The short answer
Perplexity is an AI answer engine from Perplexity AI, Inc. that searches the web and answers with numbered citations, which makes it the best everyday tool for research you need to check. Free users get unlimited basic searches. Paid plans add model choice, including GPT-6.1 Sol and Claude Opus 5.5, Deep Research and Perplexity Computer, an agent launched in February 2026 that turns multi-step tasks into finished reports and files.
| Made by | Perplexity AI, Inc., San Francisco, United States |
|---|---|
| Free plan | Yes: unlimited basic searches with citations, small allowance of Pro Searches and Research |
| Models | Models from other makers on paid plans, including GPT-6.1 Sol, GPT-6 Astra, Claude Opus 5.5, Gemini 4 Argon and Kimi K3 |
| Best at | Quick research with sources, comparisons, fact checks, sourced reports in Computer |
| Shows sources | Yes, on every answer |
| Team plans | Enterprise Pro and Enterprise Max |
| Your data | Free, Pro and Max train by default until you switch off AI data retention. Enterprise data is never used for training. |
What Perplexity is in November 2026
Perplexity, made by Perplexity AI, Inc. in San Francisco, started as an answer engine: you ask a question, it searches the web, and it writes a short answer with numbered citations you can click. That is still the core, and it is still the best reason to use it. Where ChatGPT or Claude answer from what they learned plus an optional search, Perplexity searches first by design, so the habit of showing sources is built into every reply.
In 2026 it grew a second half. Perplexity Computer launched on 27 February 2026, first for Max subscribers, and it is now on every paid plan. Perplexity's own help centre puts the difference simply: the search side answers your questions, Computer does your work. Computer runs in a cloud sandbox, chains multi-step workflows, writes documents, presentations and small apps, and connects to tools such as Gmail, Outlook, Slack, Notion, GitHub, Linear and Salesforce. It is paid for with credits, where 100 credits equal one US dollar, and Perplexity's own examples put a multi-step research report at roughly 350 to 950 credits.
The other big difference is model choice. Perplexity does not build one flagship model; it rents the best ones from other makers. On paid plans in autumn 2026 that includes GPT-6.1 Sol for search and light Computer tasks (added 5 October), Claude Opus 5.5 for standard Computer work, Claude Fable 5.1 for demanding tasks, GPT-6 Astra (21 September), Google's Gemini 4 Argon (19 October) and the open-weight Kimi K3, which Perplexity hosts only on US servers. Free users get automatic model selection instead of a choice.
Where Perplexity is genuinely strong
Quick research you can check
Ask "What changed in the EU AI Act obligations for general-purpose models this year?" and you get a few paragraphs with a citation after almost every sentence. The value is not that the answer is always right. It is that you can see in ten seconds which claim rests on which page, and open the two that matter. For a briefing before a meeting, that beats both a search engine with ten blue links and a chatbot that gives you no trail at all.
Comparing options with sources side by side
Perplexity is good at "compare X, Y and Z on these four criteria and cite each cell". Vendor shortlists, competitor pricing pages, policy positions of three organisations: anything where the facts live on public pages and you need them in one table. Ask for a table and tell it to write "not found" when a source does not say.
Several models, one question
Model Council runs one question through several models and shows where they agree, where they disagree and what only one of them raised. In Search it needs Max or Enterprise Max and uses three models; inside Computer, every paid plan can pick two to eight models and pays in credits. For a decision where one model's blind spot could cost you, a second and third opinion in one screen is a real advantage.
Agent work that ends in a file
Computer is where Perplexity now competes with ChatGPT agents and Claude's Cowork. Give it a brief such as "build a 10-slide market overview of heat pump installers in Germany with sources on each slide" and it plans, searches in parallel, writes and hands back a file. Effort Mode, added on 21 September, lets you pick Light, Standard, High or Ultra, and Perplexity chooses a fitting model for each level. Automations, added on 5 October, run recurring work on a schedule or when something happens in a connected app, with review checkpoints before anything consequential.
Reaching into your work files
Since September, Pro, Max and Enterprise users can type @Microsoft 365 in a question to search SharePoint and OneDrive and read Excel workbooks through a work or school account. Projects, introduced on 4 August as shared workspaces that replaced Spaces, keep files, memory and connectors together so a team can pick up the same research thread.
Where Perplexity falls short
- A citation is not a guarantee. Perplexity can cite a real page and still summarise it wrongly, or lean on a weak source. Open the sources behind any number, quote or date you plan to repeat.
- Writing in your voice. It is built to answer, not to draft. For a long article, a careful email or a document that has to sound like you, ChatGPT or Claude usually give better first drafts.
- Credits are hard to predict. Computer tasks draw from a credit balance. Perplexity's own examples range from a few dozen credits for a quick lookup to several thousand for a data dashboard, and a task pauses if credits run out.
- The plan rules are confusing. Perplexity's pricing page and help centre describe limits and features differently in places, for example whether Model Council in Search is on Pro. Check inside the app before you rely on a feature.
- Training is on by default. More on that below.
How to get better answers from Perplexity
- Ask a full question, not keywords. "Best CRM" gets a listicle. "Which CRM suits a 12-person B2B agency that lives in Gmail, and why?" gets an answer you can use.
- Say what counts as a good source. Add "use official pages, regulators and reputable press, not blogs or listicles". Perplexity follows this well, and it cuts out most of the noise.
- Ask for dates. "Give the publication date of each source" lets you spot stale pages at a glance, which matters for prices, laws and product features.
- Force honesty about gaps. "If you cannot confirm something, say so instead of guessing" reduces confident filler.
- Use Computer only when the output is a file or a multi-step job. A normal search does not use credits. Save Computer for reports, decks and recurring tasks, and start on Light or Standard effort.
- Check two sources before you quote anything. Click through, read the paragraph, and only then paste the claim into your work.
Our free Perplexity lessons at /learn?tool=perplexity turn each of these into a short exercise.
Perplexity privacy settings and where to find them
On the Free, Pro and Max plans, AI data retention is switched on by default, which means your questions can be used to train Perplexity's models. To turn it off, open Account settings, go to Preferences and switch off AI data retention. The opt-out only covers data from that moment on; Perplexity says data already collected cannot be removed. It applies to your account alone, so each person on a shared team of personal plans has to do it themselves.
Enterprise Pro and Enterprise Max are different: Perplexity says enterprise data is never used for training, uploaded files are kept for seven days, and larger organisations can set their own retention period. Perplexity also says its contracts with outside model makers such as OpenAI and Anthropic bar them from training on Perplexity data. If you research anything confidential, such as an unannounced deal or customer names, use an enterprise seat or keep the identifying details out of your questions.
Which Perplexity plan suits whom
The free plan is enough if you mainly want quick cited answers a few times a day. You get unlimited basic searches, but only a small allowance of Pro Searches and Research, and no choice of model.
Pro is the plan for anyone who researches every week. You choose the model, get Deep Research, a one-time bonus of 4,000 Computer credits and, as it rolls out, a weekly allowance of Computer usage, plus premium data sources such as Statista and PitchBook. Verified students and teachers can get Education Pro, which adds Learn Mode.
Max is for people who live in Computer: it comes with 10,000 credits a month, Model Council in Search and priority access to new features. Most professionals do not need it.
Teams should look at Enterprise Pro or Enterprise Max rather than a stack of personal plans, mainly because of the training exclusion and admin controls; audit logs and SCIM come with Enterprise Max. Current prices are on Perplexity's pricing page.
Perplexity compared with the other big assistants
Pick Perplexity if your work starts with "find out": market scans, fact checks, vendor comparisons, briefings. Pick ChatGPT or Claude if it starts with "write" or "analyse this file". Many people use both, with Perplexity for the facts and another assistant for the draft. See ChatGPT vs Perplexity, Claude vs Perplexity and Gemini vs Perplexity, or our pick of the best AI for research. If Perplexity is not for you, our Perplexity alternatives page lists the options, and tool stacks show how people combine them.
Perplexity also makes Comet, a Chromium-based browser with Perplexity built in, free to download since October 2025. You do not need it to use Perplexity, but it lets the assistant read the page you are on.
Prompts to copy
Brief me before a meeting
I have a meeting with [company] tomorrow about [topic]. Using official pages and reputable press from the last 12 months, give me: what they do, their three most recent announcements with dates, and two questions I should ask. Cite every point and say if you could not confirm something.
You get: A one-screen, dated briefing with sources to click
Compare vendors in a cited table
Compare [vendor A], [vendor B] and [vendor C] for a [team size] team that needs [requirement]. Make a table with rows for pricing model, free plan, main integrations, data location and contract terms. Cite each cell from the vendor's own site and write 'not found' where it does not say.
You get: A comparison table where every cell has a source
Fact-check a claim
Someone told me: '[paste claim]'. Is it true? Find the original source of the number or quote, say who published it and when, and tell me whether it is accurate, outdated or misquoted.
You get: A verdict traced back to the original source
Track what changed
What has changed about [law, product or policy] since [month year]? List each change with its date and an official source, newest first. Ignore commentary and blog posts.
You get: A dated changelog built from primary sources
Find the evidence on both sides
What does the evidence say about [question]? Give me the strongest findings for and against, each with the study or report, its year and who funded it. End with what is still uncertain.
You get: A balanced evidence summary that shows its gaps
Get several model opinions
Use Model Council. Question: should a [type of business] with [constraint] choose [option A] or [option B] in [year]? Show where the models agree, where they disagree and what each one raised that the others missed.
You get: A decision view with the disagreements made visible (Search needs Max; in Computer it uses credits)
Hand Computer a research report
Build a [number]-page report on [market or topic] for [audience]. Include market size estimates with sources, the five main players, recent news with dates and a one-paragraph recommendation. Use official and reputable sources only, and list every source at the end. Start on Standard effort.
You get: A finished, sourced report file from Perplexity Computer
Set up a weekly watch
Every Monday at 8 a.m., search for news about [competitor or topic] from the past seven days. Send me five bullet points with a source link each, and flag anything about pricing or new products first.
You get: A recurring Monday digest via Automations