, what we, we respect robust attacks, but we do use summaries of web articles that are provided by others, just the factual content, you know, the metadata around that, to answer certain questions where it's relevant. And so that's, you know, we're not looking at any full texts, or partial tax from those from those sources, but we are looking at a summary of it. And I think that's, that's important, for, you know, information not to be, you know, for people to get accurate answers. And honestly, it's the basis of an open society is that, you know, once a fact is reported, it is in the ether. And I think that's been an important, you know, that this has been an important principle copyright practices going back to the 17th century. And, you know, it's not really a novel issue to AI, I mean, you know, with with the the kind of blogging, ecosystem and aggregation, like, you know, these issues have been fresh for a while. And again, I think the, the particular relevance here is people don't use perplexity in a way that is competitive or cannibalistic to going to a new site, people come to perplexity asking a very specific question. Nobody comes to perplexity asking, like, what's the news today? That's just not that's not what it's designed for. It's not what it's good at. And people ask, you know, these again, 10 word median length questions. And, and it's more like, you know, how many home runs and Babe Ruth hit in 1924? Less so like, you know, you know, what are you what happened to the Dodgers game? Today? Right, like, so? I think there's, there's a lot of difference in user intent. That is worth noting here, too.
Cool. That's a very comprehensive answer. There's actually a lot there, I'd love to unpack and go deeper on but the point about articles and material being copyright, but information isn't is certainly well taken. It's sort of the basis of the aggregation side of media that's always existed. The I think, AI and whether it's perplexity or other products bring to the table is a certain degree of scale. instantaneousness that we haven't seen before. I would also add that there's a certain amount of muscle and we'll call it muscle memory. But set of unwritten standards when journalists tend to do it to each other, right, there's a lot of sort of citing of the original source of sort of not necessarily doing a complete rewrite, there's a bunch of sort of little parts of that, do you think perplexity could maybe do a little better job there in terms of when there's like, say, a scoop or something that really is only available? Or was was clearly the product of some hard work on the part of journalists? You know, obviously, this ad model that you're talking about, and and the partnership program is gonna go a long way. But is there are more that could be even done in the product itself?
Yeah, I think the our original insight again, was you start with the sources, right? We actually list the sources before we even go into the answer. You know, I think some of the critical coverage came out of this this thing with Forbes, where that actually wasn't about the core perplexity product. We had a experimental product called perplexity pages, which a bit more of a stylized Answer that that's curated. And it literally it was the product was 10 days old. And we were not displaying sources in the same format as we do on a typical answer. And it was fair feedback. And we actually changed the UI within 24 hours of that. We also changed the system prompt that four pages, if it does rely on a scoop, it won't only list the source of the top, but like a, like a journalistic convention convention, it will in the body of the page, say, you know, as reported by, you know, the New York Times, right, as reported by Bloomberg. And we actually got feedback from journalists that like, and we're incorporating this now, like, take it a step further, say, as reported by John Smith at Bloomberg, as you know, so attribute the the individual reporter, so I, yeah, I mean, I think we're,
So it’s already happened.
We’ve incorporated that feedback, I think there's it's the perplexity pages products is just like a very minor part of our experience. And so I think there was some conflation between that what people actually use us for, where it's not as relevant a question. But, but yeah, I think it was, listen, we think it's, you know, I've worked at tech companies that have touched media for the bulk of my career, I was at Facebook, I was at one of the first mobile news aggregators called Pulse news.
I remember Pulse news. I was kind of a power user.
So I was I was then at LinkedIn through the acquisition. And, and so, but publishers have a legitimate right to be anxious. I don't think any of the challenges publishers are facing today and their business models are tied to AI. But this is, you know, it's not unreasonable to kind of get concerned about where this goes a few years from now. And, you know, again, that that was why we took the approach, unlike some others of these long term revenue sharing structures, where we align incentives that way over kind of a one time lump sum.
Nice. Okay, so I could go even deeper on some of that. But I do want to ask you about at least one other piece of big news that dropped this week. So open AI, as you probably saw, recently unveiled its search product. And its quote, coincidentally, right before your guys announcement and said that, and it seems to have a similar idea. It's essentially an AI powered search engine, it's got publisher partners. How do you regard their approach? And even just competitors in this space more broadly, who are putting AI together with search?
Yeah, so from our outset, we've, we've effectively been competing against, you know, trillion dollar companies or companies bankrolled by trillion dollar companies. This announcement, you know, it's been teased for, you know, since the beginning of this year, and honestly, the chat GPT product has actually had even in the free version, you know, web search, going back to the launch of GPT-4o. So this, there's actually not that much new here. I mean, that this has been, this has been live for a while, and Bing has been live and Google AI overviews. All I can share is every time a competitor announces something similar, our traffic just keeps going higher. So it hasn't impacted our business. I think, you know, at our core, any startups core, there's only two advantages you have over larger, less, more distracted incumbents, velocity and focus, right? So our operating philosophy is 1% better every day, we make the product 1% better every day, incremental improvements, and that compounds over a year, you end up with a product that's 30 times better over the course of a year. And so that's been you know, you talk to somebody who use perplexity this point last year, and they use it today and it feels like a very different product, even though day to day. It's just small changes that compound on each other. And we we think competition is great. It sharpens our focus even even further. And and I think the more people you know, get trained to ask natural language questions on the internet. And that's good like our success is ultimately tied to growing global curiosity. Right people ask more question shins today than they did yesterday, and are trained to kind of want to ask more questions. That's a good environment for for perplexity. And you know, open AI is doing many things they're trying to build AGI they're trying to train new foundation models are trying to do video. All we wake up and do and we dream of doing is answering people's questions. And there's a lot of you know, there's a lot of benefit that comes from that focus.
Yeah, it's got to be, I guess, maybe even a little flattering, since you know, you're right, that they've had search in some form for quite some time now, at least since the fall, though, with regard to an official, quote, unquote, search product. They seem to acknowledge they needed to rework their UI into something that's a little more visual, and have more sort of provenance studied of sources, which obviously, are just features built into perplexity that said, they've got a hell of an Onboarding Tool, which add GPT. I don't know if there's a more recent stat. I think they said something that once they had 100 million active users, how do you compete with that? I have you guys do you? Can you tell me how many active users you have? Like, what is? And what is sort of the percentage that maybe do pro? Like, what is your user profile look like?
So we tend to, as I said, think of the world through questions as per day. You know, just in the last month, over 250 million questions were asked and perplexity all of last year, it was only 500 million. So we're seeing very rapid growth. I think chat, GBT is an amazing tool. And you know, we have all the respect in the world for the open AI team. But one of the challenges with it, is it. It's so useful for many different things you don't know, what's the one thing to come to it for? And so I think that that's going to be a persistent challenge. In kind of these very broad AI chatbots is, people don't want a chatbot. They want specific problem solved, right, the problem we solve for people is we answer their questions. And so people will know what to use perplexity for. And that's very powerful. And so yeah, I mean, for if we're going to, it's one of those things, if like, We were so worried about an upfront distribution advantage, you know, don't compete against Google, right like that, if if there's a company we should be, like, terrified of, it's that one. And yet, we just keep growing right. And like, we actually grow faster, the more launches, they announced, because in many ways, it validates that we're onto something big. And, you know, we're the only player that is well resourced, but only has this to focus on. And it has a product and brand that our users love. So we like our odds.
well, to pick out just a couple of details in the announcements. So you mentioned, you know, you're sharing analytics with the publishers of the links to people click on that sounds like a big deal for publishers, they always want more data on what their audiences doing. How do you think they'll use that to adapt their strategy? I mean, for a long time publishers, you know, we would look at their traffic from search and social and adapt strategies around that. How do you how do you adapt strategy around AI search?
The thing that is really interesting about this new paradigm is in contrast to traditional SEO, right, so we want publishers to know, which of their articles are useful to answering people's questions because of the facts laid laid in them. And then of those, what are the ones that kind of show up in the categories that advertisers want to reach? And that's a very powerful because because if you know, the articles that are generating revenue from a rev share point of view, then you can create more of that, right? And we want to create these healthy feedback loops, where publishers are motivated and all kinds of publishers, small ones, big ones, to create content that has the types of facts in it verified knowledge, that that is useful to answering people's questions, right, because if there's more of those facts in the internet, perplexity becomes more useful. And it's, it's this virtuous kind of feedback loop. And so I think that that's kind of what what we and so that's kind of what's in it for us. I think publishers, there's all kinds of other great things, you know, I mean, they can infer based off of how their content appears and perplexity and when, how it appears in other MLMs and other kinds of AI products. So I think there's a lot of indirect benefits that come from from us kind of being open kimono there that they'll appreciate over time.
Just to clarify By, you wouldn't have suffered from a favourability problem, right? Well, what I mean by this is that you obviously have some deals in place.
Yeah, we're not we're not going to the search index and ranking system is blind to who's in the publisher program. That said, if you make more of your full text content visible to our search index and ranking system, it's more likely that when it's relevant, it will be cited as a source, just because we, you know, there's a higher a larger surface area to do that semantic match.
Gotcha. Cool. Thanks for clarifying that. So, one of the other things I noticed in the agreement was the, you're gonna be sharing some API's, publishers, presumably will be able to essentially create their own maybe mini perplexities, I guess, on their own sites so that they can sort of train on their own data. That seems really interesting to me. Certainly chat on sites, there's there's a lot of solutions for that. But what is the advantage of using your guy's API and in this case.
I mean, we're really good at it. And it doesn't require, you know, because we can already do that indexing, like, you don't have to do anything different, right? Like the onboarding is kind of painless. And you can have a solution up and running right away. I think the other functionality that's really powerful is actually the related questions. So 40% of questions and complexity of a follow up question. And that's, I think, you know, that's a powerful insight that can be relevant to publishers of you know, how do you get people to spend more time on an article, and so opening up that API, so that, you know, you can point not a question, but a URL, and generate related, you know, kind of perplexity powered content to it. And is, and keeping fit, you know, it's not about linking back to perplexity, it's just about creating kind of complementary content, and you can pick which domains, so that that content can just be, you know, leveraging your your historical content. It doesn't have to, like use the broader web. And so I think that's something we're we're excited to, and publishers are excited to take advantage of.
Cool. When I thought about this, I thought about the perplexity product itself. And you know, as a tech guy, I've always thinking about ways I can tweak things and customize them, would this search itself perplexity, allow for certain ever allow for a certain amount of tailor and only have filters, which is cool. But some people sort of suggest, like, you know, what, if you could tell it, I want this type of site only in my results? Yeah. Just through prompting, like, I just want blogs or whatever, you know,
I mean, that was the API, you can't do that. I mean, you can try to do the prompt tacking and corporate laxity, but it won't work perfectly. But with the API, you can whitelist or blacklist certain domains that will will or won't show up as source inputs. Nice.
Interesting. And what about those filters, because I noticed that you had one that was called Reddit on my version anyway. And now it says social, but it's only shows me Reddit stuff. Like, is there more coming into that? And second, similar question for video and YouTube? You only get YouTube links and that one, but the label video would seem to imply would get other stuff.
I'll be honest, like those focus views don't aren't really used that often. We're constantly a be testing things, but it is I think we're just trying to reimagine what focus may look like. But it's not it's not a main area of investment for us at the moment. Gotcha.
Makes sense. Cool. So I've kind of wondered, why is it called perplexity?
So so we're gonna go above my technical paygrade here, but perplexity is actually a concept in artificial intelligence, like academic literature that I will not begin to explain, but what I will do and maybe you can include in the show notes that I'll send you a good perplexity answer for what the technical concept of perplexity means, but our founders were AI PhDs. And so it was, it was kind of a geeky reference to a technical, scientific concept within AI research.
Just to close things out, I always in most of these interviews with a question like this, think about the future think about five years from now we have a more of an AI mediated information ecosystem. What is that that look like to the average user and then what is perplexity’s role in it?
So I always I hesitate to make predictions about the future. I like the more Bezos view of like what are the The things that won't change. And and as it relates to perplexity, you're the thing we have conviction on is the uniquely human aspect of our business is curiosity and the desire to ask questions like, you know, you could train an AI to generate questions, but you can't train an AI to want to ask questions, right. So that innate desire to learn more, to be curious, you know, that that will be the driver of our business, you know, you know, 510 2200 years into the future. And so, you know, that's the thing we focus on cultivating, you know, the total addressable curiosity. And, and I think that that's going to be you know, that that will be the Evergreen part of what we do.
That's a new one for me total addressable curiosity, but somehow it's pretty fitting. Dimitri, this has been great. Thanks so much for taking the time. This is really illuminating.
Really enjoyed the conversation looking forward to coming back against it.
Perplexity's Master Plan, With Dmitry Shevelenko