What is Perplexity AI?

Why — the first-principles explanation

A language model on its own generates a plausible continuation from learned patterns. That is useful for drafting, but it may not know what changed yesterday and can state an invented fact fluently. An answer engine changes the order: it searches first, retrieves source material, gives that material to a model and returns a synthesized answer with links back to the sources. Perplexity describes this as an answer engine that searches the internet in real time.

Citations are the product's checking interface, not a truth certificate. A cited page can be outdated or wrong; the model can misread it, use a snippet without enough context or make a sentence stronger than the source. Open the cited page, check the publication date and scope, and prefer primary documents for important claims. A citation shortens the path to verification; it does not outsource judgment.

There are several product layers. In the consumer app, a quick search is different from a reasoning task or a broader research report. Perplexity's developer documentation groups model use cases into Search, Reasoning and Research, with different trade-offs between speed, depth and cost. Use the smallest mode that can answer the question, then escalate when the evidence or analysis is insufficient.

The individual pricing page currently displays Free at $0/month, Pro at $20/month and Max at $200/month for the listed US plans. The paid tiers add some combination of advanced model access, deeper research, premium sources, uploads, connectors, Computer or other credit-based features. Exact quotas, regional prices, taxes and included credits can change, so the live plan page is the source of truth rather than a cached comparison article.

A consumer subscription is not the same thing as API access. Perplexity publishes separate Search, Gateway, Agent and Sonar API billing. Search is metered by successful request; Gateway and Agent usage is token or tool metered; Sonar has its own token and search-context pricing. Budget the API from the workload and read the endpoint-specific rate table before shipping a feature.

Privacy also depends on the surface. The consumer privacy notice covers websites, apps and related services and says Perplexity may use collected data to improve or create services, including AI models, subject to its stated choices and legal terms. The API privacy page separately states zero retention and no training for the Chat Completions API, while the enterprise page says enterprise customer data is not used to train its LLMs. Those are different contractual contexts: do not infer that a consumer setting, every API endpoint or every connected app has the same treatment.

For a team, the question is governance as much as answer quality. Perplexity Enterprise advertises controls such as SSO and SCIM, user management, configurable file retention, SOC 2 Type II and connectors to internal knowledge. Confirm the security and retention commitments in the order form and trust documentation that apply to your region and plan before uploading confidential material.

A defensible workflow is simple: state the question, date range and acceptable sources; run the least expensive suitable mode; open several citations; triangulate important claims against primary sources; save the evidence and prompt; and have a qualified person approve high-stakes decisions. This is where Perplexity is strongest—as a fast research interface with a visible trail—not as an automatic authority.

An example that makes it click

How to do it

1. Define the job: a quick current fact, a multi-source report, a file analysis, a team workspace or an API feature. The job determines the right product surface.
2. Start with a question whose answer you can evaluate. Use a precise scope, date range and source preference instead of asking for an unbounded summary.
3. Click several citations and check that each one actually supports the nearby sentence. Read the surrounding context, not only the highlighted snippet.
4. Use Search for straightforward retrieval, Reasoning for multi-step analysis and Research for broad reports when the official model documentation says that trade-off fits your task.
5. For important claims, triangulate against primary documents and record the source URL, publication date and any disagreement. Treat uncited or weakly sourced statements as leads.
6. Try Free first, then identify the bottleneck that would justify Pro or Max: model choice, research volume, premium sources, uploads, connectors, Computer features or credits.
7. Before paying, check the live plan page for price, limits, credit resets, cancellation, regional availability and data choices. Do not treat a subscription as API access.
8. If building software, price Search, Gateway, Agent and Sonar separately, set rate and spend limits, and test latency and citation behavior on representative queries.
9. For private work, classify the data first. Review consumer privacy choices, endpoint-specific API retention terms or enterprise controls before uploading sensitive files or connecting services.
10. For medical, legal, financial, employment or other high-stakes decisions, use the cited sources and qualified review; an AI answer is not the final authority.

Key facts

• Perplexity's official introduction describes an answer engine that searches the internet in real time and returns answers with clickable citations.
• The displayed US individual pricing page currently shows Free at $0/month, Pro at $20/month and Max at $200/month; regional, tax, annual and promotional terms can differ.
• The consumer plan page describes different access and credits across Free, Pro and Max; included features and quotas can change.
• Perplexity's model documentation separates Search, Reasoning and Research use cases so users can trade off retrieval speed, multi-step analysis and report depth.
• Developer billing is separate from the app subscription: Search API uses request pricing, while Gateway, Agent and Sonar have their own token, tool or search-context charges.
• The consumer privacy notice and the API/enterprise policies are separate documents. The API page states zero retention and no training for Chat Completions API requests; enterprise materials state enterprise customer data is not used to train LLMs.
• Enterprise materials advertise SSO, SCIM, user management, configurable file retention, connectors and SOC 2 Type II; verify the terms that apply to the purchased plan.
• A citation makes a response easier to audit, not automatically correct. Verify the source's date, context, authority and support for the exact claim.

What authoritative sources say

People also ask

Related questions