In February, The New York Times profiled Perplexity, a buzzy year-old startup whose AI-powered search engine is loosening Google’s grip on the market. “Can This AI-Powered Search Engine Replace Google?” the article asked.
Only three months later, Perplexity announced a $62.7 million funding round that valued the company at over $1 billion. It drew heavyweights from OpenAI, Amazon, Nvidia, and other tech giants, cementing Perplexity as a key player in AI.
The company offers an alternative to users frustrated by Google’s ad-heavy and SEO-driven search results. Instead of serving up links, Perplexity aggregates information from credible sources into a single, cohesive answer.
This method, according to CEO Aravind Srinivas, represents “a better way to experience the internet.”
At Jefferies’ 2024 Private Internet Conference in Los Angeles, Srinivas sat down with Gaurav Kiuttur, Global Co-Head of Internet Investment Banking, to discuss the shifting search landscape, Perplexity’s growth, and competition from Google and OpenAI.
Their conversation comes on the heels of Perplexity announcing Enterprise Pro, its first B2B offering that delivers the platform with security and control at enterprise scale.
The following Q&A was lightly edited for clarity and length.
There’s been a ton of buzz around Perplexity over the last year. Can you tell us about your platform and business model?
Perplexity is a conversational answer engine. Instead of serving you ten links, it cuts the signal out of noise and gives you the answer you’re looking for.
Our current business model is subscriptions. A lot of people made fun of us for using subscriptions, asking “why would someone pay for search when Google is free?”
Well, Google is filled with spam, ads, and SEO. For people who value their time, $20/month is a worthwhile investment. We’ve successfully challenged conventional thinking around search, bringing a viable new business model to the space.
For people who haven’t used Perplexity, can you describe the user experience and how it differs from Google?
With Google, you get links. With Perplexity, you get answers.
Perplexity aggregates and resolves several links at once, giving you a summarized answer with citations. And you can ask follow-ups. You can converse with it.
Using Perplexity is like being in dialogue with the world’s smartest person who has all the internet’s knowledge at their fingertips. It’s like Wikipedia and ChatGPT had a kid.
How replicable is the technology? What’s Perplexity’s secret sauce?
Building the prototype itself isn’t that difficult. What’s hard is delivering it at scale.
All the minor details we manage – UX, speed, latency, accuracy, mobile responsiveness, iterative improvements. That, and the scale of users we have.
The way that we tackle hundreds of small details at scale is very hard to recreate. And by the time someone recreates it, we’ll only be further ahead.
Talk about OpenAI. What are the big differences with ChatGPT, their consumer product?
There is a simple way to understand this. For a product like ChatGPT, hallucination is a feature. For a product like Perplexity, where the goal is accurate answers, hallucination is a bug.
If I prompt ChatGPT with a creative task, like writing a poem or a joke, it has to hallucinate. The more it hallucinates, the more entertaining its response.
With Perplexity, we offer something similar in our “writing mode,” but the default functionality is to answer questions with rigor and citations. In that sense, we approach our product differently than ChatGPT.
Note: “Hallucinations” are misleading results generated by AI models, due to flaws in their training data, incorrect assumptions, or biases.
Where does Perplexity get its data? How do you ensure sources yield accurate answers?
From the beginning, we’ve been very mindful about which data sources we allow to contribute to answers. We want high-authority sources of information.
One useful proxy for this is peer review. You can’t publish something to The New York Times or The Wall Street Journal without it being approved by your editor and colleagues. That peer review gives content a high trust score.
Content on Twitter or Medium, on the other hand, is not really reviewed for accuracy. We take these factors into account and create a core of trusted domains on the web.
Then, as the product is used by more users and you get a sense of the best domains, you continue updating those trust scores to build a more and more refined product.
So far, how are you finding the market for users willing to pay for search? Is it big?
It’s a huge market, and we’re just getting started. There are millions of people using our product for free today, which gives us a massive funnel to convert from.
And beyond that, platforms like Google have billions of users, so there are so many more people to acquire at the top of the funnel. There are billions of dollars in revenue to make here.
And are there plans for an ad-supported version?
Definitely.
I’m a big fan of how Google built a very high-margin business through advertising. Unfortunately, it’s come at the cost of their core mission of “organizing the world’s information and making it accessible and useful.” Now, the product is designed more for the advertiser than the user.
We have ideas for integrating ads without compromising our core mission. That might mean building a lower-margin business than Google, but that’s okay with us as long as we are profitable and successful.
Finally, what does Perplexity look like in the next couple of years? How do you think about scaling?
We want to have an order of magnitude more users. Our queries should be more accurate. We should offer more diverse formats of answers – not just paragraphs but knowledge panels, scores, and more. This will be important for the Olympics and the election.
We also want to enable multimodal experiences: voice form factors, image form factors, distribution across several devices and powering every knowledge worker’s day-to-day workflow. People should be able to do more with Perplexity answers. They shouldn’t just read it and walk away.
The AI market needs to transition from chatbots to actually supporting workflows. If Perplexity can succeed at that, we have a really, really good shot at a big chunk of the AI profits.