The Perplexity Playbook: Why Google Still Can't Beat a 250-Person Startup

On Thursday, I told you how Perplexity got built, the boy from Chennai who missed his CS branch change by 0.01 GPA, read a book about Google in a London research library, interned at the three best AI labs on earth, and came home to build an answer engine none of them had bothered to make.

If you missed that one, go read it first. Today's issue hits differently if you know where this started.

Today is the teardown. How the product actually works. Why Google with 25 years of infrastructure, 180,000 employees, and the most profitable advertising business ever built still hasn't properly responded. And what any builder can steal from how Perplexity won a category nobody thought existed.

Let's get into it.

📌 No sponsor this week - Projects of AI just relaunched. If your product belongs in front of 2,400+ AI founders and builders with a 69% open rate, grab a slot here →

The insight that sounds obvious but wasn't

Here's the thing about the "answer engine vs search engine" framing that most coverage gets wrong.

Perplexity didn't build a better search engine. They didn't compete with Google on Google's terms faster indexing, better ranking, more relevant links. They looked at what Google was actually doing and asked a different question entirely.

Google's job, structurally, is to send you away. You arrive, type a query, and Google's entire business model is built around pointing you at ten other websites. The more you click, the more ad impressions get served. Google has a fundamental incentive to not answer your question because the moment you get your answer, you leave, and the transaction is over.

Perplexity is built on the exact opposite premise. You ask a question, the engine searches the web in real time, reads multiple sources simultaneously, and delivers one synthesized answer with clickable citations. The job is to end your search, not extend it.

Think of it this way: Google hands you a grocery bag full of ingredients. Perplexity hands you a finished meal.

That's the product insight. But the business insight underneath it is what makes this story interesting: Perplexity bet that a meaningful segment of users researchers, analysts, builders, operators, would pay a subscription for that finished meal rather than deal with the grocery bag anymore.

They were right.

How the product actually works

Most people who haven't used Perplexity assume it's just ChatGPT with search bolted on. It's not. The architecture is meaningfully different.

When you submit a query, Perplexity runs a multi-step process:

Step 1 - Query decomposition. The system breaks your question into sub-queries, identifying what information it actually needs to answer properly.

Step 2 - Real-time retrieval. It searches billions of web pages in real time, identifying the most relevant sources based on authority, recency, and relevance. Not a cached index - live retrieval at query time.

Step 3 - Synthesis. Multiple sources are read simultaneously, cross-referenced, and synthesized into a single coherent answer.

Step 4 - Citations. Every factual claim is tagged with a numbered source you can verify. Not buried at the bottom integrated into the answer itself.

For research-heavy tasks, Perplexity reduces work that takes 30 minutes on Google to under 2 minutes. That's not a marketing claim, it's why enterprise teams are switching.

One venture capital firm uses Perplexity's Deep Research to generate preliminary due diligence reports on startups work that previously took an analyst half a day now takes twenty minutes. That's the kind of productivity delta that makes a $200/month subscription feel cheap.

The three-tier product that does three different jobs

What most people miss about Perplexity is that it's not one product. It's three, stacked on top of each other, each serving a different job.

.Tier 1 - The Answer Engine (Free)

The base product. Type a question, get a cited answer in seconds. Processes over 1 billion queries per month in 2026.

This is the free tier and it's genuinely good. Good enough that 100 million monthly active users use it without paying. The free tier isn't a crippled demo. It's a fully functional product that builds the habit, creates the dependency, and eventually pushes power users upward.

Tier 2 - Perplexity Pro ($20/month)

Access to GPT-4o, Claude, and Gemini 2.5 Pro - user's choice of model. 300 Pro searches daily. File analysis. Image generation. Internal knowledge search.

The key feature here isn't any single capability, it's the Model Council. Pro users can pick which AI model powers their queries - the best model for the job, not a fixed choice. Perplexity becomes the interface layer on top of the entire AI model ecosystem. You're not buying access to one AI, you're buying access to all of them through a single, research-optimized interface.

Tier 3 - Perplexity Max ($200/month)

This is where it gets serious.

Deep Research runs on Claude Opus 4.5 for Max users. It doesn't just search once, it searches dozens of times across hundreds of sources, reads full articles, compares data points, flags contradictions, and produces a structured report. A typical Deep Research query visits 100+ web pages in 2–5 minutes and delivers a structured report with sections and citations.

As of March 2026, Deep Research can generate deliverables directly - PowerPoint presentations, spreadsheets, dashboards, and websites from your research prompts. No more copy-pasting findings into other tools.

That last feature is the one that changes the ROI calculation entirely. It's not a research tool anymore. It's a research-to-output pipeline.

The Comet browser - the most ambitious product bet

In July 2025, Perplexity did something that looked, at first glance, like a distraction.

They built a browser.

Not a search widget. Not a Chrome extension. A full browser Comet available free on macOS, Windows, Android, and iOS. Comet combines Perplexity's search-focused AI with full browser capabilities. It can fill forms, compare products across websites, and complete basic transactions autonomously.

The strategic logic is easy to miss if you think about this as a product decision. It's not. It's a distribution decision.

Google's entire moat is built on Chrome. Google's global search share has dipped below 90% for the first time in 2025, but it's still 90%. The reason? Chrome. Install Chrome, Google is your default search. Uninstall Chrome, almost nobody does. Google's search dominance is downstream of browser dominance.

Perplexity building Comet is Perplexity saying: the only way to permanently displace Google is to own the layer below search.

Comet for Enterprise launched in March 2026 - deployable across macOS and Windows devices via MDM. Enterprise administrators can push Comet silently to every machine in their organisation.

Silent enterprise deployment. That's not a product feature. That's a go-to-market strategy.

Why Google still hasn't properly responded

This is the question I keep coming back to. Google has more data, more engineers, more compute, and more distribution than any company in history. Google launched AI Overviews, their answer to the answer engine and it now reaches 2 billion monthly users.

And yet Perplexity is at $21 billion and growing 335% year over year.

Three reasons Google can't simply copy their way out of this.

The incentive problem. Google is consolidating around an ad-supported AI layer that monetizes commercial intent at billion-user scale. Perplexity formally abandoned advertising in February 2026, going subscription-first to protect user trust. Google cannot kill their ad business to compete with Perplexity. The thing that would make Google genuinely competitive with Perplexity would destroy Google's $237 billion annual revenue. They're trapped by their own business model.

The trust problem. Internal Perplexity research showed that ads caused users to question the objectivity of AI-generated answers, undermining the product's core value. That's why they killed ads. Google's AI answers will always carry the suspicion that commercial intent influenced the result. Perplexity's subscription model is literally its credibility mechanism.

The user problem. Perplexity drives a different kind of traffic than Google, fewer visitors, but visitors who convert at multiples of organic, skew educated and senior, and arrive pre-researched. This is the 1% that matters. Researchers. Analysts. Engineers. VCs. The people who influence what tools their entire organizations use. Losing them to Perplexity now means losing the enterprise contract decisions five years from now.

Three things any builder can steal right now

1. Reframe the job your product does - not the category it's in.

Perplexity didn't position as "AI-powered search" that's Google's category, Google's game, Google's win. They positioned as an "answer engine", a completely different job description. The user isn't searching, they're asking. That single reframe changed who the competition was, what success looked like, and what the product needed to do.

Before you name your product's category, ask: what is the actual job the user is hiring this for? The answer might not be the category you assumed.

2. Build the free tier to create dependency, not just acquisition.

Perplexity's free tier is genuinely good. Not crippled. Not a teaser. A fully functional product that builds a daily habit around asking questions and getting cited answers. By the time a user hits the limits of the free tier, they've already restructured how they research. The upgrade isn't optional at that point, it's a refusal to downgrade.

Most free tiers are designed to frustrate users into paying. Perplexity's is designed to make users dependent before they ever see a paywall.

3. Own the layer below your product.

Comet isn't a product diversification. It's a defensive moat. By owning the browser layer, Perplexity removes Google's ability to use Chrome as a distribution weapon against them. Whatever you're building, ask yourself what the layer below your product is, and whether someone else owning it creates an existential risk. If yes, that's the next thing to build.

Where Perplexity goes from here

Perplexity in 2026 looks fundamentally different from Perplexity in 2025. The product expanded from "answer engine" to a full browser, a deep-research agent, and a multi-platform AI ecosystem.

The 2026 revenue target is $656 million. Management is targeting IPO readiness within two to three years.

The company that launched seven days after ChatGPT, with four people and no LLM of its own, is now building the browser that might eventually be what you use to read this newsletter.

And the boy from Chennai who read a book about Google in a London office at midnight?

He's not done.

That's the Perplexity teardown. Next Tuesday: Runway AI. Three art students at NYU built the video AI tool that Hollywood couldn't live without and they did it by being artists, not engineers. That story is one of my favourite so far.

Hit reply - do you use Perplexity? Has it replaced Google for you yet? I read every reply..

- Projects of AI

P.S. - The detail I keep coming back to: Perplexity tested ads from late 2024, built a whole team around monetising them, then walked away completely in February 2026. The executive quote to the Financial Times says it all: "The challenge with ads is that a user would just start doubting everything. We are in the accuracy business, and the business is giving the truth, the right answers." Monthly revenue jumped 50% the month after they killed ads. Most companies would have found a way to rationalise keeping them. Perplexity deleted them. That kind of conviction is either visionary or reckless. Given the growth curve, I am going with visionary.

Enjoying this? Forward it to one person building in AI. That's the only growth strategy I have.

Want to reach 2400+ AI founders and builders? Sponsor Projects of AI