An AI search engine that summarises the web with clickable citations.
Perplexity Review: The Citation-First Search Engine
Critics often dismiss Perplexity as nothing more than a parasitic interface wrapper that scrapes genuine journalism and passes off stolen summaries as its own work. There is merit to that accusation when bots scrape paywalled publishers without permission or credit, and anyone who relies on original investigative journalism ought to be skeptical of automated scrapers. Yet framing the platform merely as an intellectual property pirate misses why software engineers, postgraduates, boutique agency operators, and independent researchers keep turning to it over standard engines. Traditional search has become clogged with sponsored placements, synthetic affiliate articles, and useless content mills designed purely for ranking algorithms.
Perplexity attempts to resolve that rot by pairing conversational generation with live internet index lookups and immediate citations. Whether an independent developer is tracking down changes between framework updates, a biology student is parsing academic preprints, or a shop proprietor is checking regional commercial compliance rules, the value does not stem from conversational chatter. It comes from having a machine read five modern documents, synthesize an answer, and hand over the exact references used to produce each sentence. It fundamentally changes how one interrogates the open internet.
Think of Perplexity as a hybrid between an answer engine and an LLM aggregator. When you type in a question, it does not just rely on static training data frozen months ago. It pings a search index, pulls down relevant live web pages, reads through them, and generates an answer backed by those specific sources.
Every single claim comes with a little numbered link right beside it. You can click that number, jump straight to the source, and verify if the AI understood the context or just grabbed a stray quote out of context. For anyone whose job requires getting facts right, this single mechanical difference makes standard chatbots feel blind by comparison.
Behind the scenes, Perplexity does not rely solely on its own custom models. While it defaults to its own fine-tuned system (called Sonar, built on top of open-source models), the paid plan lets you swap the underlying engine to flagship models from OpenAI and Anthropic. That flexibility makes it far more than a basic wrapper.
The core interface looks clean, like a minimal search bar with a few toggles underneath. You ask a question, read the synthesized answer, look at the sources lined up across the top, and then use follow-up prompts to dig deeper.
The standout feature is how it presents information. You type in a query, such as asking for recent conversion benchmarks in B2B SaaS, and it gives you a clean summary broken into neat paragraphs.
Above the text, you see small cards showing the websites it scraped. Within the text, tiny superscript numbers link directly to those specific URLs. You can hover over a citation to preview the site, or click through to read the original article. This fixes the biggest headache with generative text tools: the guesswork. If an answer looks weird, you do not have to wonder if the model made it up. You just check citation number three. If citation number three points to a spammy forum comment or an outdated case study, you know to ignore it.
Directly beneath the search bar, you will find a dropdown labeled Focus. This is one of the most practical tools in the entire product because it restricts where Perplexity goes looking for information:
On the search bar, there is a small toggle called Pro Search (formerly known as Copilot). When you turn this on, Perplexity does not just run a single query and dump an answer. It acts like a research assistant.
If your prompt is vague, Pro Search will pause and ask you clarifying questions before running the search. Once it has enough detail, it performs multiple multi-step searches behind the scenes, combining findings from several angles, cross-referencing points, and summarizing the entire investigation. It takes a few extra seconds, but this is the bit I would actually pay for. For complex research projects, the difference in answer quality between regular quick search and Pro Search is massive.
If you are paying for the Pro tier, you do not have to settle for one model. Through the settings menu, you can toggle between several engines:
Being able to switch the model means you can test how Claude interprets a set of search results versus how GPT-4o formats the same data, all without leaving the dashboard or paying for three separate subscriptions.
Perplexity lets you organize your research threads into spaces called Collections. You can set specific system prompts for a Collection, so every query inside it follows your rules. For instance, you can create a collection that forces every output into a three-bullet executive summary with a skeptical tone.
There is also a feature called Perplexity Pages. It takes your research threads and converts them into formatted, shareable articles with headers, callouts, and integrated images. You can edit the text manually, add new sections with specific prompts, and publish a public URL for your team or clients. It is handy, though most serious writers will still end up pasting the findings into Google Docs or Notion anyway.
You do not necessarily have to spend a dime to get value out of this tool, but the line between the tiers is drawn firmly around depth and access.
The free version costs nothing and does not require a credit card. It includes: Unlimited standard quick searches. 5 Pro Searches every four hours. Access to the default Sonar model. Basic web search and citations.
For casual lookup queries or replacing your morning Google habits, the free plan is plenty.
The Pro subscription matches the standard twenty-dollar monthly rate seen across the AI industry. It includes: Over 300 Pro Searches per day. Full access to switch between GPT-4o, Claude 3.5 Sonnet, Claude 3 Opus, and newer Sonar models. Unlimited file uploads for analyzing PDFs, spreadsheets, CSVs, and images. Image generation using tools like DALL-E 3, Stable Diffusion, and Flux. A $5 monthly credit toward Perplexity's API.
If your daily responsibilities demand extensive investigation, hitting that five-query limit on the free plan takes about fifteen minutes. The Pro tier removes that friction entirely.
For companies wanting administrative oversight, Perplexity offers an Enterprise plan: $40 per seat per month, or $400 per seat per year. Single sign-on (SSO) integration. Centralized user management. SOC2 certification and guaranteed data retention controls ensuring company queries are not used for model training.
In my work, Perplexity has taken over roughly seventy percent of the searches I used to run on Google.
If I need to know a simple fact, a stock price, or an address, I still use Google. If I need to buy a pair of running shoes, I use Google because e-commerce filtering is still better there.
But if I need to answer questions like "What are the common churn drivers for mid-market CRM software according to recent post-mortems?" or "Summarize the regulatory changes affecting privacy in California this year," Google is practically useless. It gives you SEO farms written by agencies trying to sell you something. Perplexity bypasses that junk, pulls the actual text from three distinct sources, compiles the summary, and shows me the original links.
It also beats vanilla ChatGPT whenever timeliness matters. ChatGPT can browse the web now, sure, but it is slow, clunky, and prone to failing when a site blocks its crawler. Perplexity was built from scratch around web retrieval, and it shows in how fast the pages load.
First, the speed is outstanding. Even when it runs multi-step searches, it consistently delivers answers faster than ChatGPT's browsing mode.
Second, the citation system builds trust. You never have to blindly accept what the screen says. A quick scan of the cited domains tells you immediately if you are looking at verified reporting or biased promotional copy.
Third, the file analysis inside Pro Search is top-tier. You can drop a 60-page PDF report into the box and ask, "Pull out every mention of customer acquisition cost and cross-reference it with the revenue figures on page 12." Claude 3.5 Sonnet handles those tasks inside Perplexity with pinpoint accuracy.
Fourth, Reddit focus mode saves hours. Finding real human discussions without scrolling past five affiliate blogs is genuinely difficult on modern search engines. Filtering directly through Reddit threads with natural-language synthesis is a massive relief.
Perplexity is not magic, and it has some glaring blind spots.
It can still be fooled by bad sources. If the top three results on the web are wrong, Perplexity's summary will be wrong too. It summarizes what it finds; it does not possess human judgment to tell whether an industry blogger made up a statistic. You still have to look at the tiny citation numbers and vet the domains.
The writing tone can feel repetitive. Sonar tends to write in a very stiff, predictable cadence with lots of introductory fluff and bullet points. If you want high-grade prose, you have to switch the model to Claude 3.5 Sonnet and prompt it aggressively to cut the robotic habits.
Skip it if you primarily need an AI to write creative long-form copy, generate code inside an IDE, or manage multi-step project workflows. Dedicated code editors like Cursor and general writing sandboxes like Claude or ChatGPT are still better suited for raw creation. Perplexity is built to look outward at the world, not to polish your internal drafts.
There are also occasional scraping blocks. Some major publishers block Perplexity's bots. When that happens, the tool will sometimes hallucinate an answer based solely on the URL text or fall back on lower-quality scraper sites that re-publish the same story.
If you use search engines to gather facts, track competitors, or summarize documents, Perplexity AI is worth every penny of the twenty-dollar fee. The free plan is generous enough to test for a week without feeling boxed in, but the Pro search capabilities combined with Claude 3.5 Sonnet make it a fixture on my browser toolbar. It does not replace every AI tool in the stack, but it has completely replaced Google for deep research.
freemium