Perplexity AI: Complete Guide to Search, Research, Comet and Computer

by David Porter

Friday, 31 July 2026 at 08:55

thumbnail_what-is-perplexity-ai-complete

Perplexity is an AI answer engine that searches current sources, synthesizes what it finds and places citations beside its claims. It can answer a quick factual question, conduct a longer research project, analyze uploaded files, browse with an AI assistant, complete multi-step work through an agent and supply search infrastructure to developers.

That makes the 2026 version of Perplexity much broader than the familiar description “Google with an AI summary.”

Perplexity now has four distinct layers:

Search and Research turn questions into sourced answers.
Projects, files and creation tools organize knowledge and produce reports, spreadsheets, presentations, images and other assets.
Comet and Computer move from answering toward browsing and acting.
Enterprise and developer products connect private company knowledge or put Perplexity’s search technology inside other software.

The common thread is retrieval. Perplexity is at its best when a task starts with finding, comparing and checking information. It is less distinctive when no external evidence is needed, and it should never be treated as an infallible source merely because citations are visible.

This cornerstone explains the complete Perplexity ecosystem without turning every feature into a separate tutorial. Where pricing, privacy, the API or a major product needs more depth, it links to a dedicated guide with a different search purpose.

Features, limits, prices and company information were checked on July 24, 2026. Perplexity changes model access and usage allowances frequently, so the account interface and current official documentation remain authoritative.

Perplexity at a glance

What it is	An AI answer, research and agent platform built around current search and citations
Company	Perplexity AI, Inc.
Founded	2022
Headquarters	San Francisco, California
CEO and co-founder	Aravind Srinivas
Core experience	Ask a question, search relevant sources, receive a synthesized answer with links
Main consumer products	Search, Pro Search, Research, Create, Projects, Comet and Computer
Business products	Enterprise Pro, Enterprise Max, connectors and administration
Developer products	Agent API, Sonar API, Search API and embeddings
Free plan	Yes
Entry paid personal plan	Pro at $20 per month or $200 per year in the United States
Mobile access	iOS and Android
Browser	Comet for macOS, Windows, iOS and Android
Publicly traded?	No; Perplexity is privately held
Official website	perplexity.ai

Key takeaways

Perplexity is an answer engine rather than a conventional list-of-links search engine, although it still exposes the sources behind its synthesis.
Its core advantage is not that every answer is correct. It is that current web retrieval and inline citations make many answers faster to inspect.
Search, Pro Search, Research and Create represent increasing depth and effort. The right mode depends on the job; the most expensive mode is not automatically the best.
Perplexity can use several frontier model families. The product’s search, retrieval and citation system matters at least as much as the selected language model.
Projects combine instructions, collaborators, uploaded material and saved research. They are useful knowledge spaces, but not a substitute for the original document repository.
Comet is a Chromium-based browser with an assistant that can understand and act across open pages. Browser agents require stricter permission and confirmation habits than ordinary chat.
Computer is a cloud-based digital worker that can coordinate tools, connectors and subagents to produce outcomes rather than only answers. Its work consumes credits in addition to the subscription.
Consumer searches may be used to improve Perplexity’s models when the account setting is enabled. Enterprise data is handled under different commitments.
A citation is a route to evidence, not a quality certificate. Important claims still require checking the source, date, context and original publisher.

What is Perplexity AI?

Perplexity describes itself as an AI answer engine. A normal web search typically ranks pages and leaves the user to assemble the answer. Perplexity attempts that assembly itself: it interprets the question, searches one or more sources, extracts relevant material, generates a response and attaches citations.

The result looks conversational, but the product is built around a search process. Follow-up questions retain context, so a user can narrow a topic without constructing a new query from scratch. A thread can progress from “What changed?” to “Compare the three proposals” and then “Show only primary sources published this month.”

Perplexity is not one model. It is a system that can route tasks across models, search indexes, retrieval tools, code execution, file parsers and interfaces. Paid users can choose among supported models for some modes, while automatic routing can select one for them.

The distinction matters. When an answer is strong, credit may belong to the retrieved evidence and the way Perplexity organized it—not only to a model. When an answer fails, the problem could be an ambiguous question, weak sources, incomplete retrieval, incorrect synthesis, stale indexing or model hallucination.

Is Perplexity a search engine, chatbot or AI agent?

It is now all three, but through different products.

An answer engine

The core Perplexity interface is closest to an AI search engine. It is optimized for questions that benefit from current information, source discovery and concise synthesis.

A conversational assistant

Threads, file uploads, Projects, model choice and creation tools let Perplexity handle tasks that resemble a general chatbot: summarizing a document, explaining a topic, drafting a report or transforming research into a presentation.

An agent platform

Comet can take actions inside a browser. Computer can coordinate a longer workflow across cloud tools, files and connectors. The Agent API gives developers similar building blocks for their own applications.

These labels should not be collapsed. Asking Perplexity for a cited overview has a different risk profile from allowing an agent to send email, alter a record or make a purchase.

How Perplexity works

Perplexity does not publish every detail of its ranking and orchestration systems, but its public documentation and observable interface support a practical five-stage explanation.

1. It interprets the question

The system identifies entities, time constraints, comparisons and the likely form of the answer. A request for “the latest” should trigger a different retrieval strategy from a request to explain an established concept.

Good prompts make those constraints explicit. “Compare the July 2026 US prices shown on the vendors’ own pages” is more actionable than “Which is cheaper?”

2. It searches relevant sources

Perplexity can perform one or several searches rather than relying entirely on information learned during model training. Depending on the mode, the user can focus on the general web, academic material, financial sources or uploaded files.

3. It retrieves and ranks passages

The system selects pages and passages that appear relevant. Retrieval quality is a major bottleneck: a polished synthesis cannot recover evidence that was never found, was inaccessible or came from a poor source.

4. A model synthesizes the answer

A language model combines the retrieved material into prose, a table, a calculation or another requested format. The output can include useful inference, but that inference may go beyond the literal text in the sources.

5. Citations connect claims to pages

Inline references let the reader open the supporting material. They make verification possible, not automatic. The cited page may support only part of a sentence, reflect an old version of a policy or repeat another outlet’s reporting.

Our guide to using Perplexity turns this process into repeatable search, prompting and verification workflows.

Perplexity’s search modes explained

Mode names and placement can change, but the basic progression is from speed to depth.

Search or Ask

The ordinary experience is suitable for definitions, current facts, recommendations to investigate and straightforward comparisons. It is fast and available on the free plan.

Use it when a good answer can be built from a small number of clear sources. Do not start a long research run for a question that one official page can answer.

Pro Search

Pro Search plans and performs a more involved search. It can use multiple queries, reason across results and let eligible users choose an advanced model. Perplexity currently offers source focuses that include the web, academic material, finance and files.

This mode fits product comparisons, technical investigation, market scanning and questions where the first search is unlikely to find the complete answer.

Research

Research mode spends more time conducting a sequence of searches and produces a longer report. It is designed for a topic that needs breadth, synthesis and an evidence trail rather than an instant response.

Research is useful for an initial landscape review. It is not the same as a systematic literature review, legal opinion, financial audit or professional due-diligence process.

Advanced Deep Research

Advanced Deep Research expands the research process with deeper source exploration, cross-referencing, uploaded-document analysis and a code sandbox for quantitative work. Availability and allowances depend on the plan.

Its best use is a bounded, demanding assignment with a defined output: evaluate a market, compare a body of evidence, analyze a dataset alongside public sources or produce a decision brief.

Create

Create shifts the objective from answering a question to producing an asset. Depending on access, Perplexity can generate reports, presentations, spreadsheets, dashboards, simple applications, images or video.

The output still needs functional and factual review. A presentation can look finished while containing a weak source. A spreadsheet can calculate correctly but encode the wrong assumption.

Why model choice is not the whole product

Perplexity has offered models from multiple providers, including its own systems and frontier families from companies such as OpenAI, Anthropic and Google. The exact list changes too quickly to serve as a permanent feature table.

Model choice can affect:

writing style and concision;
reasoning on technical or mathematical tasks;
how instructions are followed;
the ability to work with long context;
latency;
and the amount of plan capacity consumed.

It does not replace source selection. Two models grounded in the same incomplete evidence may produce different prose but share the same blind spot.

For normal use, start with automatic selection. Choose a specific model only when testing shows a repeatable advantage for the task. Record the model and date when reproducibility matters.

Model Council, available on eligible high-end plans, runs three models and synthesizes their perspectives. It can expose disagreement, but agreement among models is not independent evidence if they rely on the same sources.

Citations: Perplexity’s biggest strength and its most misunderstood feature

Visible citations reduce the friction between an answer and its evidence. They are particularly valuable for:

finding the original announcement behind a news summary;
checking when a price or policy changed;
separating a company claim from independent reporting;
following a technical explanation into documentation;
and building a reading list quickly.

They do not guarantee:

that the source is authoritative;
that it supports the complete claim;
that the quoted context was preserved;
that the information is current;
that conflicting evidence was searched;
or that the model’s conclusion follows from the evidence.

A 2025 Tow Center test of eight AI search products used a narrow news-retrieval task rather than every kind of Perplexity question. Perplexity answered 37% of those test queries incorrectly, and the study found broader problems with confident responses, attribution and access across the category. The exact percentage should not be generalized into a universal accuracy score, but the lesson is durable: inspect the cited page.

For a material decision, use this sequence:

Open the citation beside the exact claim.
Prefer the original document, official database, research paper or direct announcement.
Check the publication date and the date of the underlying event.
Look for qualifications omitted from the synthesis.
Search for a credible source that could contradict the conclusion.
Record what is established, disputed and inferred.

Threads, Projects and files

Threads

A thread holds a conversation and its sources. Follow-ups can refine the same inquiry without restating every detail. Keep one thread focused; unrelated questions contaminate context and make later review harder.

Projects

Projects, previously described as Spaces in some documentation and URLs, combine instructions, saved research, files and collaborators.

They work well for:

a recurring competitor review;
a course or research topic;
an editorial evidence folder;
a client or consulting engagement;
and a team knowledge area.

Paid plan limits differ. At the time of review, Pro allowed up to 50 files in a project, Max and Enterprise Pro up to 500, and Enterprise Max up to 5,000. A limit is not a recommendation to upload everything. Curate the smallest source set that improves the answer.

Connected files inherit permissions from the source provider, but an answer can still reveal information to a user who is allowed to run the query. Organizations need connector, sharing and output policies in addition to storage permissions.

Uploaded files

Perplexity accepts common documents, PDFs, text and code, as well as supported images, audio and video. Audio and video can be transcribed; visual scenes in video are not necessarily indexed as if every frame had been inspected.

File limits vary by upload location and plan. The interface’s limit for an individual session should not be confused with a project’s file allowance.

Uploads are also a privacy decision. Ordinary session files and files stored in a project can follow different retention rules. The Perplexity privacy and safety guide maps those differences.

What can Perplexity create?

Perplexity has grown beyond text answers.

Eligible accounts can use asset creation to turn research into structured deliverables. Image and video generation are available in supported plans and regions. Learn mode provides guided explanations, while Education Pro adds student-oriented access at a discounted price after verification.

Perplexity also offers specialized experiences:

Finance-focused search for markets and companies;
Health for supported US Pro and Max users, including optional connections to medical records and wearable data;
shopping and Instant Buy for supported US transactions;
mobile voice and assistants for hands-free questions and selected actions.

These experiences have different consequences. A generated image needs an originality and rights check. Health information is educational rather than a diagnosis. A shopping agent can affect money and an external account. Treat the product label as a starting point for risk analysis, not a universal safety guarantee.

What is the Comet browser?

Comet is Perplexity’s Chromium-based browser for macOS, Windows, iOS and Android. It combines normal browsing with an assistant that can summarize pages, compare tabs, use page context and perform supported actions.

Comet’s value is contextual continuity. Instead of copying text from several tabs into a separate chatbot, the assistant can work with the browsing session. Possible uses include:

summarizing a long page;
comparing products across tabs;
extracting dates or action items;
finding information within a site;
preparing an email from research;
or guiding a multi-step web task.

The same context creates risk. A malicious page can attempt prompt injection. Broad permissions can expose browsing context or account data. An agent can click the wrong control or act on an inaccurate interpretation.

The complete Comet browser guide covers setup, permissions, privacy, security findings and a safe operating method.

What is Perplexity Computer?

Computer is a cloud-based digital worker. Perplexity summarizes the distinction as: its search product answers, while Computer does work.

Computer can plan a task, use connected applications, delegate subtasks, work with files, create an output and remember useful context. Examples include:

researching prospects and updating a sales system;
assembling a recurring market brief;
building a spreadsheet from several sources;
creating a small application;
reviewing a backlog and preparing a project update;
or coordinating material across email, documents and databases.

It operates in an isolated cloud environment and can connect to supported email, collaboration, code, project, CRM and data services, depending on availability and permissions. Scheduled tasks can run recurring work. Skills package reusable procedures. Personal Computer extends the model to local Mac files and applications.

Computer usage is measured in credits, separate from ordinary search. Work can vary dramatically in cost because an open-ended assignment may run many steps. Our Perplexity Computer guide explains credits, connectors, workflows and approval boundaries.

Perplexity plans and pricing

The individual range starts with Free and extends through Pro and Max. Education Pro discounts the Pro experience for verified students and educators. Organizations choose Enterprise Pro or Enterprise Max.

Plan	Standard US price at review	Best fit
Free	$0	Basic cited search and occasional advanced use
Education Pro	$10/month	Verified students and educators
Pro	$20/month or $200/year	Frequent research, files, models and creation
Max	$200/month or $2,000/year	Heavy individual research, Computer and early/high-capacity features
Enterprise Pro	$40/user/month or $400/user/year	Managed organizational search and knowledge access
Enterprise Max	$325/user/month or $3,250/user/year	Maximum enterprise capacity and advanced controls

These prices do not explain the entire bill. Computer can consume credits. API usage is separate. Taxes, currency, app-store billing, promotions and negotiated enterprise terms can change the total.

Free is sufficient to learn the core product. Pro is the sensible paid starting point for most individuals. Max should solve a measured capacity or feature constraint, not a vague desire for “the best.”

The dedicated Perplexity pricing guide compares current allowances, Computer credits, cancellation and plan-selection decisions.

Perplexity for developers

Perplexity sells search and research infrastructure independently of the consumer subscription.

Agent API

The Agent API is the broadest developer surface and Perplexity’s recommended starting point for many new projects. It can orchestrate supported third-party and Perplexity models with search, URL retrieval, people and finance tools, a sandbox, functions and other integrations.

Sonar API

Sonar returns generated answers grounded in web search. It suits an application that needs a readable response with citations rather than only raw results.

Search API

The Search API returns ranked web results. It is the better primitive when the developer wants to control retrieval, ranking, filtering and generation separately.

Embeddings

Embedding models convert text into vectors for semantic search, clustering and retrieval systems.

API billing does not come with Pro or Max. The Perplexity API guide includes current prices, selection logic, code examples and production controls.

Perplexity for organizations

Enterprise plans add a managed environment around the answer engine. The value is not simply more searches.

Depending on plan and deployment, organizational capabilities include:

SAML single sign-on and SCIM provisioning;
team administration;
controlled file and app connectors;
internal knowledge search;
source-permission inheritance;
audit and retention options;
larger project and file limits;
premium data sources;
enterprise data commitments;
and access to Computer with enterprise controls.

The difficult part is deployment. A connector can make useful knowledge searchable while also magnifying an old permission problem. A sourced answer can still be wrong. A capable agent can perform an unintended action more quickly than a human.

The Perplexity Enterprise guide covers pricing, identity, connectors, retention, governance and a practical rollout checklist.

Does Perplexity use your data for training?

For consumer Free, Pro and Max accounts, Perplexity says AI data retention for model improvement is enabled by default. Users can turn it off in account preferences. The setting affects model-improvement use, but it does not stop every form of processing required to operate, secure or legally maintain the service.

Perplexity says Enterprise data is not used to train its models. It also says contracts prohibit third-party model providers from training on data sent through Perplexity.

Retention is a separate question from training. Turning off model improvement does not instantly delete search history. Deleting a thread may not remove a copy stored somewhere else. Session files, project files and account information can follow different schedules.

The safe rule is simple: do not upload data merely because an opt-out exists. First decide whether the account, plan, connector and retention controls are appropriate for that data class.

Is Perplexity accurate?

Perplexity can be highly useful and still be wrong.

Its search-first design reduces one common problem of static language models: answering a current question only from training data. It also gives the reader an unusually visible path to verification. Neither feature eliminates:

hallucinated details;
weak or circular sources;
incorrect calculations;
overconfident conclusions;
outdated pages;
selection bias;
missing paywalled or inaccessible evidence;
and errors in the underlying source.

Accuracy depends on the task. Finding an official price is different from diagnosing a medical condition, identifying an article from an excerpt or forecasting a market.

Use Perplexity as an efficient research assistant. Do not delegate final judgment in high-stakes medical, legal, financial, safety or employment decisions.

Privacy, copyright and crawler disputes

Perplexity’s use of online material has generated legal and publishing disputes. News organizations have accused the company of copyright infringement or unauthorized distribution. Perplexity has pursued licensing and revenue-sharing relationships with publishers. Cloudflare reported that undeclared crawlers associated with Perplexity appeared to evade some sites’ blocking directives; Perplexity disputed the characterization.

These are not all the same issue:

Copyright concerns whether protected expression was copied or used unlawfully.
Crawler control concerns how a site’s access preference is communicated and respected.
Attribution concerns whether the original source receives accurate credit and traffic.
Answer accuracy concerns whether the synthesis represents the source correctly.
User privacy concerns the information people submit to Perplexity.

Several disputes remain unresolved, so allegations should not be presented as final legal findings. Readers who publish Perplexity-assisted work remain responsible for fact-checking, attribution and rights clearance.

Who owns Perplexity?

Perplexity AI, Inc. is a private company co-founded in 2022. Aravind Srinivas serves as chief executive alongside a founding group with machine-learning, infrastructure and startup experience.

The company has raised money from venture firms, technology companies and individual backers. A September 2025 report attached a $20 billion valuation to $200 million of investor commitments, although Reuters said it could not independently verify the report.

A valuation is not revenue, cash in the bank or the value of publicly tradable shares. The company does not publish a complete current cap table, so outsiders cannot reliably state each founder’s or investor’s ownership percentage.

Perplexity has said 2028 is its earliest planned IPO date. It is not currently listed on a stock exchange. The Perplexity ownership and business guide separates confirmed company facts from reported funding snapshots.

Perplexity versus Google and ChatGPT

Perplexity versus Google Search

Google remains a broad discovery and navigation system with maps, shopping, local results, news, images and an enormous advertising ecosystem. Perplexity begins with a synthesized answer and makes follow-up conversation central.

Choose traditional search when you want to explore many results, navigate to a known destination, inspect local listings or control the reading path. Choose Perplexity when you want a fast, sourced synthesis of a clearly framed question.

Often the best process uses both: Perplexity maps the topic, while a traditional search checks coverage and finds sources the synthesis missed.

Perplexity versus ChatGPT

ChatGPT and Perplexity now overlap in web search, research, files and agents. Their centers of gravity still differ.

Perplexity is optimized around source discovery and cited answers. ChatGPT is a broader general work environment for extended drafting, coding, data analysis, image work and iterative deliverables.

The right choice depends on the workflow rather than one benchmark. Our detailed Perplexity versus ChatGPT comparison tests the products across research, writing, files, agents, privacy and price.

What Perplexity is best for

Perplexity is particularly strong for:

learning the vocabulary and major sources of an unfamiliar topic;
checking current product, policy and company information;
comparing claims across several pages;
finding primary sources behind a summary;
building a first-pass literature or market map;
turning a bounded source set into a brief;
monitoring a topic on a schedule;
and adding search-grounded answers to an application.

It is a weaker default for:

work where no external evidence is needed;
sensitive data on an unsuitable account;
decisions that require a licensed professional;
exact reproduction of a complex spreadsheet or application without testing;
irreversible browser actions without human approval;
and any workflow where a visible citation will be mistaken for proof.

How to get started

Open the web app or install the mobile app.
Ask one specific current question.
Include the geography, time period and desired output.
Open at least two citations.
Ask a follow-up that challenges the first answer.
Save the thread only if it has lasting value.
Test Pro only when free limits interrupt real work.
Create a Project when the same topic will recur.
Connect an external service only after reviewing permissions.
Use Comet or Computer first on reversible, low-risk tasks.

A strong first prompt is:

Explain [topic] for [audience]. Use sources published or updated since [date] where recency matters. Separate confirmed facts, company claims and reasonable inference. Prefer primary sources, cite every time-sensitive claim and end with three uncertainties I should investigate.

Common Perplexity mistakes

Treating every citation as confirmation

Open the page and verify the relevant passage. A credible domain can still be cited for a claim it does not fully support.

Using Research for a one-page fact

Longer runs cost time and may introduce unnecessary sources. Match the mode to the question.

Asking for “latest” without a date or region

State what current means. A price in the United States can differ from a price in Europe; a policy announced today may not be effective today.

Uploading a complete drive because a connector exists

Use least privilege. Connect the smallest relevant folder or repository and review inherited sharing.

Buying Max before measuring Pro

Max is expensive. Upgrade because a named feature or capacity limit blocks valuable work.

Assuming a consumer subscription includes API usage

Consumer plans and developer billing are separate.

Letting a browser agent perform an irreversible step silently

Require confirmation before sending, buying, deleting, publishing, granting access or changing security settings.

Citing Perplexity instead of the source

For published work, cite the underlying primary or authoritative source unless the Perplexity output itself is the object being discussed.

Frequently asked questions

Is Perplexity AI free?

Yes. The Free plan provides core cited search and limited access to more advanced modes. Paid plans add capacity, model choice, files, creation tools and other features.

Is Perplexity a search engine?

Perplexity calls itself an answer engine. It searches and retrieves web material like a search product, but returns a synthesized conversational answer instead of only a ranked result page.

Who owns Perplexity AI?

Perplexity is privately held by its founders, employees and investors. Its complete current ownership percentages are not public.

Does Perplexity use ChatGPT?

Perplexity has offered access to model families from OpenAI and other providers, but it is not a version of ChatGPT. Perplexity operates its own search, retrieval, citation, interface and orchestration systems.

Is Perplexity better than Google?

It is better for some source-led synthesis and follow-up research. Google is stronger as a broad navigation, local, shopping, image and discovery system. Important research benefits from checking both.

Is Perplexity better than ChatGPT?

Perplexity is often the cleaner starting point for current cited research. ChatGPT can be the better general environment for extended creation, coding and iterative work. Both products now cover more than those simple labels.

Can Perplexity read PDFs?

Yes, subject to supported format, size and plan limits. Scanned or visually complex PDFs may require checking because extraction can miss layout, charts or image-only text.

Does Perplexity have an AI browser?

Yes. Comet is its Chromium-based browser with an integrated assistant and agentic features.

What is the difference between Comet and Computer?

Comet works in the browser context. Computer is a broader cloud worker that can coordinate multiple tools, connectors, files and subtasks. Personal Computer can also work with supported local Mac resources.

Are Perplexity citations always reliable?

No. They make an answer inspectable but can be incomplete, misapplied, outdated or secondary. Open the cited source before relying on a material claim.

Does Perplexity train on searches?

Perplexity documents a default-on consumer control governing whether activity can improve its AI; an individual can switch it off for future collection. Its stated Enterprise commitment excludes customer data from training.

Is Perplexity safe for confidential work?

That depends on the plan, data, retention settings, connectors and organizational policy. A managed Enterprise deployment is materially different from placing confidential data in a personal account.

Does Perplexity Pro include API credits?

No. Developer API usage is billed separately.

Can I buy Perplexity stock?

Not through a public stock exchange. Perplexity is private and has said 2028 is its earliest planned IPO date.

The bottom line

Perplexity’s original proposition remains compelling: ask a question, search current information and show the reader where the answer came from. Its 2026 product is now a much larger system around that idea.

Search and Research help people understand. Projects and creation tools organize and transform knowledge. Comet brings assistance into the browser. Computer attempts complete work. Enterprise and APIs extend the same retrieval foundation into organizations and software.

The right way to evaluate Perplexity is therefore not “Is it a better chatbot?” Ask a narrower question: does its combination of retrieval, citations and action improve this particular workflow enough to justify the cost and risk?

Start with a real question on the free plan. Inspect the sources. Measure what the paid mode changes. Add permissions and agents gradually. Perplexity is most valuable when its evidence trail strengthens judgment—not when its confident answer replaces it.

byDavid Porter

AI Agents AI tools Perplexity

Perplexity Enterprise: Pricing, Security, Connectors and Deployment Guide

Perplexity Computer: Complete Guide to the AI Agent, Credits and Workflows

Write a comment