Perplexity — The Complete Guide

What is Perplexity — in plain English?

You know how Google works. You type a question, you get ten blue links. Then you have to click on one, read through the article, come back, try another, piece together an answer yourself. It works — but it takes time and effort.

Perplexity is different. You type the same question — and instead of giving you links, Perplexity reads those pages itself and writes you a clear, direct answer. Then it shows you the sources underneath so you can check everything.

It is available free at perplexity.ai. You do not need to create an account to start using it.

Who made Perplexity?

Perplexity AI was founded in August 2022 in San Francisco by four people: Aravind Srinivas (CEO), Denis Yarats (CTO), Johnny Ho, and Andy Konwinski.

Their backgrounds are remarkable. Aravind Srinivas had worked as a research scientist at OpenAI — one of the people who helped build the early GPT models. Denis Yarats had been a research scientist at Meta AI. Andy Konwinski co-founded Databricks, one of the most successful data infrastructure companies in the world. These were not first-time founders or students — these were experienced AI researchers and entrepreneurs who saw a specific problem and built a focused solution.

The investors who believed early

Perplexity’s investor list reads like a who’s-who of technology: Jeff Bezos invested personally. NVIDIA invested — unusual for a hardware company to back an AI application startup. Sequoia Capital and NEA led rounds. Former Google CEO Eric Schmidt invested. IVP, Elad Gil, and Nat Friedman (former GitHub CEO) also backed the company early.

By early 2024, Perplexity had raised over $500 million at a valuation approaching $3 billion — remarkable for a company just 18 months old.

The history of Perplexity — from idea to 10 million daily users

The problem they saw

In 2022, when Aravind Srinivas was still at OpenAI, he was already thinking about the next generation of search. The insight was simple but powerful: large language models had become good enough to not just retrieve relevant documents but to synthesise answers from them. The bottleneck was not AI capability — it was the interface. Search engines were still showing links when they could be showing answers.

The specific frustration: when you search something factual, you know what you want. You want the answer. Not ten guesses at which page might have it.

Launch and early growth (2022–2023)

Perplexity launched its beta in December 2022 — just weeks after ChatGPT’s launch. The timing was perfect. ChatGPT had made the public aware that AI could have conversations and give direct answers. Perplexity offered something slightly different: an AI specifically designed for factual, sourced answers to search-style queries.

Early users were enthusiastic. Researchers loved the academic paper integration. Journalists loved being able to fact-check quickly with sourced answers. Students used it for research in a way that felt more trustworthy than asking ChatGPT directly (because every claim had a source attached).

By the end of 2023, Perplexity had tens of millions of monthly users and was growing rapidly.

2024: The year Perplexity challenged Google

In 2024, Perplexity began to be discussed seriously as a threat to Google Search — something that had seemed unthinkable just a year earlier. Several factors contributed:

• Google’s AI Overviews feature launched with embarrassing errors (recommending people eat rocks, suggesting adding glue to pizza) — creating an opening for a competitor perceived as more careful
• Perplexity launched Perplexity Pages — a feature that generates comprehensive, formatted research documents on any topic
• The Deep Research feature launched, enabling multi-step research that could take 30 minutes of human work and produce it in minutes
• Partnerships with telecom companies including SoftBank meant Perplexity was pre-installed on new phones in some markets

Controversy: the publisher issue

2024 also brought significant controversy. Several major publishers — including News Corp, The New York Times (via legal action against OpenAI, which raised related questions), and others — raised concerns about AI search tools summarising their content without compensation. Perplexity was specifically accused of generating answers that closely paraphrased premium content, potentially reducing traffic to original publishers.

Perplexity responded by launching a publisher revenue-sharing programme, but the debate about AI and content economics remained unresolved.

2025–2026: Perplexity as a daily habit

By 2025, Perplexity had over 10 million daily active users. The product expanded significantly: voice search, an iOS and Android app with on-device understanding, integration with third-party tools, and the Perplexity API allowing developers to build applications using Perplexity’s search-grounded AI. The company positioned itself not as a search engine replacement but as an “answer engine” — a distinct category.

What Perplexity can do — real examples for everyone

For a student writing a report

Instead of spending two hours in the library and on Google cobbling together sources, a student types: “What were the main causes of the First World War? I need academic sources for my A-level history essay.” Perplexity gives a structured answer with five cited academic sources — each one linked. The student can then go and read the full sources for their bibliography. Two hours becomes twenty minutes.

For someone following the news

Instead of reading three different news websites and trying to understand what is actually happening, you type: “What is happening with [current news story]? Explain it from the beginning.” Perplexity reads recent news sources and gives you a coherent narrative — with sources from multiple outlets so you can read further.

For a small business owner

You want to know: is [competitor] still in business? What are their current prices? What are customers saying about them recently? Perplexity searches the current web and gives you up-to-date intelligence — far more current than asking ChatGPT, which has a knowledge cutoff.

For a parent

Your child has been diagnosed with something you’ve never heard of. You type the medical term into Perplexity. It gives you a clear explanation, what the treatment typically involves, what questions to ask the doctor, and links to NHS, Mayo Clinic, and peer-reviewed sources. Not scary forum posts. Not outdated articles. Sourced, current, clear.

When to use Perplexity vs ChatGPT vs Google

Getting started right now

1. Go to perplexity.ai — no sign-up required to start
2. Type any question in the search box — write it as a full question, not just keywords
3. Read the answer and check the numbered sources underneath each claim
4. Ask follow-up questions — Perplexity maintains the context of your search thread
5. Use the Focus menu to restrict sources (Academic, News, YouTube, Reddit)

Free vs paid — what do you actually need?

For most people, the free tier is genuinely sufficient for everyday research. The 5 daily Pro searches covers most casual use. Power users — researchers, journalists, analysts, students doing serious academic work — will find Pro worth it, particularly for Deep Research.

Source: perplexity.ai/pro — April 2026

Getting the most from Perplexity

Perplexity rewards specific, well-formed questions far more than keyword searches. The shift from “Google mode” (keywords) to “Perplexity mode” (complete questions with context) is the single most important thing you can do to improve your results.

The Focus filters — use them intentionally

Before searching, consider which Focus mode is right for your task. The default is Web. Switch to:

• Academic — for peer-reviewed research, scientific questions, medical evidence
• News — for current events, breaking stories, recent developments
• YouTube — to find and summarise relevant video content
• Reddit — for community opinions, product experiences, practical tips from real users

20 high-value prompts for Perplexity

How Perplexity works technically: RAG at scale

Perplexity is built on Retrieval-Augmented Generation (RAG) — an architecture that combines the knowledge synthesis capabilities of large language models with real-time information retrieval from the web. Understanding RAG is essential to understanding what Perplexity can and cannot do.

The standard LLM limitation RAG solves

A standard LLM like GPT-4 or Claude has a knowledge cutoff — it cannot know about events that happened after its training data was collected. It also cannot verify facts against current sources; it can only recall patterns from training. RAG addresses both limitations: at query time, it retrieves relevant documents from the current web, then passes those documents as context to the LLM, which synthesises an answer grounded in retrieved content rather than training memory alone.

Perplexity’s RAG pipeline — step by step

When you submit a query to Perplexity, a multi-stage pipeline executes:

The Sonar model family

Perplexity has developed its own model family — the Sonar series — specifically optimised for search-grounded generation. Unlike general-purpose LLMs, Sonar models are fine-tuned on data that trains them to:

• Prioritise information from retrieved context over training-time knowledge when the two conflict
• Generate accurate inline citations that map correctly to source documents
• Express appropriate uncertainty when retrieved sources disagree or are incomplete
• Handle multi-hop reasoning across multiple retrieved documents

The Sonar Pro model is used for Pro searches; Sonar is used for standard searches. Pro users can also choose GPT-4o or Claude as the generation model, with Perplexity’s retrieval layer providing the web grounding.

Deep Research: multi-step agentic retrieval

Perplexity’s Deep Research feature (Pro) goes beyond single-step RAG. It implements an agentic research loop:

1. The initial query is decomposed into a research plan — multiple sub-questions that need to be answered to address the overall topic
2. Each sub-question triggers an independent retrieval-and-synthesis cycle
3. The model evaluates whether the retrieved information is sufficient or whether additional searches are needed (a form of self-critique)
4. All retrieved information is synthesised into a comprehensive research report with full citations

This can involve dozens of retrieval steps and takes several minutes — hence the “30 minutes of human research in minutes” positioning. The output is a structured report with a full bibliography.

Perplexity API — technical reference

The Perplexity API uses an OpenAI-compatible interface, making it straightforward to integrate for developers already using OpenAI’s Python SDK.

from openai import OpenAI

client = OpenAI(
    api_key="your-perplexity-api-key",
    base_url="https://api.perplexity.ai"
)

response = client.chat.completions.create(
    model="sonar-pro",
    messages=[
        {
            "role": "system",
            "content": "Be precise and cite sources for all factual claims."
        },
        {
            "role": "user",
            "content": "What is the current state of AI regulation in the EU?"
        }
    ],
    # Perplexity-specific parameters:
    # search_recency_filter: "hour" | "day" | "week" | "month" | "year"
    # return_citations: True (included by default in sonar models)
)

print(response.choices[0].message.content)

# Access citations if available
if hasattr(response, 'citations'):
    for i, citation in enumerate(response.citations):
        print(f"[{i+1}] {citation}")

Key API parameters

• model — sonar-pro, sonar, sonar-reasoning-pro
• search_recency_filter — Restrict retrieved sources to a time window (hour/day/week/month/year)
• search_domain_filter — Whitelist or blacklist specific domains for retrieval
• return_related_questions — Boolean, returns follow-up questions for the query

Full API documentation: docs.perplexity.ai

Limitations of RAG systems — important to understand

Understanding the failure modes of Perplexity specifically helps you use it more effectively:

• Source quality inheritance — If the top-ranked web sources are wrong, Perplexity’s answer will likely be wrong. It retrieves and synthesises; it does not independently verify claims against ground truth.
• Citation hallucination — Despite grounding, models can occasionally generate a claim that seems cited but either misrepresents the source or maps to the wrong citation number. Always click through and verify important citations.
• Recency vs quality tradeoff — Very recent events may have fewer high-quality sources indexed. Perplexity may return lower-quality sources for breaking news than for well-covered topics.
• Index coverage — Perplexity’s index, while substantial, is smaller than Google’s. For very obscure topics, it may return fewer relevant results.