250 Million Answers per Month
Before ChatGPT's launch, former OpenAI research scientist Aravind Srinivas founded Perplexity, positioning it as an AI search engine using real-time information extracted from the web (including news sites).
As a startup competing with larger, earlier-established rivals, Perplexity is about to complete a $250 million investment round, including SoftBank Vision Fund 2, with its valuation increasing from $1 billion in April to $3 billion.
Existing investors include NVIDIA and Amazon founder Jeff Bezos, as well as AI industry leaders like Andrej Karpathy and Yann LeCun.
Besides investor confidence, Perplexity's monthly revenue and usage have surged sevenfold since the beginning of this year.
According to company insiders, Perplexity's annualized revenue at the start of the year was $5 million (based on the most recent month's sales), but current revenue expectations exceed $35 million.
Statistics show that Perplexity's search engine answered about 250 million questions last month, while the total query volume for 2023 was only 500 million, achieving half of last year's KPI in just one month.
This explosive growth trend highlights that Perplexity is another fastest-growing generative AI application after ChatGPT, although Perplexity's data collection techniques are somewhat controversial.
Perplexity's growth also indicates people's increasing preference for AI-powered search engines—whether products from Google, OpenAI, or Perplexity.
Integrating AI into internet search will not only revolutionize new search methods but may also lead to changes in user search habits.
Core Competitiveness: Focus and Speed
Aiming to become a leader in the search field, the "cultivated" Perplexity faces significant challenges.
Despite rapid growth, Perplexity still lags far behind market leader Google.
As a search engine giant, Google has dominated the market for years, holding over 90% of the global market share and processing about 8.5 billion queries daily.
Moreover, Google has enormous financial resources and vast amounts of data, enabling continuous improvement of its AI search capabilities.
After addressing initial market entry issues, Google's multimodal Gemini model now tops benchmark tests.
However, in Perplexity CEO Aravind Srinivas's view, Perplexity's purpose is not to replace Google but to do things Google overlooks.
As an "answer engine," as their homepage slogan states, Perplexity is "the starting point for knowledge."
Perplexity doesn't aim to be another Google but to change how we find answers on the internet.
Perplexity's core goal and positioning are to satisfy users' curiosity and provide them with the answers they seek.
Users don't care if Perplexity has the most powerful model. They only care about getting good answers.
Perplexity's Chief Business Officer Dmitry Shevelenko states, "Ultimately, small players in the search field have two advantages: speed and focus."
"The Perplexity team only considers one thing: how to quickly answer users' questions. Intense competition makes us more focused on this point."
Having firmly established its positioning and goals, Perplexity is not intimidated by competition from tech giants.
OpenAI's business is diverse and not originally focused on search, so OpenAI doesn't concentrate on answering user search queries through high-quality information sources.
Perplexity focuses on the search track, which is why initial feedback on SearchGPT suggests it doesn't have an advantageous position compared to Perplexity.
Shifting Towards Advertising Revenue
So far, Perplexity's revenue mainly comes from consumer and enterprise subscriptions. Recently, Perplexity announced on its official website that it will introduce advertising on its platform by the end of next month.
Shevelenko stated that Perplexity will share a "double-digit" percentage of revenue with news publishers for each sponsored article, and they have already signed agreements with companies like Time Magazine, Der Spiegel, and Fortune.
As part of this program, publishers will also gain access to the Perplexity API for creating custom "answer engines" and "enterprise pro" accounts.
Additionally, Perplexity will provide all employees of these publishers with one year of enterprise pro product usage, featuring enhanced data privacy and security functions.
Within two weeks of launching the partnership, 50 publishers have requested to join the program, and Perplexity hopes to include as wide a range of websites as possible.
However, before announcing recent collaborations with publishers, Perplexity was accused of plagiarism by media outlets like Forbes and Wired in June.
These companies criticized Perplexity for copying content without clear source links and scraping information from websites that explicitly block crawlers.
Perplexity subsequently modified its user interface to make citations more prominent and took measures to ensure that response interface content wouldn't be a "mishmash" of information scraped from other websites.
To make Perplexity a competitive search engine, maintaining operations with a good business model is essential.
From a long-term perspective, based on Perplexity's positioning and specific company situation, revenue sharing is a more effective approach than one-time payments. However, OpenAI has adopted a one-time payment scheme.
Another difference between Perplexity and Google or OpenAI is that Perplexity doesn't build its own AI large models.
Instead, it obtains licenses to use AI systems from companies like OpenAI.
Like many potential Google competitors, Perplexity's search engine was initially supported by a licensed version of Microsoft Bing's web index but later stopped using Bing as its core system.
Although using technical support from various engines, Perplexity has always maintained its proprietary search index and ranking system.
Earlier this year, a Perplexity employee stated that compared to traditional search engines like Google, Perplexity has more professional and reliable information sources, better suited for search needs in journalism and academia.
Training with unreliable junk information can only yield large amounts of junk information, which has been a problem plaguing most companies. Therefore, using more diverse data sources is necessary when training models.
However, some have raised concerns that introducing advertising might deter users, who may question whether the search environment and results are trustworthy. Advertisements could make the webpage appear less professional and credible.