The volume of online data has been growing exponentially ever since the early days of the internet. In 2023 alone, an estimated 120 zettabytes of data were created worldwide, with global data creation projected to reach 150 zettabytes by 2025. The 120 zettabytes of data created in 2023 marks a nearly 60x increase in the amount of data created in a year, from the same year 2010 — representing a 60-70% annual growth rate. Despite this massive influx of data, the efficacy of traditional search engines may be deteriorating, making it harder to find the desired information. For example, a study published in 2024 by a team of German researchers after a year-long review of results for 7.4K product review queries on Google, Bing, and DuckDuckGo, concluded that search engine query results have become increasingly populated with low-quality, highly optimized content that it characterized as “SEO spam”.

In addition to this, the average position of the top organic link on a Google search engine results page lowered from 375 pixels down the page in 2013 to 615 pixels in 2020. Recent studies indicate that these changes are more than just a minor inconvenience: over 50% of consumers reported in 2020 that they sometimes feel misled by search results, with 25% saying that they often end up somewhere unexpected by clicking on a search result. As a result of increasingly cluttered search experiences, some people have even started looking for answers in places like Reddit and TikTok instead of traditional search, with others touting that new advances in AI might put the future of search engines in peril.

In the late 2010s, a pivotal shift occurred in natural language processing with the introduction of transformer architecture. Unlike previous recurrent neural networks and convolutional neural networks, transformers revolutionized the field by introducing self-attention mechanisms. This architectural innovation enabled models to process entire sequences of data in parallel rather than sequentially, leading to significantly faster training times and improved performance on complex language tasks.

This breakthrough, combined with enhanced processing power and the wealth of data available online, led to significant advancements in the size and scope of LLMs. With extensive training, these models become applicable to a wide variety of tasks. In November 2022, following its initial launch, ChatGPT crossed 1 million users in just five days. The product then became the fastest-growing consumer application in history, reaching 100 million users within two months of its launch.

However, LLMs have their drawbacks, particularly in the realm of information retrieval. They are expensive to train on the most up-to-date knowledge and are susceptible to producing outputs that are occasionally factually wrong — a phenomenon known as hallucination — that stems from overfitting on the available data.

Perplexity, also known as Perplexity AI, is an AI-powered, conversational search engine that answers queries using natural language predictive text utilizing sources from around the internet. Perplexity performs real-time web searches to gather the most current information available across the Internet. Then, it summarizes the information in a cited format, using LLMs mainly only to preprocess the questions and summarize search results, which is intended to deliver up-to-date and well-sourced information. Perplexity describes itself this way: “unlike traditional search engines that present a list of links, Perplexity delivers synthesized information in a natural language format, complete with citations and follow-up suggestions for more refined searches.” The company’s mission is to “serve the world’s curiosity”.

Perplexity was founded in 2022 by Aravind Srinivas (CEO), Denis Yarats (CTO), Johnny Ho (Chief Strategy Officer), and Andy Konwinski.

Growing up, Srinivas was fascinated with computers and mathematics. After earning a Master’s in Engineering from the Indian Institute of Technology in 2017, he pursued a PhD in Computer Science at UC Berkeley, specializing in reinforcement learning and predictive coding. During his postdoc, Srinivas interned at Deepmind and OpenAI.

Srinivas and fellow cofounder Denis Yarats first connected through email after the two published nearly identical research papers at UC Berkeley and NYU, respectively. They stayed in touch even as Yarats started working as an AI researcher at Facebook. In July 2022, Srinivas and Yarats joined together with former Quora engineer Johnny Ho and Databricks cofounder and fellow UC Berkeley alumnus Andrew Konwinski to formally pursue a startup together.

Perplexity’s first product, Bird SQL, launched in December 2022. It used a tool from OpenAI called Codex to help turn everyday language into computer code and search through databases like Twitter. The tool turned natural language search queries into SQL, which would then return Twitter-specific results, and even got the attention of Jack Dorsey.

However, when Twitter announced it would stop allowing free access to its API in February 2023, the team decided to switch gears to focus on a core search product. By this time, ChatGPT was rapidly gaining popularity, having hit 100 million monthly active users within two months of launch, and the team saw an opportunity to create a product that addressed the potential of AI chatbots to source inaccurate information with unclear sourcing by adding citations for generated answers.

Speaking on this topic, Srinivas noted that “citations are a great way to marry search and LLMs.” Perplexity developed a core vision “to become the best platform for answers and information” and launched an updated version of its Ask product in January 2023 focused on adding sources and follow-on questions to initial queries. The company described the launch as “the world’s first conversational search engine”, and it reached 2 million monthly active users within its first four months.

Perplexity’s core product is a conversational search engine that leverages advanced LLMs to provide direct, sourced responses to user queries in natural language. It offers a personalized and interactive search experience, with features like document importation for reference, image generation, and educational, professional, and content creation assistance.

Source: Perplexity

Search Interface

Perplexity’s core offering is its answer engine is intended to bridge the gap between conventional search engines and AI-driven chat interfaces. As the CEO Aravind Srinivas described it in a blog post in January 2024:

In contrast to conventional search engines, which typically yield a list of hyperlinks, Perplexity’s core engine uses LLMs to summarize real-time search results into a succinct, text-based format with inline citations. In doing so, the engine prioritizes contemporary sources. Compared to traditional search, conversational AI provides an interactive search experience by enabling users to pose follow-up questions within the same conversational context.

Inline citations allow the product to confront two more significant challenges inherent in LLM-based approaches to information retrieval: outdated data and hallucinations. Perplexity’s answer engine conducts real-time web searches to fetch the most up-to-date information to combat outdated data, ensuring that current data takes precedence. Additionally, the ask engine cross-references model output with contemporary sources to verify accuracy and reliability in tackling hallucinations.

Multiple LLMs

Perplexity's approach to its answer engine is characterized by a hybrid model strategy, which uses both internally developed models (based on open-source models) and third-party foundational models. Initially, Perplexity relied on OpenAI's GPT-3.5 model and Microsoft Bing for its search capabilities. However, it has since announced the development of its own LLMs that it built on top of open-source models like LLaMA-2 (released by Meta in July 2023) and Mistral 7B (released by Mistral AI in September 2023). This approach has enabled Perplexity to concentrate on optimizing its search platform without the necessity of developing and training proprietary models from the ground up.

Perplexity Pro users have the option to interchange between different series of models provided by Perplexity, including OpenAI’s GPT series, Anthropic's Claude models, and offerings from Mistral AI. This versatility allows users to select models that precisely align with their specific research objectives and preferences, including writing and image generation.

Source: Perplexity

Integrations

Perplexity provides various tools intended to enhance the search and browsing experience across different platforms, including mobile devices and web browsers.

Perplexity has developed a mobile application compatible with both iOS and Android. It offers various features designed to facilitate easier and more effective information retrieval, with an emphasis on voice search. The iOS app can be downloaded from the Apple App Store, and the Android version is available on Google Play. Both versions are free to download and use.

Perplexity also offers a Chrome extension that integrates into the user’s browser. This extension can provide instant page summaries, enable queries directly from the toolbar, and offer a contextual understanding of the content being viewed.

Source: Chrome

The Arc browser also partnered with Perplexity in January 2024, integrating Perplexity into the Arc Browser as a default search engine option.

Perplexity Pages

Source: Perplexity

Perplexity Pages, launched in May 2024, is a tool that helps turn research into content that is intended to be visually engaging and comprehensive. It allows users to create, organize, and share information by generating formatted articles on any topic. With features like customization for different audiences, adaptable article structures, and enhanced visuals, Pages is intended a wide range of creators, including educators, researchers, and hobbyists. The tool is aimed at simplifying the content creation process, enabling users to focus on effectively sharing their knowledge and publishing their work in Perplexity's expanding library of user-generated content.

Customer

Perplexity serves a wide range of customers, including individual users through its free and pro tiers, and organizations via its business tier. In January 2024, Perplexity said it had over 10 million weekly active users and had served over half a billion queries in 2023.

Perplexity initially focused on refining its technology through a small user base on Discord. Over time, it began to offer a more comprehensive UI and product offering. As an AI search engine, Perplexity's target customer is anyone seeking accurate information from credible sources.

Source: Perplexity

While Perplexity targets a broad market segment, it offers advanced “focus” modes tailored to specific aspects of web search. These include an academic mode for searching through cited articles, a writing mode for pure text generation, a math answer engine, and specialized searches for entertainment and discussion boards.

In April 2024, Perplexity launched its formal Enterprise Pro offering. This launch followed a period of early access testing with select companies across a number of industries including Zoom, Stripe, Bridgewater, Snowflake, the Cleveland Cavaliers, Universal McCann, Thrive Global, Databricks, Paytm, ElevenLabs, HP, Vercel, and Replit. Those early adopters helped shape the enterprise solution. Through this phased approach, Perplexity has established a foundation in serving both individual and enterprise customers.

Market Size

Perplexity is a player in the search market because it functions as an AI-powered search engine, allowing users to input queries and receive relevant search results. The global search engine market was valued at $167 billion in 2021 and is expected to grow to $529 billion by 2032, representing an 11% CAGR. The rise in internet connectivity globally has significantly contributed to the growth of the search market. For instance, the share of adults using the internet in the US rose from 86% in 2015 to 95% in 2023. Likewise, digital advertising spend remains robust and continues to grow, and was forecasted to reach $350 billion by the end of 2023.

Additionally, with the development of its own models as well as its API offerings, Perplexity has also entered the generative AI market. The generative AI market was valued at $29 billion in 2022 and is projected to grow at a CAGR of 47.5% to reach $668 billion by 2030. Although much of the value of the market is being captured by leading players such as OpenAI, Midjourney, and Anthropic, Perplexity may be able to carve out its own sizable position, especially with its focus on enterprise clients and future model offerings.

Google: As of February 2024, Google remained the dominant search engine globally, holding a market share of approximately 91.6%. It has been the leading search engine for well over a decade and answers approximately 2 trillion global searches per year. Founded in 1998 by Larry Page and Sergey Brin, Google owes much of its success to the groundbreaking PageRank algorithm, which revolutionized how search results are ranked and retrieved. Initially conceived as a research project at Stanford University, PageRank propelled Google to prominence by prioritizing web pages based on their relevance and authority rather than simply relying on keyword density. In November 2023, Google began experimenting with Generative AI in its own search product under an opt-in experience from its internal division, Search Labs.

In May 2024, Google announced that it would begin rolling out a very similar experience to Perplexity, with AI Overviews at the top of its native search feature. AI Overviews are built on top of and integrated with Google’s Shopping Graph, which allows the Overviews to populate suggested advertising products for queries related to search natively, potentially lessening the pushback from advertisers that might lose click-through rates with an advanced search experience. While Google’s AI overview feature is similar to Perplexity, including suggested questions and citations, it is worth noting that Google has received extensive criticism of the feature. The rollout has been mocked on social media, with users able to widely reproduce hallucinations on the overview feature that suggests users should eat rocks, put glue on pizza, or that dogs have played in the NBA. In each case, the answer was generated due to an overreliance on user-generated content from blogs or Reddit, creating a larger question of what sources on the web should be treated as authoritative.

Although Perplexity and other AI-native search products also struggle with hallucinating answers, Google faces significant scrutiny and higher expectations relative to startup competitors given the scale of its user base. Google has encountered substantial pushback with its AI releases, highlighting the innovator's dilemma that some see it currently facing. The launch of Gemini, Google's generative AI tool, was particularly criticized for its handling of racial demographics in image generation. Google admitted it had "missed the mark" after Gemini produced historically inaccurate images by overcompensating for diversity, such as depicting nonwhite individuals in contexts where they wouldn't historically appear. Google’s parent company, Alphabet, has a market capitalization of approximately $2.2 trillion, making it the fourth most valuable company in the world as of June 2024.

OpenAI: Founded in 2015, OpenAI was originally a non-profit organization which became for-profit in 2019. The company is known for creating the Generative Pre-trained Transformers (GPT) series of AI models, which it first introduced in 2018. OpenAI has raised $11.3 billion in total funding as of June 2024. In February 2024, OpenAI reportedly completed a deal that valued the company at $80 billion.

In September 2023, OpenAI announced that its ChatGPT product would be able to access the internet and provide real-time information to users through Bing web search. This feature is only available to ChatGPT Plus subscribers. Meanwhile, in contrast to Perplexity, ChatGPT does not provide citations for its answers to user queries. As of June 2024, ChatGPT was estimated to have 200 million monthly active users, after having reached 100 million users within two months of its launch in November 2022.

Meta: Established in 2013, Meta AI develops the LLaMA series of open-source foundation AI models. In April 2024, Meta announced that its Meta AI assistant would be integrated into the existing product suite (Facebook, Instagram, WhatsApp) in order to bring information retrieval and LLM querying closer to the product experience for the 3.6 billion people who were active on at least one Meta platform as of April 2024.

Bing: Bing, owned by Microsoft, holds a smaller portion of the search market, with a share of about 3.3% globally as of 2024. Despite significant investments in integrating AI into the search experience through features like Bing Copilot in the last year, Bing has struggled to significantly erode Google’s market share. Bing’s market position is bolstered by its integration into Microsoft’s products and services.

You: You is a newer entrant in the search engine market that also positions itself as an AI-powered search engine. It was founded in 2020 by Richard Socher and Bryan McCann, both former employees at Salesforce, with the mission to create a more customizable and private search experience. Similar to Perplexity, You also has citation ability and uses other AI models to simplify the search feedback experience. You.com has raised a total of $45 million in funding as of June 2024.

Source: Perplexity

Perplexity operates on a subscription-based freemium business model where basic functions are offered at no cost and additional features may be added with a monthly fee. Perplexity had three pricing tiers as of June 2024.

1. Free: Unlimited Quick searches (searches using Perplexity’s own model that offer faster and less in-depth information on a query), personalized answers, and five Pro searches per day.

2. Pro: The pro tier, which costs $20 per month, offers everything in the free plan, plus 600 Pro searches per day. Pro searches use a variety of models to provide extensively researched answers with improved accuracy, like GPT-4 Turbo and Claude 3. It also allows for file uploads, and a $5 monthly credit to use on Perplexity’s API.

3. Enterprise: In April 2024, Perplexity announced that it had developed a suite of B2B offerings designed to cater to business clients. Users of Enterprise Pro have access to team management, single sign-on integration, and increased data privacy. In addition, Perplexity noted that it already has seen enterprise adoption from Databricks, Nvidia, Zoom, and the Cleveland Cavaliers basketball team, among others. The Enterprise Pro plan has a self-serve plan for companies with fewer than 250 employees with pricing at $40 per month or $400 per year per seat, and custom pricing for companies with over 250 employees.

In addition to its subscription services, Perplexity is exploring revenue generation through advertising. In April 2024, the company announced future plans to integrate ads into its platform. This move comes as a reversal of its previous stance on its website, where it had previously stated that search should be “free from the influence of advertising-driven models.” A line that, in March 2024 was removed from Perplexity’s site. In this announcement, Perplexity’s Chief Business Officer, Dmitry Shevelenko, said, “advertising was always part of how we’re going to build a great business.” Although the future method of employing advertising in the product is not yet definite, Srinivas noted that “when ads are done right it’s amazing, and generative AI is going to help us build even better targeting.”

Perplexity launched its flagship answer engine in December 2022. Four months later, in March 2023, the company reached 2 million monthly active users. Throughout 2023, Perplexity would grow to process half a billion queries.

By January 2024, Perplexity reported that it had expanded its user base to 10 million monthly active users. The company has also made strides in global market penetration, particularly in India, which currently has 1 million users.

As of April 2024, Perplexity was serving 169 million monthly queries and was reported to have $20 million in ARR. Perplexity has also attracted several major enterprise users for its Perplexity Enterprise Pro product. Notable companies include Databricks, Zoom, Hewlett Packard, the Cleveland Cavaliers, Stripe, and Thrive Global.

In April 2024, along with the launch of Enterprise Pro, Perplexity raised a $62.7 million Series B1 funding round, which the company said doubled its valuation to “over $1 billion”. The investment was led by Daniel Gross (former head of AI at Y Combinator) with participation from new investors Stanley Druckenmiller, Garry Tan (CEO of Y Combinator), Dylan Field (CEO of Figma), Brad Gerstner (Founder & CEO of Altimeter Capital), Laude Capital, Lip-Bu Tan (former CEO of Cadence), and Jakob Uszkoreit (co-inventor of Transformers). Many of Perplexity’s existing investors, including Jeff Bezos, NVIDIA, Tobi Lutke, Elad Gil, Nat Friedman, Naval Ravikant, Andrej Karpathy, IVP, and NEA, also participated in the round.

The Series B1 followed closely after a $73.6 million Series B the company raised in January 2024 led by IVP. Other notable investors in the round included NVIDIA, Balaji Srinivasan, and Austen Allred. Perplexity has raised a total of $165.3 million in funding as of June 2024. Although Perplexity’s Series B1 round was finalized in April 2024, subsequent reports in May 2024 indicate that this might not be the end of fundraising for the year 2024. Multiple sources in April 2024 reported that Perplexity was on the verge of securing an even larger round of at least $250 million at a $2.5 billion to $3 billion valuation.

Shifting Paradigms in Internet Search

The exponential growth of data on the internet has necessitated the evolution of information retrieval technologies. Approximately 120 zettabytes of data created globally in 2023, with projections reaching up to 150 zettabytes by 2025. This massive influx of information comes as traditional search engines are increasingly returning low-quality, highly search engine optimized content. The deterioration in the quality of search results, coupled with the physical displacement of top organic links to lower positions on the search pages, has led to consumer dissatisfaction and a shift towards alternative platforms like Reddit and TikTok for information retrieval.

In response to these challenges, advancements in AI have paved the way for more sophisticated AI-driven platforms that benefit from big data. Transformers, which utilize self-attention mechanisms, allow for the parallel processing of data sequences, significantly enhancing the speed and efficiency of training LLMs. These models, exemplified by the adoption and growth of ChatGPT, have shown remarkable capabilities in generating human-like text and processing vast amounts of information quickly. However, they suffer from drawbacks such as expensive training costs and the tendency to produce factually incorrect “hallucinations.” Perplexity represents a new approach in this landscape, combining real-time web searches with the processing power of LLMs, and can benefit by continuing to position itself as encompassing the best of both worlds.

Customer Expansion

Perplexity’s launch of Enterprise Pro, a service tailored for business clients, in April 2024 marks its entry into the B2B sector. This move into the enterprise market represents a significant opportunity for growth, diversifying the company's revenue streams beyond its consumer base and Pro subscriptions.

Additionally, the company announced plans for global expansion in April 2024, with partnerships with major telecom companies, SoftBank in Japan and Deutsche Telekom in Germany. The collaboration with these telecom providers is expected to distribute and market Perplexity's services to over 365 million additional users, spanning both mobile and broadband sectors. Perplexity can continue to expand to encompass enterprise customers and partnerships like those with Softbank to differentiate itself against other AI answer engines and search engines in a competitive market.

Profitability in Search

As of April 2024, much of the online advertising economy still hinges on users navigating through Google to click on links, subsequently landing on publishers' websites where they are presented with ads. By leveraging AI to preprocess the search results, Perplexity eliminates the need for users to directly visit websites. Instead, the AI sifts through the internet on behalf of the user, aggregating and presenting all necessary information directly within the answer page. This offers a more efficient search experience but diverges from the ad-based revenue model that has historically fueled the search industry — which is almost entirely based on actually interacting with the advertising results.

Although Perplexity had grown its user base to 10 million monthly active users as of January 2024 and its ARR was only $20 million as of April 2024. The transition from an ad-supported to an AI-curated search experience raises questions about the sustainability of traditional search revenue models in the face of higher computational costs. A report from 2023, for instance, suggested that running a search query through an advanced neural network like Google Gemini “likely costs 10 times more than a standard keyword search”. This indicates that Perplexity will have to both navigate higher costs than traditional search engines while also not being able to rely on advertising revenue. In this sense, the search industry is at a crossroads, exploring new revenue models to adapt to AI. It remains to be seen how Perplexity will modify Google Ads traditional ad model to build a business that is sustainable in the long term.

Questions Around Content Ownership

A December 2023 lawsuit filed by the New York Times against OpenAI and Microsoft underscores a critical challenge faced by traditional media with the rise of AI-driven content generation. The core of the issue lies in the unauthorized use of copyrighted material to train AI models, such as ChatGPT, which can then produce content that competes with or even replicates the original sources without directing traffic or revenue back to them. This situation poses a threat to the financial sustainability of news organizations, which depend on website visits for advertising revenue, subscriptions, and engagement metrics.

Perplexity, by providing direct answers and attributing sources without necessitating a visit to the original website, has the same issue. While attribution is a step towards acknowledging the source of information, it does not compensate for the loss of direct engagement and revenue that would have been generated from a visit to the publisher's site. This model, therefore, raises concerns similar to those highlighted in the New York Times lawsuit: another case of the use of content by AI models without adequate compensation or control over its dissemination by the content’s original creators.

If users can get the information they need directly from AI-driven platforms without visiting the original content creators' websites, their revenue models based on advertising, subscriptions, and pay-per-view content could be severely undermined. This could lead to reduced financial resources for these publications, impacting their ability to produce high-quality journalism to train new models, which could in the long-term negatively impact the data available to AI tools like Perplexity. Furthermore, it raises questions about the fair use of copyrighted material, the ethical considerations of AI content generation, and the need for business models that ensure content creators are compensated for their work.

The rapid growth in online data has challenged traditional search engines. Studies reveal deteriorating search result quality, with users sometimes feeling misled. Meanwhile, advancements in natural language processing, particularly the transformer architecture, have led to advancements in AI with the advent of LLMs. Perplexity, founded in 2022, leverages real-time web searches combined with LLMs to provide sourced answers in a conversational format, intended to address issues like misinformation and inaccurate answers. This approach aims to transform the search experience by prioritizing transparency and reliability and represents a shift from traditional search engines towards AI-driven, interactive search solutions.