Perplexity Explained: How Its AI Search Answers Questions Differently from Google and ChatGPT

Human Verified 0

Share to: , and

A person asking a question to a laptop whose screen shows a single clean answer with small citation markers, soft daylight, modern home office. No readable text, no logos.

Perplexity Explained: How Its AI Search Answers Questions Differently from Google and ChatGPT

Perplexity is an AI answer engine. You ask a question in plain language, it searches the live web for that question, and it writes back a single conversational answer with numbered citations that link to the original sources. It is not a search engine in the Google sense — it does not hand you ten blue links and leave you to do the reading — and it is not a chatbot in the ChatGPT sense, because it does not answer from what it remembers being trained on. Every answer starts with a fresh search. That one difference shapes everything else about the product: its strengths, its price, its browser, and the ways it can still get things wrong. Checked 15 November 2026 against Perplexity’s own help centre, pricing pages and API documentation, plus Google’s and OpenAI’s own help pages for the comparisons.

Two corrections belong at the top of this guide, because most explainers still get them wrong. First, the model family called Sonar — Sonar, Sonar Pro, Sonar Reasoning and Sonar Deep Research — is retired. Perplexity ended support for Sonar Chat Completions on 27 September 2026 and replaced it with the Agent API. Any guide that still lists the Sonar lineup as Perplexity’s current models is describing a product that no longer exists. Second, Comet, Perplexity’s browser, is free. It launched for Max subscribers on 9 July 2025, but it has been free for everyone worldwide since 2 October 2025. Any guide that still says Comet costs $200 a month is a year out of date. Both corrections are documented in full later in this guide, with dates and sources.

The short version, before the detail, is this. Perplexity’s free plan answers unlimited basic questions, with 3 Pro Searches a day and 1 Research query a month, and it picks the model for you. Pro costs $17 a month when billed annually ($204 a year) or $20 a month on monthly billing, and buys manual model choice across 7 models, image and video generation, file uploads, and a monthly allowance of deeper searches that Perplexity describes only as “average use” — it publishes no exact number. Max costs $167 a month when billed annually ($2,000 a year) or $200 a month on monthly billing, and buys 11 models, the Model Council, 10,000 Computer credits a month, and the highest limits Perplexity sells. Comet, the browser, is free to download, but its agent features are metered by plan. Perplexity publishes no accuracy rate for its answers, so the honest way to use it is as a fast first draft of an answer with the sources attached — and to open those sources before you trust anything important. This guide explains how the answer pipeline works, how to check citations, what each plan actually buys, how Perplexity compares with Google AI Overviews and ChatGPT Search, and who should pay and who should stay free.

A person asking a question to a laptop whose screen shows a single clean answer with small citation markers, soft daylight, modern home office. No readable text, no logos.

What Perplexity Is — and What It Is Not

<img src="

A clean modern workspace with a laptop screen glowing with a single concise answer panel and blurred browser tabs behind it — one answer, not ten links.

” alt=”A clean modern workspace with a laptop screen glowing with a single concise answer panel and blurred browser tabs behind it — one answer, not ten links.”>

Perplexity AI, Inc. is a San Francisco company founded in 2022. Its product, Perplexity, is described in its own help centre as “an AI-powered search engine” that returns “conversational answers backed by verifiable sources” — but the company’s own framing, and the way the product actually behaves, is better captured by a different label: an answer engine. The distinction is not marketing. A search engine’s job is to find pages. A chatbot’s job is to talk. An answer engine’s job is to read the pages for you and write the answer, with the evidence attached. Perplexity does that one job, and it does it on every single question, including the ones a chatbot would answer from memory.

That makes Perplexity different from Google in a specific, testable way. Google Search returns a ranked list of links; its AI Overviews feature adds a summary at the top of some results, but only when Google’s systems judge the summary “additive to classic Search” — and often it does not trigger at all. Perplexity has no “sometimes” mode. Every question triggers a live search, and every answer is written from what that search found. The trade-off is that you give up the control Google gives you: on Google you choose which result to open, weigh the sources yourself, and decide what to believe. On Perplexity you read one synthesised answer first, and the sources sit behind numbered markers that you have to choose to click. The product is faster than Google for a straight question and slower to verify than Google for a question where the answer matters.

It is also different from ChatGPT, in the opposite direction. ChatGPT is a conversation-first product: it decides, question by question, whether your question would benefit from current information, and if it decides it would not, it answers from its training data with no citations at all. Perplexity never answers from training data alone. It searches first, every time, and cites what it found. That makes Perplexity the better tool for questions about things that happened after a model’s training, about prices and figures that change, and about anything where you want to see the source. It makes ChatGPT the better tool for long-form writing, brainstorming, coding and extended conversation, where the search is incidental to the work rather than the work itself. The two products overlap, but they are built around different defaults, and the rest of this guide is largely about what follows from those defaults.

Who is Perplexity for? The honest answer is narrower than the marketing. It is for people who ask factual questions all day and want the answer plus the source in one place: researchers, students, journalists, analysts, and anyone who fact-checks claims for a living. It is for people who would otherwise open five tabs to answer a question and would rather open one. It is not, on the evidence of its own pricing and limits, a replacement for a general-purpose chatbot, a writing tool, or a coding assistant. It does some of those things — it generates images and video on paid plans, it creates documents and apps, it has an agentic mode called Computer — but those are additions to an answer engine, not the reason to buy one. If you want one answer with sources, Perplexity is the most direct tool for the job. If you want a long conversation, a draft essay, or a coding partner, the tools built for those jobs are still better at them.

One more thing to fix before the detail begins. Perplexity is sometimes described as a “Google killer”. It is not, and the comparison in this guide does not pretend it is. Google answers billions of questions a day across shopping, local results, maps, images and video; Perplexity answers text questions with text answers. The interesting question is not whether Perplexity replaces Google. It is when an answer with citations beats a list of links, when it does not, and what each plan costs for the privilege. That is the question this guide answers, section by section.

How Perplexity Answers a Question, Step by Step

An abstract pipeline visual: a question mark flowing through three soft glowing stages into a tidy cited answer card, blue-teal palette. No words.

Perplexity’s help centre, in an article updated on 12 November 2026, describes the answer pipeline in four steps, and it is worth knowing them because they explain both what the product does well and where it can fail. First, it understands the question with a language model. Second, it searches the internet in real time, gathering from what it calls “authoritative sources like articles, websites, and journals”. Third, it summarises and compiles the relevant insights into an answer. Fourth, it attaches numbered citations linking to the original sources. The whole sequence happens in seconds for an everyday question, and in two to four minutes for a Research report, which performs dozens of searches and reads hundreds of sources. Nothing in the pipeline is a mystery. The mystery, if there is one, is in the parts Perplexity does not publish: which sources it ranks highest, how it decides a source is authoritative, and how often its summary of a real source is wrong. Those gaps matter, and the citations section of this guide deals with them directly.

The first practical consequence of the pipeline is that Perplexity has no memory-only mode. A chatbot can answer “what is the capital of France?” from its training without touching the web. Perplexity searches anyway. That is wasteful for a question with a settled answer, and it is the point for a question whose answer changes: a stock price, an election result, a product launch, a court ruling. The live search is the product. The second consequence is that the quality of an answer is bounded by the quality of the search results. If the live web’s best pages on a topic are thin, outdated or wrong, Perplexity’s answer will be built from thin, outdated or wrong material — and it will still arrive written in the same confident tone as an answer built from excellent sources. The citations are the only way to tell the two apart, which is why this guide spends a whole section on how to read them.

On the free plan, Perplexity picks the model for you. The setting is called “Best”, and it means the product chooses whichever model it judges right for the query. Paid subscribers can choose manually from a model picker — 7 models on Pro, 11 on Max, listed in full in the pricing section below. One point of terminology is worth getting right, because Perplexity’s own pages are careful about it: “Thinking” is a setting applied to a model, not a separate model. If a guide lists “Thinking” alongside model names as though it were a model you could pick, it has misunderstood the interface. For readers who want the background on what choosing between models actually means, our guide to choosing an AI model explains the trade-offs — speed, cost, reasoning depth and context — that sit behind any model picker, Perplexity’s included.

Questions do not live alone in Perplexity; they live in threads. A follow-up question keeps the context of the conversation, so you can ask “and what did the court say about that?” without restating the case, and the answer will be built on the earlier exchange. That is the chatbot half of the product, and it is genuinely useful for research: a first question establishes the topic, a second narrows it, a third asks for the counter-argument. The risk of threads is the same as the risk of any long conversation with an AI — the model can carry a wrong premise from an early answer into a later one, and each new answer inherits the earlier answer’s mistakes. The defence is the same as everywhere else in this product: open the citations, and when an answer matters, ask the follow-up that tests it rather than the one that confirms it.

Perplexity also has a Learn Mode, an interactive study mode that asks you questions back instead of simply answering yours. It is a smaller feature than the marketing suggests, but it is a real one: for students working through a topic, being questioned is more useful than being handed a paragraph. Research mode, covered in the depth section below, is the heavier version of the same idea — a long report rather than a short answer. And on paid plans, Perplexity can create documents and apps from an answer, and its Computer mode can take agentic actions. Those are covered in the pricing sections, because they are priced separately from the search itself.

Citations: Perplexity’s Best Feature — and How to Check Them

A close-up of a hand tapping a small numbered reference marker beside a paragraph on a tablet, with source documents fanned behind it.

Citations are the reason to use Perplexity, and they are also the feature most often misunderstood. Every answer carries numbered inline citations, and each number opens the source page it points to. That is a genuine improvement on a chatbot that answers with no sources at all: you can see where a claim came from, and you can check it. But a citation proves less than it appears to. It proves that a source exists and that Perplexity presented it in support of the answer. It does not prove that the source actually says what the answer claims it says, that the source is reliable, or that the answer has not left out the part of the source that contradicts it. Perplexity publishes no accuracy rate and no citation-correctness rate for its answers, and this guide states no figure of its own, because no vendor-published figure exists. The honest framing is the one OpenAI itself uses for ChatGPT Search: “Search results and citations can be incomplete, outdated, or incorrect.” That warning applies to Perplexity’s citations exactly as much as it applies to OpenAI’s.

Checking a citation is a small skill, and it is worth learning properly because it is the whole value of the product. First, click the number, not the answer. Read the source page itself, and look for the sentence the answer is summarising. Second, ask whether the source actually supports the claim, or merely mentions the same topic. A citation to a page about a company’s revenue does not support a claim about its profit, and a citation to a press release does not support a claim about what independent analysts think. Third, check the date on the source. Perplexity searches the live web, but it can cite an old page for a new question, and an old figure presented as current is one of the commonest errors in AI answers. Fourth, check who published the source. Perplexity marks some cited domains with Government, Academic or Trusted labels, which is a useful first filter — but the label system is Perplexity’s own judgement, and Perplexity publishes no formula for how a source earns a label or how sources are ranked against each other. A label is a hint, not a guarantee.

Two limits of the citation system deserve to be stated plainly. The first is paywalls. Perplexity can cite a paywalled source, but it cannot always read the full text behind the paywall, so an answer built partly on paywalled material may be summarising a headline, a snippet or an abstract rather than the article itself. Some premium data — PitchBook, Statista, FactSet, Crunchbase and similar licensed databases — is available through Perplexity’s licensed data connectors, with some of it open to all users and some of it limited to Pro, Max or Enterprise users holding an eligible licence. This guide keeps that description general, because the exact eligibility of each database by plan was not re-read on a Perplexity page during this research. The second limit is source quality. Perplexity’s pipeline gathers from “authoritative sources”, but the live web’s idea of authoritative and yours may differ, and thin or SEO-farmed pages can be cited alongside good ones. A citation to a real page is not a citation to a good page. The number next to a sentence tells you where to look; it does not tell you what you will find.

Used properly, citations change how you should read an answer. Treat the answer as a draft and the citations as the evidence. If a claim matters — a medical fact, a financial figure, a legal statement, a statistic you plan to repeat — open the source and confirm it before you use it. If the citation does not support the claim, say so in a follow-up and ask Perplexity to correct itself; it will often do so, and the correction is itself a useful signal about how much the first answer was worth. If several citations support the same claim from independent sources, your confidence can rise. If one citation supports a claim and that source is a press release from the company the claim is about, your confidence should fall. None of this is difficult. It is simply slower than trusting the answer, which is why most people do not do it — and why this guide insists on it.

The Citation Quality Test: Real Queries, Real Answers

Three devices — a phone, a laptop and a tablet — showing the same question, with a magnifying glass over one answer, in a laboratory testing mood.

This section does not contain test scores, and it says so plainly rather than inventing them. The research for this guide read Perplexity’s own documentation in full, but it did not run a live citation test across Perplexity, Google AI Overviews and ChatGPT Search — and a test that was not run cannot have results. Any guide that publishes citation accuracy percentages, “X out of 10 citations correct” scores, or a ranked winner from a test it never ran is fabricating evidence. Perplexity publishes no accuracy rate of its own, so there is no vendor figure to fall back on either. What this section gives you instead is the test itself: six queries, the method, and the exact things to record, so you can run it in fifteen minutes and see the real results for the questions you actually ask.

Pick six queries that stress different failure modes. First, a breaking-news question — something that happened in the last 48 hours, where freshness is the test. Second, a statistic — a figure that has a single correct value from an official source, such as a central bank rate or a census number, where the test is whether the cited page actually contains the number the answer quotes. Third, a product comparison — two products with different prices and specifications, where the test is whether the answer mixes up the two. Fourth, a health or finance question — a topic where a wrong answer has consequences, and where the test is whether the answer cites a primary authority or a content farm. Fifth, an obscure fact — a question with one correct answer that appears on very few pages, where the test is whether the model guesses when it cannot find the answer. Sixth, a local question — something about your own city or region, where the test is whether the answer is specific or generic. Ask each question the same way in all three tools: Perplexity, Google AI Overviews, and ChatGPT Search with the Search tool switched on. Do not rephrase between tools, and do not ask follow-ups until you have recorded the first answer.

For each answer, record five things. One: how many citations it carries, and how many of them open to a real page. Two: whether the cited page actually supports the claim it is attached to — read the page, do not assume. Three: the date of the newest cited source, and whether it is recent enough for the question. Four: whether the answer says “I don’t know” or hedges when the evidence is thin, or whether it states a confident answer built on weak sources. Five: whether any claim in the answer has no citation at all, and whether that claim is the one that matters most. A useful shorthand is to score each answer on support rather than on style: count the claims that are backed by a source that genuinely supports them, and count the claims that are unbacked or mis-cited. Style is the least informative thing about an AI answer. Support is the most informative, and it is the only thing a citation test can measure honestly.

What you are likely to find, if you run the test, follows from what the documentation already tells us. Perplexity will cite on every question, because it searches every time. Google AI Overviews will sometimes not appear at all, because Overviews only trigger when Google judges them additive — so the test will have missing data on the Google side for some queries, and that missing data is itself a finding. ChatGPT Search will cite when it searches and will not cite when it answers from training data, so its citation count will vary by question in a way that tells you how often it decided a search was needed. The interesting comparison is not the headline count of citations. It is the support rate: of the citations each tool gives you, how many actually back the claim they are attached to, and how many are decoration. Run the test on your own questions and you will have a better answer than any published score, because it will be measured on the questions you actually ask, with the sources you can open yourself. That is the point of the method — and it is the reason this guide refuses to print a score it did not earn.

One practical note for readers who want to dig into the models behind the answers. The model names that appear in Perplexity’s picker — Claude Sonnet 5.5 and Claude Opus 5.5 on the paid tiers, Grok 4.7 on Pro — are the same models covered in our explainer of the Claude model family and our explainer of Grok, which give the background on what each model is and where it comes from. Knowing which model wrote an answer is useful context for a citation test, because different models summarise sources differently — but the test itself is about the citations, not the model, and a well-cited answer from a smaller model beats a poorly cited answer from a frontier one.

Search vs Pro Search vs Research: Picking the Right Depth

Three ascending layers as a metaphor for search depth: a shallow pool, deep water and an ocean floor, in a calm blue gradient.

Perplexity has three depths of search, and the names cause more confusion than the feature does. The first is Search, the everyday mode: one question, one live search, one answer with citations, in seconds. The second is Pro Search, the deeper mode: multiple searches, more sources, more reasoning before the answer is written. The third is Research, formerly called Deep Research: a long report built from dozens of searches and hundreds of sources, delivered in two to four minutes. The names matter less than the choice, and the choice is simpler than the product makes it look: use Search for a quick question, Pro Search for a question where the answer needs weighing, and Research when you want a report rather than an answer.

The single most misunderstood fact about Pro Search is that it is not a toggle. There is no switch to turn it on. On paid plans, Search applies Pro Search automatically when the plan includes it; on the free plan, you get 3 Pro Searches a day, and the product decides when to use them. That design is deliberate — Perplexity would rather decide the depth for you than ask you to decide it yourself — but it has a consequence worth knowing: you cannot force a deep search on a question the product judges simple, and you cannot save your daily Pro Searches for a question that deserves them by switching the mode off for easy questions. The daily limit on the free plan is the only place where the depth choice is visible to the user, and it is visible as a number counting down rather than as a control.

Research is the mode where the depth is visible in the output. A Research query returns a structured report — sections, subsections, citations throughout — rather than a paragraph, and it takes two to four minutes because it is genuinely doing more work: dozens of searches, hundreds of sources read, a synthesis written across them. The free plan includes 1 Research query a month, which is enough to try it once and not enough to use it regularly. Pro and Max include Research within monthly limits that Perplexity describes only as “average use” for Pro and “advanced use” for Max, with no published numbers. That vagueness is a real weakness in Perplexity’s pricing, and this guide states it as such rather than repeating a number from an outdated plan version. The old figure of “300+ Pro searches a day” described an earlier version of the plans and must not be used; the current help-centre tables say “average use” and nothing more.

Two more modes complete the picture. Learn Mode is the interactive study mode: instead of answering your question outright, it asks you questions back, checks your understanding, and guides you through a topic step by step. It is the right mode for a student learning a subject and the wrong mode for someone who wants a quick answer — which is why it is a separate mode rather than the default. And Projects — renamed from Spaces on 30 July 2026 — are folders for grouping related threads, sources and files, so a research project on one topic stays together instead of scattering across your history. Neither mode changes the price. Both change how useful the product is for sustained work rather than single questions.

The practical rule for choosing depth is short. If you can phrase the question in one sentence and the answer is a fact, use Search. If the question has a “compared with what?” or “is it true that?” in it, you want Pro Search, and on the free plan that means spending one of your three daily Pro Searches. If the question is a project — “give me a report on the current state of X, with sources” — use Research, and accept the two to four minutes. And if you find yourself using Research every day, that is the clearest signal in this whole guide that the free plan no longer fits you, because one Research query a month will run out on the first day of the month.

Perplexity vs Google AI Overviews: Answer vs Summary

Two contrasting search scenes side by side: a traditional results list on one side and a single summarised answer card on the other.

The most useful comparison in this guide is also the one most often stated badly. Perplexity and Google AI Overviews are not two versions of the same thing. Perplexity is a product whose whole job is to write you an answer. AI Overviews is a feature inside Google Search that adds a Gemini-powered summary at the top of some results, with supporting links, and then hands you back to the classic list of links. The difference is in who is in charge of the experience: Perplexity controls the whole answer, and Google controls only the summary at the top. Everything else — trigger behaviour, citation style, link diversity, and which tool to use when — follows from that.

Trigger behaviour is the first difference, and it is the most important. Perplexity searches and answers on every question, without exception. Google AI Overviews appear only when Google’s systems judge a summary “additive to classic Search”, and Google’s own documentation says they “often don’t trigger”. That means the same question can produce an AI Overview on Monday and a plain list of links on Tuesday, depending on Google’s judgement of the query, and there is no switch a user can flip to force an Overview to appear. For a reader, the practical effect is simple: Perplexity is predictable, and Google is not. If you want an AI-written answer to every question, Perplexity gives it to you. If you are happy to read the links yourself when no Overview appears, Google’s inconsistency costs you nothing.

Citation style is the second difference. Perplexity attaches numbered inline citations to specific claims in the answer, so you can see which source supports which sentence. AI Overviews show supporting links as chips alongside the summary, tied to the summary as a whole rather than to individual claims. Perplexity’s style is better for verification, because a number next to a sentence is a precise claim about evidence; Google’s style is better for exploration, because the chips invite you to open several pages and read around the topic. Neither style guarantees the summary is faithful to the sources — that is the same gap on both sides, and the only cure is opening the sources yourself.

Link diversity is the third difference, and it is the one where Google’s own documentation is most interesting. AI Overviews and AI Mode use a technique Google calls query fan-out: the system runs multiple related searches across subtopics and combines the results, which produces a wider and more diverse set of links than a classic results page. Perplexity’s pipeline also runs multiple searches for deeper queries, but Perplexity publishes no equivalent description of how it selects and diversifies sources, and no ranking formula at all. What can be said is that both systems now do more than return the top ten results for a single query — and that the diversity of the links is a property of the system, not of the answer, so it is visible only if you open the links and look.

PerplexityGoogle AI Overviews
What it isA standalone answer engine; one answer per questionA feature inside Google Search; a summary above the classic results
When it appearsEvery question, without exceptionOnly when Google judges it “additive to classic Search”; often does not trigger
CitationsNumbered inline citations tied to specific claimsSupporting-link chips tied to the summary as a whole
Link diversityMultiple searches on deeper queries; no published selection formulaQuery fan-out across subtopics; wider, more diverse link set than classic results
Page eligibilityAny page Perplexity’s search finds and can readIndexed and eligible to show in Google Search with a snippet; no additional technical requirements, no special markup needed
CostFree tier, then Pro $17/month billed annually or Max $167/month billed annuallyFree, inside Google Search
Best forA single sourced answer to a specific questionExploring a topic across many sources, with the summary as a starting point

Which to use when is a question of habit as much as of capability. Use Perplexity when you have a specific question and you want the answer written for you with the sources attached: a fact to check, a comparison to make, a figure to confirm. Use Google when you want to explore — when the question is open-ended, when you want to see how different sources frame the same topic, or when you want the full web’s view rather than one synthesis. And use Google’s AI Mode, the separate conversational surface for multi-step reasoning and comparison questions, when you want a conversation inside Google rather than a summary on top of it. Google’s own documentation notes that AI Mode and AI Overviews “may use different models and techniques, so the set of responses and links they show will vary” — so even inside Google, the two surfaces are not the same product. For readers who want the wider picture of how Google’s AI search features fit together, our guide to Google’s AI search features covers AI Overviews and AI Mode in detail, and our explainer of the Gemini model family covers the models behind them. The comparison that matters for site owners — how to be the page that gets cited — is in the final sections of this guide, and the framework behind it is set out in our comparison of AEO, GEO and SEO.

Perplexity vs ChatGPT Search: Always-On Search vs Sometimes Search

A conversation thread on one side and a live web search feed on the other, connected by a thin bridge — chat versus search.

ChatGPT Search launched on 31 October 2024 and opened to signed-out users on 5 February 2025. It is available on Free, Go, Plus, Pro, Business, Enterprise and Edu plans. Its design is the mirror image of Perplexity’s. Where Perplexity searches every time, ChatGPT Search decides, question by question, whether your question would benefit from current information. OpenAI’s own help page puts it plainly: the product “may search the web automatically when your question would benefit from current information”. If it decides the question does not need the web, it answers from its training data — with no citations at all. You can force a search with the Search tool, but the default is a judgement call the product makes for you, silently.

That single design choice explains most of the practical difference between the two tools. Ask ChatGPT a question about something that happened last week and it will usually search, and its answer will carry inline citations and a Sources panel. Ask it a question about a settled topic and it will often answer from memory, and you will get a confident paragraph with no evidence attached. Ask Perplexity the same two questions and both answers will carry citations, because both triggered a search. The trade-off is real in both directions. ChatGPT’s approach is faster and cheaper for the product to run, and it is better suited to a conversation where most questions do not need the web. Perplexity’s approach is slower and more expensive to run, and it is better suited to a workflow where every answer needs a source. Neither is wrong. They are built for different defaults, and the right choice depends on which default matches your work.

The pipelines differ in the details, too. When ChatGPT does search, it rewrites your query into one or more targeted queries, sends them to search providers — OpenAI’s help page links the privacy policies of Microsoft and Shopify as the providers — runs follow-up queries if needed, and then answers with inline citations and a Sources panel. Perplexity’s pipeline is described in the same shape: understand, search, synthesise, cite. The difference is not in the steps but in the transparency. Perplexity’s numbered citations are tied to individual claims, which makes verification a matter of clicking a number. ChatGPT’s Sources panel lists the pages used, which makes verification a matter of opening the panel and matching pages to claims yourself. Both work. Perplexity’s is the more precise instrument, and for a research workflow that precision is worth having.

Both companies warn, in their own words, that citations are not a guarantee. OpenAI’s help page states that “Search results and citations can be incomplete, outdated, or incorrect.” Perplexity publishes no accuracy figure, and the same warning applies to its answers by the same logic. The warning is not a reason to avoid either tool. It is a reason to treat both tools the same way: as fast first drafts with the evidence attached, to be checked before anything important depends on them. Site owners should also know that ChatGPT Search has its own eligibility rule: a site must allow the OAI-SearchBot crawler, and placement is not guaranteed — OpenAI says it “ranks search results using multiple factors” that it does not publish. Perplexity’s equivalent rule is covered in the getting-cited section at the end of this guide.

For the rest of the comparison — writing, coding, images, voice, memory, and the long conversation — ChatGPT is the stronger product, and this guide does not pretend otherwise. ChatGPT is a general-purpose assistant that happens to have search; Perplexity is a search product that happens to have a chat interface. If your work is mostly conversation, drafting and coding, ChatGPT is the tool to reach for first, and our guide to using ChatGPT is the place to start. If your work is mostly checking facts and reading sources, Perplexity is the tool to reach for first. And if you want the background on the company behind ChatGPT Search, our OpenAI explainer covers the models, the products and the pricing structure that ChatGPT Search sits inside. The verdict section at the end of this guide puts the choice by use case, because “which is better” is the wrong question and “better for what” is the right one.

Perplexity Free vs Pro: What the $17 Actually Buys

A small open gate and a larger unlocked doorway side by side — a modest tier and a premium tier.

Pro is the tier most readers will consider, and the pricing is simple to state but easy to misstate, so this section states it with the billing basis attached to every figure. Pro costs $17 a month when billed annually — $204 a year — or $20 a month on monthly billing. Those are two prices for the same plan, not two plans, and any guide that quotes “$17 a month” without saying “when billed annually” is quoting the cheaper figure without the condition that earns it. The annual price was read live on perplexity.ai/pro on 15 November 2026; the monthly price comes from Perplexity’s own help-centre articles, including its October 2026 price table, which lists Pro at $20 a month and Max at $200 a month. Prices were checked 15 November 2026, and Perplexity can change them; the in-app pricing page is the authority if the figures here have moved.

FreePro — $17/month billed annually ($20/month monthly)
Price$0$17/month billed annually; $20/month on monthly billing
Basic searchesPractically unlimitedUnlimited
Pro Searches3 per dayWeekly limits, described only as “average use” — no number published
Research queries1 per monthMonthly limits, described only as “average use” — no number published
Model choiceNo — “Best” only; Perplexity picks the modelYes — manual picker, 7 models (see below)
Image and video generationNo image generationYes, within monthly limits
File uploadsBasic, limitedExpanded, with file and app creation
Comet Assistant queriesNoneMonthly limits, described as “average/advanced use”
Computer creditsNone4,000 bonus credits (limited time, about 30-day expiry); no monthly allocation after the bonus — credits then cost extra at 100 credits = $1
Data privacyQueries may be used for AI training; no opt-outOpt-out available
Comet Plus$5/month standaloneIncluded

The most important thing about that table is the row that has no number in it. Perplexity does not publish exact Pro usage numbers. It describes Pro’s Pro Search allowance as weekly limits for “average use” and its Research allowance as monthly limits for “average use”, and it publishes no figures behind those phrases. That is a genuine gap in the pricing, and it matters, because the difference between Free and Pro is largely a difference in how much deep searching you can do — and Perplexity will not tell you, in numbers, how much that is. Do not trust any guide that quotes a precise Pro Search limit for the current plans. The old figure of “300+ Pro searches a day” described an earlier version of the plans and is outdated. The current help-centre tables say “average use” and nothing more, and this guide repeats only what the tables say.

What Pro does buy, in concrete terms, is choice and capability. The model picker is the clearest of the two: on Pro you can choose from 7 models — GPT-6.1 Sol from OpenAI, Gemini 3.8 Flash from Google, Claude Sonnet 5.5 from Anthropic, Kimi K3 from Moonshot AI, GLM 5.3 from Z.ai, Grok 4.7 from xAI, and Nemotron 3.5 Super VL from NVIDIA. That list is a snapshot from Perplexity’s help centre dated 11 November 2026, and it comes with two caveats that belong in any honest account. First, model names change fast, and Perplexity’s own pages disagree with each other: the Pro pricing page read on 15 November 2026 names Gemini 4 Hydrogen where the 11 November help-centre snapshot names Gemini 4 Carbon, and the API documentation’s examples name GPT-5.6 Terra and GPT-6 Luna. This guide uses the 11 November help-centre snapshot as canonical and says so, because the help centre is the page Perplexity itself calls the source of truth — “the model selector in Perplexity is always the source of truth”. Second, the snapshot is, in Perplexity’s own words, “not a commitment to keep any individual model available”. A model on the list today can be gone next month. If you are choosing a plan because of a specific model, check the in-app selector before you pay.

The second concrete thing Pro buys is generation: image and video generation, file uploads with document and app creation, and access to Perplexity’s Computer mode for agentic work. The third is Comet Plus, included in Pro rather than charged at its $5-a-month standalone price. The fourth, and the one most buyers overlook, is the privacy opt-out: on consumer plans, queries may be used for AI training unless the user opts out, and the opt-out is a paid-plan feature. On the free plan there is no opt-out. Enterprise plans are different — data is never used for training — and the enterprise section of the pricing table below covers that.

The Computer credits deserve a paragraph of their own, because they are the closest thing to a trap in Perplexity’s pricing. Pro includes 4,000 bonus Computer credits as a limited-time offer, with about a 30-day expiry — and no monthly credit allocation after the bonus runs out. Once the bonus is gone, Computer work costs extra, at 100 credits to $1. That means Pro’s agentic features are a trial, not a subscription benefit: you get a month of agent work to try, and then you pay per use. Max is different — it includes 10,000 Computer credits every month — and the difference is one of the real reasons to consider Max over Pro, covered in the next section. If you are buying Pro for Computer, read that sentence twice. If you are buying Pro for searching, the credits do not matter, and the question is simply whether 3 Pro Searches a day and 1 Research query a month are enough for you.

Who should pay, and who should stay free? Stay free if you ask a handful of factual questions a day, if you rarely need a deep search, and if you do not care which model answers. The free plan’s practically unlimited basic searches cover that use completely, and the 3 daily Pro Searches are enough for the occasional hard question. Pay for Pro if you research for a living — a student, a journalist, an analyst, a founder doing due diligence — and you hit the free limits every week. Pay for Pro if you want to choose your model, generate images and video, or opt out of training. Do not pay for Pro for the Computer credits, because they run out. And do not pay for Pro expecting a published usage number, because there is not one — you are buying “average use”, and you will discover what that means by using it. For readers weighing Perplexity’s tiers against ChatGPT’s, our comparison of ChatGPT’s Free, Plus and Pro tiers is the parallel guide, and our guide to choosing an AI model covers the model-picker decision that sits at the centre of the Pro purchase. Students and educators should also know about Education Pro, at $10 a month with SheerID verification, which Perplexity’s plan article describes as including unlimited Pro Searches.

Perplexity Max: Who Needs the $167 Tier

A researcher's desk at night with multiple monitors, notebooks and a coffee cup, in warm lamp light.

Max is Perplexity’s most expensive consumer tier, and its pricing follows the same two-figure pattern as Pro: $167 a month when billed annually — $2,000 a year — or $200 a month on monthly billing. The annual figure was read live on perplexity.ai/pro on 15 November 2026; the monthly figure comes from Perplexity’s help-centre articles. The gap between the two is $33 a month, or $396 a year, which is the price of paying monthly instead of annually. That is a large premium for flexibility, and it is worth stating plainly, because the annual figure is the one most guides quote and the monthly figure is the one most buyers will actually pay if they do not commit for a year.

What Max buys over Pro is, in order of importance: more models, more credits, a second opinion on every question, and the highest limits Perplexity sells. The model list grows from 7 to 11, adding four frontier models: GPT-6 Astra from OpenAI, Gemini 4 Carbon from Google, Claude Opus 5.5 from Anthropic, and Claude Fable 5.2 from Anthropic. Those four are the most capable models in their respective families, and they are the reason a researcher who needs the strongest available reasoning would pay for Max. The same naming caveat applies as for Pro — the 11 November 2026 help-centre snapshot is canonical, Perplexity’s own pages disagree on names (Gemini 4 Hydrogen on the Pro page, Gemini 4 Carbon in the snapshot), and the in-app selector is the source of truth. The snapshot is not a commitment to keep any individual model available.

The second thing Max buys is the Model Council, available on Max and Enterprise Max on the web only. Several models answer the same question, and their answers are synthesised into one. It is a genuine capability rather than a gimmick: for a question where the models disagree, the Council shows you the disagreement instead of hiding it behind a single confident answer. The third is Brain, a memory system currently in Research Preview and free during the preview, which carries context across conversations. The fourth is the Computer credit allocation: 10,000 credits every month, plus 35,000 bonus credits, against Pro’s one-off 4,000 bonus with no monthly allocation. That is the difference between agent work as a subscription benefit and agent work as a trial. The fifth is priority access to new features and dedicated support, and the highest usage limits Perplexity sells — again described in words (“advanced use”) rather than numbers, because Perplexity publishes no exact Max figures either.

Enterprise deserves a paragraph here, because the enterprise tiers are where Perplexity’s pricing finally becomes numeric. Enterprise Pro costs $40 per seat a month, or $400 per seat a year, and buys 400 Pro Searches a week, 50 Research queries a month, 80 Comet Assistant queries a month, and 500 Computer credits per seat a month, with data never used for training, single sign-on and SOC 2 Type II compliance. Enterprise Max costs $325 per seat a month, or $3,250 per seat a year, and buys 4,000 Pro Searches a week, 500 Research queries a month, 800 Comet Assistant queries a month, 15,000 Computer credits a month, 15 video generations a month (8 seconds, 16:9, with audio), 10,000 personal files, and premium security including SCIM, audit logs and configurable retention. Those are the only Perplexity plans where the limits are stated as numbers, and the contrast with the consumer plans is instructive: businesses get numbers, consumers get adjectives.

The honest verdict on Max is narrow. Buy it if you are a professional researcher, an analyst, or a team lead whose work is deep research every day — the frontier models, the Model Council and the monthly Computer credits are a real working toolkit for that person. Buy it if you need the highest limits and you have already hit Pro’s ceiling repeatedly. Do not buy it for casual use, for the occasional hard question, or for the brand value of the top tier: Pro covers the casual researcher completely, and the free plan covers the casual questioner. Do not buy it expecting a published usage number, because Max’s limits are described in the same words as Pro’s. And do not buy it on the strength of a comparison with ChatGPT Pro or Google AI Ultra, because the prices of those tiers were not verified against the vendors’ own pricing pages during this research — multiple secondary pricing reads in November 2026 corroborate figures for ChatGPT Pro and Google AI Ultra, but this guide does not repeat prices it did not read on the vendor’s own page. Compare the features: ChatGPT Pro buys a general-purpose assistant with search attached, Google AI Ultra buys Google’s most capable models with the highest limits, and Max buys an answer engine with the deepest search and the most models. For the tier-by-tier comparison on the ChatGPT side, our ChatGPT tiers guide sets out what each ChatGPT plan costs and includes, and it is the right companion to this section if you are choosing between the two ecosystems rather than within one.

Comet: Perplexity’s Browser, Explained

A sleek modern browser window floating over a desk scene, with a small assistant orb beside the tabs and a comet streak in the sky outside a window.

Comet is Perplexity’s browser, and the first thing to say about it is the thing most explainers still get wrong: it is free. Comet is a Chromium-based browser with an AI assistant in every tab. It launched on 9 July 2025 for Max subscribers, and it has been free for everyone worldwide, with no waitlist, since 2 October 2025, when Perplexity announced the change on its own blog. Android followed on 20 November 2025, iPhone on 18 March 2026, and Comet Enterprise — a version for organisations — launched in March 2026. It runs on Mac, Windows, iOS and Android, requires iOS 18 or later on iPhones, and has no Linux version. Any guide that still says Comet costs $200 a month is describing the launch pricing from July 2025, not the product as it exists in November 2026. One secondary source claimed the paywall was removed on 29 July 2026; that date conflicts with Perplexity’s own blog announcement of 2 October 2025 and with the help centre, and this guide treats the July date as wrong.

What Comet does is best understood in two halves: the browser half and the agent half. The browser half is a normal Chromium browser — tabs, bookmarks, extensions, the web as you know it — with an assistant sidebar that can summarise the page you are on, answer questions about it, and chat with your open tabs. The agent half is Comet Assistant, which can take actions: fill in forms, book things, shop, and carry out multi-step tasks across the web on your behalf. The distinction matters because the two halves are priced differently. The browser itself is free. The agent is metered by plan: the free plan gets no Comet Assistant queries at all, Pro and Max get monthly limits described as “average” and “advanced” use, Enterprise Pro gets 80 a month, and Enterprise Max gets 800 a month. That metering is the real price of Comet. Downloading it costs nothing. Using its agent costs plan allowance, and on the free plan the agent is not available at all.

Comet Plus is the publisher side of the browser, and it is worth understanding because it is Perplexity’s answer to the question of who pays for the journalism its answers are built on. Comet Plus costs $5 a month as a standalone subscription and is included in Pro and Max. Perplexity says 80% of Comet Plus revenue goes to participating publishers, from an initial pool of $42.5 million announced in August 2025, with launch partners including Condé Nast titles, CNN, the Washington Post, the LA Times, Fortune, Le Monde and Le Figaro. Payouts are based on three kinds of traffic: human visits, search citations, and agent actions. That is a real attempt to share revenue with the publishers whose work the answer engine summarises, and it is more than most AI companies offer. It is also, honestly, an unproven model: no individual publisher payouts have been disclosed, and Comet Plus subscriber counts are not published, so nobody outside Perplexity can say how much a citation is worth to a publisher. The earlier Publishers’ Program from July 2024 shared advertising revenue; that model’s engine changed when Perplexity stopped pursuing advertising, a change reported in February 2026 and not confirmed on a Perplexity page — so this guide states the advertising point only as reported, not as Perplexity-confirmed.

The security question about Comet is a real one, and it deserves a straight answer. Agentic browsers — browsers that can act on your behalf — create a new class of risk called indirect prompt injection, where malicious instructions hidden in a web page can trick the agent into doing something you did not ask for. Brave’s security team disclosed such a flaw in Comet in August 2025, and Perplexity shipped mitigations including a detector it calls BrowseSafe. This guide reports the disclosure and the mitigation as they were reported, and it does not state a current vulnerability status, because no current status was verified during this research. The practical advice is the same advice that applies to any agent with access to your accounts: the assistant pauses before sending messages, changing calendars, placing final orders, or acting on unknown details — use those pauses, read what it is about to do, and do not let an agent act on a page you have not looked at. Comet’s agent is powerful precisely because it can act; that power is the risk, and the pauses are the control.

Who should switch to Comet, and who should not? Switch if you already use Perplexity every day, if you want the assistant in every tab rather than in a separate window, and if you are on Pro or Max so the agent is actually available to you. Do not switch if you are on the free plan and the agent is the feature you want, because the free plan gets no agent queries — you would be switching browsers for a feature you cannot use. Do not switch if you need a Linux browser, because there is not one. And do not switch on the assumption that Comet will do your shopping or booking without supervision: it is an agent with guardrails and pauses, not an autopilot, and the right way to use it is as a fast assistant that shows you what it is about to do before it does it.

Where Perplexity Gets It Wrong: Accuracy, Limits, and Blind Spots

A magnifying glass revealing a small crack in an otherwise polished answer card, in muted, cautionary tones.

The most important limit in this guide is also the simplest: Perplexity publishes no accuracy rate for its answers. There is no vendor figure for how often its answers are right, how often its citations support the claims they are attached to, or how often it hallucinates. That absence is not a criticism of Perplexity alone — no major AI answer product publishes a verified accuracy rate — but it is a fact a reader needs, because it means there is no number to anchor your trust to. Your trust has to be earned answer by answer, by opening the citations. Any guide that quotes an accuracy percentage for Perplexity is quoting a third-party study, and this guide quotes none, because no study’s method was verified during this research.

The commonest failure mode is not invention but misreading. A citation proves a source exists; it does not prove the summary is faithful. Perplexity’s pipeline reads a source, compresses it into a sentence or two, and attaches a number — and in that compression, the qualifier can drop out, the date can be lost, or the claim can be broadened beyond what the source actually says. A source that says “one study found X in a small sample” becomes an answer that says “X is true”. A source from 2023 becomes an answer about 2026. A source that reports a company’s own claim becomes an answer that states the claim as fact. These are not dramatic failures. They are quiet ones, and they are the reason the citations section of this guide insists on reading the source rather than the answer. The confident tone of the answer is a property of the writing, not of the evidence.

The second failure mode is source quality. Perplexity searches the live web, and the live web contains excellent sources and terrible ones, often on the same topic. Thin content, SEO-farmed pages, content written to rank rather than to inform, and pages that repeat a claim without ever checking it can all be cited, and a citation to a real page does not tell you whether the page is good. Perplexity marks some domains with Government, Academic or Trusted labels, which helps at the margins, but it publishes no ranking formula and no list of what qualifies, so the labels are a hint rather than a system. The practical defence is the same as always: look at who published the source, and be suspicious of any answer whose citations are all to pages you have never heard of, or all to pages that agree with each other suspiciously well.

The third failure mode is recency against correctness. Perplexity’s strength is that it searches the live web, and its weakness is that the live web’s newest page is not always its best page. For a breaking story, the first reports are often wrong in the details, and an answer built from them will be wrong in the same details — confidently, with citations. For a settled topic, the newest page may be a rewrite of an old one, and the answer may cite the rewrite rather than the original. There is no fix for this inside the product. The fix is outside it: for anything that matters, check the date on the source, and prefer the primary source — the court filing, the company statement, the official statistics release — over the report about it.

Two more limits belong in this section. The first is paywalls, covered in the citations section: Perplexity can cite a page it cannot fully read, and an answer built on a headline or a snippet is weaker than it looks. The second is privacy. On consumer plans, your queries may be used for AI training unless you opt out, and the opt-out is a paid-plan feature — free users have no opt-out. Enterprise plans are different: data is never used for training. If you ask Perplexity about sensitive topics — health, finance, legal matters — know that on a free account those questions may be used for training, and consider a paid plan or a different tool for the questions you would not want used that way. And a final note on disputes: Perplexity has faced legal challenges from publishers and from Amazon over Comet’s automated shopping, with a US District Court granting a preliminary injunction in March 2026. The status of those cases after March 2026 was not verified during this research, so this guide states the March 2026 position and no further — readers who need the current status should check the court record or current reporting rather than relying on this guide.

Perplexity for Developers: The Sonar Retirement and the Agent API

A developer workspace with abstract blocks of code on a monitor and an API plug and cable motif, in a dark theme.

This section is deliberately short, because the API gets its own guide — a dedicated post on the Perplexity Sonar API will cover the developer story in full, and this section is the summary for readers who only need to know what changed and what replaced it. The change is the most important accuracy point in this whole guide, so it is stated here without softening: the Sonar model family as a consumer and API lineup is retired. Perplexity ended support for Sonar Chat Completions on 27 September 2026. Its current API documentation no longer lists Sonar, Sonar Pro, Sonar Reasoning Pro or Sonar Deep Research as priced products. Any guide, tutorial, code sample or comparison that still presents those four models as Perplexity’s current lineup is describing a product that no longer exists, and any code built on the Sonar Chat Completions endpoint needs to be migrated.

What replaced it is the Agent API, with effort presets instead of model names: fast, low, medium, high and xhigh. The mapping from the old names is roughly Sonar to fast, Sonar Pro to low, Sonar Reasoning Pro to medium, and Sonar Deep Research to high — but it is a rough mapping, not a rename, because the Agent API is a different design: you choose how much effort the agent spends, and the API chooses the model and the tools. Alongside the Agent API, Perplexity now offers a standalone Search API, an Orchestrator API, an Embeddings API and a Decisions API. The model perplexity/sonar still exists inside the Agent API, at $0.25 per million input tokens and $2.50 per million output tokens — the name survives as a model identifier inside the new API, not as a product in its own right. That distinction is the one that trips people up: “Sonar” as a model string still appears in the docs, and “Sonar” as a product lineup is gone.

The pricing is pay-as-you-go, and it is billed in two parts: per-token charges at each model’s published rate, plus tool fees for what the agent does. The tool fees, from Perplexity’s API pricing page updated 13 November 2026, are: web search $2.50 per 1,000 calls, Fast Search $1 per 1,000 calls, fetching a URL $0.50 per 1,000 calls, people and finance search $5 per 1,000 calls, and a sandbox session $0.03 each. The Search API costs $5 per 1,000 calls, or $1 for the Fast variant. Embeddings cost $0.004 to $0.05 per million tokens, and the Decisions API costs $0.01 to $0.02 per million input tokens. The Agent API resells third-party models — OpenAI, Anthropic, Google, xAI, Z.ai, Moonshot and NVIDIA — at their published rates, so the model choice on the API is a choice among other companies’ models, orchestrated by Perplexity’s search and agent layer. Developers who want the full story, including the migration path from Sonar Chat Completions, should read the dedicated API guide when it is published; this section is the correction, not the manual.

The practical consequence for anyone who is not a developer is simple: if a guide tells you Perplexity’s models are “Sonar, Sonar Pro, Sonar Reasoning and Sonar Deep Research”, stop reading it. It was written before 27 September 2026 and never updated. The current consumer model list is the 11-model snapshot in the pricing sections above, and the current API is the Agent API with effort presets. Getting this wrong would date an article on publication day, which is why this guide puts the correction in its introduction, in its pricing sections, and here — three times, in three different forms, so that no reader finishes the guide believing the old lineup is current.

Getting Cited by Perplexity: What Site Owners Should Know

A single website page rising above a crowd of similar pages toward a spotlight beam.

Perplexity is now a source of traffic and visibility in its own right, and site owners are asking the same question they asked about Google twenty years ago: how do I get cited? The honest answer starts with what Perplexity does not publish. There is no published ranking formula for which sources get cited, no list of ranking factors, and no way to buy a citation. The product’s help centre says it gathers from “authoritative sources like articles, websites, and journals”, and it marks some domains with Government, Academic or Trusted labels — but it does not say how a page earns a label, how sources are weighted, or what makes one page win over another on the same topic. Any guide that claims to know Perplexity’s citation ranking formula is guessing. This section states what is known, and it does not pretend to know the rest.

What is known starts with the basics. A page must be indexed and findable by Perplexity’s search for it to be cited at all, and a site must allow Perplexity’s crawler, PerplexityBot, to read it. Blocking the crawler is a choice some publishers make for principled reasons — if you do not want your content summarised, blocking the bot is the effective way to prevent it — but it is a choice with a cost: a blocked page cannot be cited, and a site that cannot be cited does not appear in Perplexity’s answers. The decision is a trade-off between control and visibility, and it is one each publisher has to make for itself. There is no middle setting where a page is readable by humans but invisible to the answer engine; if the bot can read it, the answer engine can cite it.

Beyond the basics, the factors that help are the same ones that have always helped a page be useful: freshness, depth, and original sourcing. Fresh pages win for questions where freshness matters, because Perplexity searches the live web and its answers are built from what the search returns today. Deep, well-sourced pages win for questions where depth matters, because a page that answers a question thoroughly, with its own sources, is more useful to a synthesis than a page that repeats the question. And original reporting, original data and original analysis win most of all, because an answer engine that summarises the web has nothing to summarise if the web only repeats itself. The pages Perplexity cites most are the pages that add something no other page has: the primary document, the interview, the dataset, the measurement. That is not a ranking formula. It is the logic of the product, and it is the reason the advice in this section is the same advice good publishers have always followed.

The wider framework for this work is generative engine optimisation — GEO — and answer engine optimisation — AEO — and this site’s coverage of both is the right next reading. Our GEO explainer sets out how AI answer engines find, evaluate and cite pages, and our comparison of AEO, GEO and SEO explains how the three disciplines relate: SEO for the ranked list, AEO for the answer box, GEO for the answer engine. The practical summary for a site owner is short. Make pages that are indexed, readable by PerplexityBot, fresh, deep, and built on original sources. Do not chase a formula that is not published. And measure the result the only way that matters: ask Perplexity the questions your readers ask, and see whether your pages appear in the citations. If they do not, the answer is almost always in the page, not in the algorithm — the page is not fresh enough, not deep enough, or not original enough, and the fix is editorial, not technical.

The Verdict: Who Should Use Perplexity — and When to Use Google or ChatGPT Instead

Three distinct paths in a landscape converging at a viewpoint, in soft morning light.

There is no overall winner in this guide, and there should not be. Perplexity, Google and ChatGPT are built for different defaults, and the right tool is the one whose default matches the question in front of you. The verdict below is by use case, because a verdict by product would be a verdict about nothing. Read it as a set of rules of thumb, and test them against your own work — the citation test in the fourth section of this guide is the way to do that, and it takes fifteen minutes.

For daily fact-checking, use Perplexity Free. A handful of factual questions a day, each answered with citations you can open, is exactly what the free plan is built for, and the 3 Pro Searches a day cover the occasional hard question. For deep research — a report, a literature review, a due-diligence task, a question with a “compared with what?” in it — use Perplexity Pro at $17 a month when billed annually or $20 a month on monthly billing. The model choice, the Research reports and the image generation are a working toolkit for that person, and the price is the price of a streaming subscription. For professional research teams, use Perplexity Max or Enterprise: Max at $167 a month when billed annually for the frontier models, the Model Council and the monthly Computer credits, and Enterprise Pro at $40 per seat a month or Enterprise Max at $325 per seat a month when the numbers, the compliance and the training-data guarantee matter.

For shopping and local questions, use Google. Perplexity answers text questions with text answers; it does not have Google’s maps, local listings, shopping results, images and video, and it is the wrong tool for a question whose answer is a place, a price in a shop, or a route. For long-form writing and coding, use ChatGPT or Claude. Perplexity can generate documents and apps, and its Computer mode can take actions, but writing and coding are not what an answer engine is for, and the tools built for those jobs are better at them. For exploring a topic across many sources, use Google with AI Overviews as a starting point, and open the links. For a question where the answer matters — a medical fact, a financial figure, a legal statement — use Perplexity for the first draft, open every citation, and confirm the claim in the primary source before you repeat it. No AI answer, from any product, is a substitute for that step.

If you only pick one tool, pick the one that matches the question you ask most often. If most of your questions are factual and you want the source with the answer, pick Perplexity Free and pay for Pro when the limits start to bite. If most of your questions are open-ended, local, or about shopping, pick Google and let AI Overviews do what they do. If most of your work is writing, coding and conversation, pick ChatGPT and use its search when you need it. And whichever you pick, remember the two facts this guide opened with, because they are the two facts most guides still get wrong: the Sonar lineup is retired, replaced by the Agent API on 27 September 2026, and Comet is free, free worldwide since 2 October 2025. A guide that gets those two facts wrong is a guide that has not been updated in a year, and it is not a guide to trust on anything else. Checked 15 November 2026.

Sources

All sources below were read in full on 15 November 2026, except where noted. Prices, limits, model names and dates in this guide come from these pages. No competitor or comparison-site pages were used as sources. Checked 15 November 2026.

  • Perplexity Help Center — “How does Perplexity work?” (updated 12 November 2026) — https://www.perplexity.ai/help-center/en/articles/10352895-how-does-perplexity-work
  • Perplexity Help Center — “Which Perplexity Subscription Plan is right for you?” (updated 12 November 2026) — https://www.perplexity.ai/help-center/en/articles/11187416-which-subscription-plan-is-right-for-you
  • Perplexity — Pro pricing page — https://www.perplexity.ai/pro
  • Perplexity Help Center — “What advanced AI models are included in my subscription?” (snapshot 11 November 2026) — https://www.perplexity.ai/help-center/en/articles/10354919-what-advanced-ai-models-are-included-in-my-subscription
  • Perplexity API docs — Pricing (updated 13 November 2026) — https://docs.perplexity.ai/docs/getting-started/pricing
  • Perplexity Help Center — Enterprise pricing FAQ (updated 6 November 2026) — https://www.perplexity.ai/help-center/en/articles/10352986-enterprise-pricing-and-billing-frequently-asked-questions
  • Perplexity Help Center — “What is Enterprise Max?” (updated 21 October 2026) — https://www.perplexity.ai/help-center/en/articles/12310544-what-is-enterprise-max
  • Perplexity Help Center — Slack connector article (updated 13 November 2026) — https://www.perplexity.ai/help-center/en/articles/12167980-using-the-connector-for-slack
  • Google Search Central — “AI features and your website” (updated December 2025) — https://developers.google.com/search/docs/appearance/ai-features
  • OpenAI Help Center — “Searching the web with ChatGPT” — https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt
  • Perplexity blog — “Comet is now available to everyone worldwide” (2 October 2025) — https://www.perplexity.ai/hub/blog/comet-is-now-available-to-everyone-worldwide — cited from search results and corroborated by the help centre; full text not re-fetched for this guide.
  • Perplexity — AI costs hub article (October 2026 prices) — https://www.perplexity.ai/hub/blog/ai-costs

Share to: , and

Faheem R. Khattak

Senior automation engineer and AI marketing journo covering LLMs, algorithms, analytics, core updates, Bing search, and technical SEO.

0 Comments

Leave a Reply

Optional Take Tag:

🛡️ Spam protection: Honeypot + Time-Trap Active

This site uses Akismet to reduce spam. Learn how your comment data is processed.