Perplexity AI, an artificial intelligence search start-up, has increased its monthly revenues and usage seven-fold since the start of the year, after closing a new $250mn round of funding.
The AI-powered search engine answered roughly 250mn questions in the last month, compared with 500mn queries for the whole of 2023, Dmitry Shevelenko, Perplexity’s chief business officer told the Financial Times.
The new figures underscore Perplexity’s position as one of the fastest-growing generative AI applications to emerge since OpenAI’s ChatGPT launched to huge acclaim in November 2022, despite controversy over the start-up’s data-gathering techniques.
San Francisco-based Perplexity, which was founded by former Google intern Aravind Srinivas just three months before ChatGPT launched, uses AI software to answer questions, using information pulled in “real time” from the web, including news websites.
Perplexity started the year with $5mn in annualised revenues — a projection of full-year revenues based on extrapolating the most recent month’s sales — and is now making more than $35mn on the same basis, according to a company insider.
Now, the start-up is pivoting its business model from subscriptions to advertising, bringing it into closer competition with Google, which dominates the $300bn search ads industry.
Its growth comes as Google steps up its integration of AI features into its core search product and OpenAI launches SearchGPT, a prototype AI search tool available to roughly 10,000 testers.
“At the end of the day the smaller player in the space has two advantages: velocity and focus,” Shevelenko said. “Our users and team only think about one thing when it comes to Perplexity: a place you get your questions answered. Competition sharpens our focus even more.”
To fuel its fight against larger rivals, Perplexity recently closed a new $250mn investment from investors including SoftBank’s Vision Fund 2, said people familiar with the deal, tripling its valuation from $1bn in April to $3bn. Bloomberg previously reported on the funding negotiations.
Its existing investors include AI chipmaker Nvidia and Amazon founder Jeff Bezos, as well as several prominent names from the AI industry, such as OpenAI co-founder Andrej Karpathy and Meta’s chief AI scientist Yann LeCun.
Shevelenko said Perplexity is not daunted by competition from better-resourced tech companies, including Microsoft-backed OpenAI, which makes the world’s most-popular AI chatbot.
“OpenAI are doing so many different things . . . They’re not focused on answering people’s questions with high-quality sources,” Shevelenko said. “That’s why side-by-side feedback of SearchGPT says it doesn’t stack up favourably to Perplexity.”
Perplexity’s revenues have been mainly from its consumer and enterprise subscriptions but the start-up recently announced it would introduce advertising on to its platform by the end of next month.
“Unlike OpenAI, we always knew our main monetisation engine was going to be advertising,” Shevelenko said.
It will split a “double-digit” percentage of its revenues on every sponsored article with news publishers cited, said Shevelenko. It has signed deals with Time, Der Spiegel and Fortune, among others.
However, prior to announcing its recent partnerships with publishers, Perplexity in June was accused of plagiarism by publications Forbes and Wired, which criticised the start-up’s reproduction of stories without clear attribution and its scraping of websites that had explicitly blocked its crawlers.
Shevelenko acknowledged the allegations and said the company had taken on board the criticism. Perplexity has subsequently made changes to its user interface to make citations more prominent and taken steps to ensure its responses do not summarise any websites.
He adds 50 publishers have asked to join Perplexity’s revenue-sharing programme in the two weeks since its launch. The company hopes to include as wide a pool of websites as possible.
“For Perplexity to be a useful product on the open web, there need to be good business models for publishing new and updated facts about the world,” Shevelenko said. “If you want to align incentives [with journalism] in the long term, revenue-sharing is a more powerful way of doing so than one-time lump sum payments, which is the route OpenAI went.”
Unlike Google and OpenAI, Perplexity does not build its own AI models, which has become increasingly costly. Instead it licences a combination of AI systems from the likes of OpenAI and others.
Perplexity’s search engine was originally powered by a licensed version of Microsoft’s Bing index of the web, like many would-be Google competitors. But Shevelenko said it no longer uses Bing as its core system.
“We have our own proprietary search index and ranking system,” Shevelenko said. “We use signals from all kinds of engines but we have our own crawler and ranking system.”
The platform has targeted journalism and academia because of their vast amounts of reliable information and data. One person who worked for Perplexity earlier this year said it saw that source material as its advantage over traditional search engines such as Google, which draws on a much wider range of websites.
“Trash in and trash out is a problem that plagues companies, so you need to consult a better variety of sources when training models,” this person said.
However, this person suggested the introduction of advertising could deter users: “There is an untrustworthy environment when you see ads. People are a little sceptical of results on Google now.”
Google, which a US judge this week ruled was a “monopolist” in a landmark antitrust case, has fended off many attempts to challenge its dominance of search over the past 20 years.
Nonetheless, Joseph Teasdale, head of tech at Enders Analysis, a consultancy, said the AI search market is “hotting up”.
“The risk from AI is . . . that general web search as a whole is made redundant by new ways of matching users with information, products and services,” he said. The “big unknown”, however, is whether it can be reliable enough for mainstream use.
“AI is stubbornly prone to confabulation,” Teasdale said. “At the scale of billions of queries per day, serious failures are inevitable.”
