Could you imagine that you could make a product that was better at Googling than Google?

Well in 2022, Aravind Srinivas, Denis Yarats, Johnny Ho, and Andy Konwinski believed exactly that, and being ex-OpenAI employees helps.

And with ~10M monthly users & 53M website visits in under 2 years, Perplexity is challenging that. It’s recently raised a $73.6M funding round which included Nvidia and Bezos.

After using Perplexity for 3 months, it’s one of the most user-centric AI products out there. I’ve even switched from ChatGPT’s paid plan to Perplexity’s paid plan.

Let’s breakdown how Perplexity is growing so fast despite Google & ChatGPT’s dominance.

Crafting a startup to take on a Googliath

According to Aravind Srinivas, the CEO of Perplexity, there are five areas where AI startups can effectively compete against larger, established companies:

1. Accuracy (how often the AI gets the right answer)

2. Reliability (how consistently the AI gets the right answer)

3. Speed (how fast does the app return an answer)

4. Product UX (how easy can a user get value from the app)

5. Personalization of experience (how the product gets better the more you use it - eg. Spotify recommends better music the more you listen)

Perplexity competes at product UX & speed.

Product UX: A simple focus on “effortless prompts with credible answers”.

Perplexity’s product UX have solved some of the most annoying problems users experience using LLMs like ChatGPT or Claude.

Here are 3 of them below:

Flagship feature: Perplexity Pro helps users write better prompts without any prior knowledge.

I was terrible at prompting ChatGPT when it first came out.

It wasn’t until I dug into prompt engineering blogs & subreddits did I learn how to add context for better prompts. But most people won’t have the patience to learn.

Instead of forcing users to learn how to prompt (a terrible user experience), Perplexity Pro builds adds “prompt clarity” intuitively into the product experience.

How? If a user’s answer is too vague, Pro will ask a follow up question to add context. Then they take in the user’s reply, adds the clarifying prompt to the original prompt to create a better prompt, and returns a more accurate answer to the user.

Perplexity Pro reduces the most intimidating barrier for new users, using the right prompt, and allows users to get the most value out of Perplexity.

Sources drives more credibility in users, discourages hallucinations.

ChatGPT and AI models are notorious for hallucinations. Hallucinations are when the AI model makes up facts.

But Perplexity solves this using “Sources”, displaying the citations in the answer at the top of every query. This gives the user confidence that the query is accurately cited.

Bing has a similar UX. But order matters.

Because of hallucinations, users worry about the factuality of a model’s answers. By putting the sources at the top, Perplexity is saying, “this answer is reliable, accurate, and grounded in truth, check it yourself.”

Focus filters out biased sources.

Imagine you’re looking for new running shoes to buy.

You ask ChatGPT or Bing. Since both of those models are trained on millions of webpages, including blogs or review sites that are paid by that brand to write great reviews, you’re likely to get a biased answer that says “Nike” is the best brand for shoes.

But how do you remove this bias?

Enter “Focus”.

Perplexity’s “Focus” feature makes it easy to filter for sources you want the model to search from, including academic sources, Reddit, and Youtube.

1. When I’m looking for the best brand for shoes, I don’t want paid review sites, I want unsolicited & unhinged user reviews from Reddit.

2. When I’m looking for highlight videos for last night’s basketball game, I don’t want sports articles, I want highlight reels from Youtube.

3. When I’m looking for the best way to build build muscle, I don’t want Quora answers, I want academic papers.

Focus is one of the reasons why I prefer the Perplexity UI over the ChatGPT Plus UI for market research.

Growth: Quick activation flows & retention loops to build habits

Suggested prompts help activate new users.

In traditional SaaS products, there’s usually 1-3 use cases (eg. you go to AirBnb to book a stay).

For AI products, there’s endless use cases.

If users are coming to AI products with different intents, this makes it hard to solve a common activation problem - how do I get the user to get value as fast as possible?

Perplexity uses suggested prompts. By clicking on a suggested prompt, a new user can experience the “aha" moment of Perplexity, get to the value of an answer, with less friction.

“Related” feature get users to develop a habit.

Michael Easter states that a slot machine works on 3 main principles:

1. Opportunity

2. Unexpected rewards

3. Quick repeatability

Perplexity’s related feature works the same way.

It creates the opportunity to roll for more answers (opportunity), with the reward of learning something new (unexpected rewards), and makes it easy to repeat by automatically creating related prompts without needing to think (quick repeatability).

And just like Wikipedia, users can easily find themselves into a rabbit hole - for better or for worse.

Perplexity’s Discover drives repeat use and content loops.

What does Twitter, Reddit, and New York Times have in common? They all have high repeat use. And they’re all in the news business.

News retains.

Built into Perplexity’s web and mobile app is a tab called “Discover”.

It’s an editorially-curated newsfeed, with pre-generated Perplexity answers about today’s most popular topics, even using push notifications to drive users back to the Discover page and a new AI-generated podcast.

But why even dabble in news?

1. Encourage curiosity.

Let’s look at Perplexity’s use case. Currently, users use Perplexity if they have a question or task they need an answer to.

But I could go a few days or week without needing to ask a question.

But with Discover, I come back to Perplexity for the news. And once that news drives new questions that I answer using Perplexity. It’s become as addicting as Reddit.

2. News is a strong viral loop.

News is easy to share, and Discover threads are better news articles.

Perplexity’s threads have the news sources cited at the top. I can investigate the credibility and bias of each piece of news - which I can’t do with other news sites. I’ve shared more Discover threads to friends than Reddit threads/news threads because they’re transparent.

There’s a beautiful content growth loop that allows current Perplexity users to bring more Perplexity users to the platform:

3. Looming publisher problem.

Perplexity has become part search app, part news aggregator.

But there’s a problem. If you’re no longer clicking into webpages, the publisher doesn’t get your traffic. And less traffic means less ad revenue. And less revenue means angry publishers.

When Perplexity gets large enough, how will publishers respond to this? We’ll have to wait and find out.

Perplexity’s strategy to compete with giants

The goal of growth is to get users to repeatedly experience value from the app.

For Perplexity, getting users to the right answer without having to click through a bunch of blue links is the core value.

Brian Balfour, CEO of Reforge and ex-VP of Growth at Hubspot, came up with the concept of reducing input friction to get users of an AI app to value, faster.

We can breakdown input friction into more variables, allowing us to understand how Perplexity reduces input friction:

Perplexity focuses on reducing user prompt ambiguity and user prompt time.

Tech giants like OpenAI & Google will have more resources (compute, talent, data) to refine their models to reduce inference error and inference time. However, smaller startups can focus on specific use cases, reducing user prompt ambiguity & creating UX to reduce user prompt time.

In doing so, Perplexity can repeatedly deliver value to users by reducing input friction.

Closing: Perplexity has a beautifully simple, focused UX, but can’t Google just copy Perplexity’s UX?

Firstly, if Google starts creating AI-generated search results, it would completely destroy Google’s $45B search ads business, which investors and employees at Google wouldn’t tolerate.

Secondly, the amount of compute required for Google to turn its billions of user searches into AI searches would require so much hardware that the current supply chain wouldn’t be able provide it.

Currently, Perplexity’s 10M users pale in comparison to Google’s - so there’s a long way to go. But Perplexity’s beautiful UX, smart engagement loops, and a creative monetization strategy has shown there’s a path for startups to take on tech giants.

[Bonus] Monetization: freemium strategy, a predatory Perplexity Pro bundle, and referrals.

Freemium with limited uses.

Perplexity offers unlimited searches.

But their advanced model called Perplexity Pro offers better search results, if users pay for the Perplexity Pro plan.

However, Perplexity lets free users try Pro, but limits them to 5 searches every 4 hours.

More importantly, Perplexity can retarget these free users across email, paid ads, and in-app notifications to convert users to the Pro plan in the future.

A predatory Perplexity Pro bundle.

If you were on ChatGPT Plus plan, why would you move to Perplexity? Well it’s because with Perplexity, you can use GPT-4, Claude 2, and Perplexity’s model, whenever you want.

Cable companies have done this for decades, bundle highly desirable TV channels with less desirable TV channels. The former will bring attention to the latter.

In this case, ChatGPT is the industry favourite. But even I switched from ChatGPT Plus to Perplexity Pro because Perplexity’s bundle has more models I could choose from.

Referrals for viral growth.

Perplexity has a referral program.

For a product that’s well suited for virality (their Twitter has over 100K followers with tweets gathering thousands of likes and comments), there’s a great fit between referrals and the product.

Got a question? Feedback? I love both.
👇 Drop a comment below.