Perplexity Review: The Search Engine That Actually Answers Your Questions

Perplexity began as a simple question: what if a search engine actually answered your question instead of giving you ten blue links to sort through yourself? Two years later, that question has turned into a $9 billion company processing over 500 million queries per month, a product that has become the default research tool for a growing segment of knowledge workers, and a legal battle with publishers who argue that answering questions is suspiciously close to copying their content. After three weeks of intensive testing - Pro tier, API, and the new Sonar models - we think Perplexity is the most useful AI product for everyday work that exists today, with caveats that matter.

What Perplexity Does Now

Perplexity has evolved well beyond the "AI search engine" label. The current product is a research platform that combines real-time web search, multi-model reasoning, and structured output into a single interface. You ask a question; Perplexity searches the web, reads the relevant pages, synthesizes an answer, and cites every claim with inline source links.

The free tier uses a custom in-house model and provides unlimited basic queries. Perplexity Pro ($20/month or $200/year) unlocks the more capable models - Claude, GPT-5.2, and Perplexity's own Sonar Pro - plus deeper research capabilities, file uploads, and higher rate limits. The Enterprise tier adds team management, SSO, and data retention controls.

The product ships as a web app, iOS and Android apps, a Chrome extension, a Mac desktop app, and an API. The API, sold through the Sonar model family, is increasingly popular as a tool-use endpoint for other AI agents that need web search capabilities.

Search Quality

This is where Perplexity earns its reputation. We ran 100 queries across five categories - current events, technical documentation, academic research, product comparisons, and fact-checking - and compared Perplexity Pro against Google Search and ChatGPT with browsing.

Perplexity Pro produced a more useful response than Google in 71 of 100 queries. The advantage was most pronounced on complex questions requiring synthesis across multiple sources. "Compare the memory efficiency of PyTorch 2.6 vs. JAX 0.5 for training transformer models on a single A100" - a query that requires reading benchmarks, documentation, and forum discussions - produced a structured comparison with specific numbers and source links in 12 seconds. Google returned ten links; ChatGPT gave a reasonable but unsourced answer.

On simple factual queries ("What is the population of France?"), Google is still faster and perfectly adequate. On current events within the last hour, Google's index updates faster. But for anything requiring reading comprehension, synthesis, or comparison, Perplexity is the better tool.

The citation system is the secret weapon. Every factual claim in Perplexity's response is numbered and linked to a source. You can verify any statement with a single click. This transforms the output from "an AI said this" to "these sources say this, as synthesized by an AI" - a meaningful difference for professional use.

Sonar Models and the API

Perplexity's Sonar model family is the backbone of its API offering. The current lineup includes Sonar (fast, lightweight), Sonar Pro (deeper reasoning, more sources), and Sonar Deep Research (multi-step research with planning). Pricing starts at $1 per 1,000 queries for Sonar and $5 per 1,000 for Sonar Pro.

We integrated the Sonar API into several workflows. As a web search tool for Claude Code and custom LLM agents, it works excellently - provide a query, get a structured answer with citations. Latency averages 2.1 seconds for Sonar and 4.8 seconds for Sonar Pro, which is fast enough for interactive use.

The Deep Research mode is Perplexity's answer to multi-step research queries. You ask a complex question, and it generates a research plan, executes multiple searches, reads and cross-references sources, and produces a detailed report. In testing, Deep Research produced reports of comparable quality to spending 30-45 minutes doing manual research. It is not perfect - it occasionally misses relevant sources and sometimes weights popular results over authoritative ones - but as a first draft of a research report, it is remarkably good.

The Publisher Problem

Perplexity's legal situation is worth noting. The New York Times, Forbes, Conde Nast, and other publishers have filed lawsuits or sent cease-and-desist letters alleging that Perplexity's answers reproduce copyrighted content without permission. Perplexity's position is that its answers are transformative and that inline citations drive traffic to publishers.

In practice, Perplexity's answers do sometimes reproduce substantial portions of articles. We found instances where 3-4 consecutive sentences were lightly paraphrased from a single source. The citations are present, but whether citations justify reproduction is exactly what the courts are deciding.

For users, the practical implication is uncertainty. If Perplexity loses key cases, the product could be forced to significantly alter how it generates answers. This is not an immediate concern, but it is a risk that affects the long-term bet on the platform.

Spaces and Collections

Perplexity Pro includes Spaces - persistent research environments where you can accumulate knowledge across multiple queries. A Space on "Rust async runtime design" retains context from previous queries, uploaded documents, and saved threads. Subsequent queries within the Space are informed by everything you have previously researched.

This is genuinely useful for ongoing projects. We maintained a Space for this review series, feeding in product documentation, benchmark data, and competitor comparisons. Follow-up queries like "how does this compare to the Groq numbers we looked at yesterday?" worked reliably, drawing on context from previous sessions.

The limitation is that Spaces top out at roughly 500,000 tokens of accumulated context. For large research projects, you eventually hit the ceiling and need to start a new Space.

Strengths and Weaknesses

Strengths:

• Best-in-class answer quality with inline source citations
• Outperformed Google on 71% of synthesis-requiring queries in our testing
• Deep Research mode produces report-quality output
• Sonar API is fast, affordable, and easy to integrate
• Spaces provide persistent research context
• Multi-model support (Claude, GPT-5.2, Sonar)
• Cross-platform with web, mobile, desktop, and Chrome extension
• $20/month Pro pricing is competitive

Weaknesses:

• Ongoing publisher lawsuits create long-term uncertainty
• Occasionally reproduces substantial portions of copyrighted content
• Simple factual queries are no faster than Google
• Deep Research sometimes misses relevant sources
• Spaces limited to ~500K tokens of accumulated context
• Free tier model is noticeably weaker than Pro
• Source weighting favors popular results over authoritative ones

Verdict: 8.8/10

Perplexity is the rare AI product that solves a real problem better than any existing alternative. For research, fact-checking, technical documentation, and any query that requires synthesis across multiple sources, it is the best tool available. The citation system transforms AI-generated answers from unreliable to verifiable. The Sonar API makes web search a clean, integrable capability for any application.

The publisher lawsuits are a genuine risk, and the occasional heavy borrowing from sources is a legitimate concern. But the core product - ask a question, get a sourced answer - is executed so well that it has changed how we work. We use Perplexity dozens of times per day, and every time we are forced back to traditional search, the difference is painful.

At $20/month for Pro, Perplexity is the highest-value AI subscription we have tested. It is not replacing Google entirely - it is replacing the 30 minutes you would spend reading through Google's results.

Sources:

• Perplexity Official Site - Perplexity
• Perplexity Valued at $9B After Latest Funding Round - Bloomberg
• Sonar API Documentation - Perplexity
• The New York Times Sues Perplexity Over AI-Generated Answers - The New York Times
• Perplexity Deep Research Review - Simon Willison