Perplexity’s AI got some answers wrong, but it helped Aravind Srinivas find his first users

Perplexity did not need a big marketing campaign to get its first users talking. Some people discovered the AI search startup by typing their own social media handles into its search bar. The answers were not always accurate but they were often surprising enough that made people share screenshots online.

That unexpected curiosity helped put Perplexity on the map and gave co-founder Aravind Srinivas an early reason to believe the product could work. The company had started with a different idea: making Twitter easier to search. The response to that experiment eventually pushed the team toward a broader web-search product.

During a conversation with Lex Fridman, Srinivas explained how the early tool attracted attention and why the founders decided to build on what they had learned.

Perplexity’s first experiment focused on Twitter

When Srinivas joined forces with co-founders Dennis and Johnny, the goal was to build useful products with large language models. At the time, the team was still exploring where AI could create the most value for users.

One early project focused on Twitter’s social graph. The team collected tweets through academic API accounts and built a search experience that let people ask questions about the platform’s data.

Users could explore conversations, look for tweets about artificial intelligence and identify connections between people through mutual followers. The system also used language models to turn natural language questions into SQL queries that could run against the database.

Srinivas said the team demonstrated the tool to people including Yann LeCun, Jeff Dean and Andrej Karpathy. Their interest helped the founders attract potential recruits at a time when the startup was still unfamiliar to most people.

Searching for yourself helped the startup spread

The first version of Perplexity’s Twitter search did not cover the entire platform. The team had collected only a portion of Twitter’s data, so some users could not find their accounts in the index.

The company had also released its regular search product about a week apart from the Twitter-search tool. When a Twitter handle was missing from the index, Perplexity could fall back on regular search, retrieve some of that person’s tweets and generate a summary of their social media profile.

The answers sometimes went wrong. Srinivas recalled that the AI could produce amusing descriptions because it hallucinated as some users were startled by what the system appeared to know about them, while others laughed at its mistakes.

They began sharing screenshots in Discord communities & online forums. People who saw the posts wanted to try the tool themselves, helping the little-known startup gain its first wave of attention.

Srinivas shared that the team understood that searching for one’s own profile was largely a one-time activity. It was unlikely to keep users returning on its own. But the response revealed something more important, people were interested in a tool that could retrieve information from links and turn it into a readable summary.

Why Perplexity moved beyond Twitter

The founders also recognised that relying on Twitter data was not a sustainable long-term approach. Their collection method was difficult to scale, and access to the platform’s data was becoming a concern as Elon Musk took over Twitter.

The team shifted its attention toward regular web search. Rather than limiting the product to social media conversations, it could retrieve information from across the internet and answer a much wider range of questions.

That became the foundation of Perplexity’s answer-engine approach. The product combines traditional search with a large language model, using relevant web material to generate answers and provide citations that allow users to check the sources.

An unfamiliar name did not stop early growth

Perplexity launched on December 7, 2022. He initially expected usage to fade, but people continued using the product during the Christmas holiday period, despite the startup being largely unknown and having what he described as an obscure name.

The first version offered a single answer with a summary and citations. Users had to enter a new query to ask another question.

About a week after New Year, the company introduced a conversational version with suggested follow-up questions. Srinivas highlighted that usage then began growing exponentially, with users clicking on related questions to explore topics further.

The early Twitter experiment was not the business Perplexity ultimately set out to build. It helped the founders discover that people wanted more than a search box returning links. They were interested in having information gathered, summarised and presented in a way that made it easier to understand.

That insight helped shape the product’s next step, turning search into a starting point for discovering knowledge.