Perplexity

What it is

Perplexity is an "AI answer engine" — a hybrid product that combines a frontier LLM with real-time web search and inline citations. Ask Perplexity a question and it returns a synthesized answer with numbered links back to the sources it used. The positioning is deliberately different from ChatGPT and Claude: where those tools are designed around conversation, Perplexity is designed around answers.

Under the hood, Perplexity runs a mix of its own Sonar models and third-party frontier models (GPT-5.2, Claude Opus 4.6, Gemini 3.1 Pro, Grok). On paid tiers, users pick which model answers each query, which means one subscription effectively gets you access to every major frontier model behind a search interface. That routing is a real advantage — if Sonar isn't strong enough for a query, you can re-run it on Claude Opus without leaving the product.

The company also ships Comet, an AI-native browser that integrates Perplexity search directly into browsing workflows. Comet is currently the flagship feature for Max and Enterprise Max tiers, and it's probably the clearest signal that Perplexity is positioning itself against Google Search at the interface layer, not just at the results layer.

On the developer side, Sonar API exposes Perplexity's grounded-search capability programmatically. You send a query, get back an answer with citations and raw search results. It's priced per-query plus per-token, and it's the cleanest way we know to add grounded live-web context to an AI application that doesn't already have one.

Positioning-wise, Perplexity competes with Claude with web tools, ChatGPT with browsing, and — increasingly — Google Search itself. Among those, Perplexity wins on UX polish for research queries specifically, ships more citation-grounded by default, and gives you a pure-LLM option for followups that Google can't.

What we tested

We've used Perplexity daily across Pro and Enterprise Pro for over eighteen months, and integrated the Sonar API into two client projects that needed grounded web search in production. On the consumer side, we've tested Deep Research mode on research-heavy topics, compared Pro search quality across the available underlying models, tried Comet browser on macOS, and run Spaces (shared threads with persistent context) for team research projects.

On the API side, we've benchmarked Sonar against alternatives — Claude + Anthropic web search tool, GPT-5 with browsing, Google Programmable Search Engine — on matched research tasks. We've measured query latency, citation accuracy, and cost at real volume for applications that need live-web context as part of their core value proposition.

On the team side, we've walked a few clients through Enterprise Pro evaluation: team workspaces, SSO, zero-retention settings, the compliance posture, and the admin controls. For shops where research and grounded search are a daily team activity (legal, finance, consulting), Enterprise Pro is a surprisingly clean fit.

None of what follows is a formal benchmark. What we can offer is the texture of running Perplexity in real workflows for sustained periods and living with the results.

What's good

The single biggest reason to use Perplexity is citation quality. Every non-trivial answer comes with numbered links to the sources the model used, which changes how you can use the output. For research, you can verify claims in seconds; for writing, you can follow up on the sources that look most interesting; for any fact-dependent workflow, you have a trail back to primary sources that ChatGPT and Claude don't ship by default.

The model picker on Pro and above is quietly one of the best deals in AI. Instead of picking between Claude, ChatGPT, and Gemini as separate subscriptions, you get all three behind one search interface for $20/mo. Ask the same question of Sonar, Claude Opus, and GPT-5 and compare the outputs. For anyone who isn't sure which frontier model they prefer, Perplexity Pro is the cheapest way to run the comparison.

Deep Research is the third killer feature. Toggle it on, ask a complex question, and Perplexity runs a multi-source agentic research session — browsing dozens of pages, extracting, cross-checking, and producing a multi-page report with citations. We've seen Deep Research outputs that would have taken a human analyst half a day to compile. It's not perfect, but it's genuinely useful.

The Sonar API fills a legitimate product gap. Claude's web search tool and ChatGPT's browsing are fine for consumer use, but as production APIs they're expensive and lack Perplexity's citation formatting. Sonar gives you grounded search as a structured API output — clean JSON with answer, citations, raw results. We've built production apps on top that would have been much harder without it.

Where Perplexity earns its keep

• Citations on every answer — traceable back to primary sources.
• Model picker means one subscription covers Claude, GPT, Gemini, Grok, and Sonar.
• Deep Research produces multi-source reports that replace hours of manual work.
• Sonar API is the cleanest production-grade grounded-search API we've tested.
• Comet browser is a legitimate bet on AI-native browsing as the next interface.
• Spaces (shared threads) work well for team research projects.
• Education Pro at $30/seat/mo is aggressive discounting for schools.

For research and grounded-search workflows, Perplexity isn't a competitor to Claude or ChatGPT — it's a different product category. Treat it as such and it becomes very hard to replace.

Spaces are an underrated feature for team research. A Space is a shared thread with persistent context, scoped to a project or topic. For legal teams working on a case, finance teams tracking a deal, or consulting teams researching a client, Spaces provide the collaboration layer that ChatGPT's chat history doesn't.

Common pitfalls

A few failure modes recur in Perplexity projects — worth naming.

Trusting citations without reading them. Perplexity does a solid job of pointing to sources, but the quality of those sources varies. SEO-optimized content sometimes ranks above more authoritative pages, and Perplexity doesn't always notice. For any fact that matters, click through to the citation — don't assume the source is trustworthy just because it's linked.

Using Perplexity as a general-purpose LLM. It isn't one. Asking Perplexity to write a 500-word marketing email, refactor a function, or brainstorm product names works, but it's not what the tool is designed for — and the output shows. Use Claude or ChatGPT for those; use Perplexity when the query has a factual answer that needs grounding.

Ignoring the model picker. On Pro and above, you can switch the underlying model per query. Most users never touch this. For hard questions, asking the same query across Sonar, Claude Opus, and GPT-5 and comparing the outputs catches failures that any single model alone would miss. The picker is the feature that justifies the subscription.

Building production apps on consumer Perplexity behavior. The consumer product evolves faster than the Sonar API does. Features move, rename, get absorbed. For anything production-critical, build on Sonar — don't assume consumer UI or behavior will stay stable.

Over-buying Max. $200/mo is a lot. Unless you're actively using Comet daily, generating Sora video weekly, or running unlimited Labs workflows, Pro at $20/mo is almost certainly enough. Max is priced for a narrow audience that's already extracting real daily value from the extra features.

Not checking zero-retention on Enterprise. Pro and Max retain data by default (with opt-out available). Enterprise Pro and Enterprise Max ship zero-retention as default. For compliance-sensitive work, confirm your tier's default before uploading sensitive documents.

What's not

Perplexity is not a general-purpose LLM. It's a grounded-search product, and it's great at that — but asking it to write long-form marketing copy, refactor code, or do creative writing gets you output that's worse than what Claude or ChatGPT would produce. If your workflow isn't primarily fact-retrieval or research, Perplexity isn't the right primary.

Citation quality is the persistent caveat. Perplexity is only as good as the pages it cites, and the web is full of SEO-optimized content that looks authoritative but isn't. For critical facts, you still have to click through and verify the source. This isn't Perplexity's fault — it's the web's fault — but it's a real limitation for anyone expecting the tool to replace careful human judgment.

The Max tier at $200/mo is priced competitively with ChatGPT Pro but the value proposition is narrower. Most users who would be candidates for Max are better served by Pro at $20 — the marginal value of Comet, Sora 2 Pro, and unlimited Labs is real but narrow. Unless you have a specific daily use case for those features, stay on Pro.

The product's direction has been less stable than competitors'. Features ship, rename, move around, or get absorbed into other features. This is normal for a company this size, but it means you shouldn't bet production workflows on a specific consumer feature holding its current form.

Perplexity has no meaningful coding or agentic-coding story. If your day-to-day is code, use Claude Code or Cursor. Perplexity can answer "how does this API work" questions reliably, but it's not the tool you want to pair-program with.

The Free tier's Pro search cap (around 5/day) is stingy, and it limits the value you can extract before upgrading. This is deliberate — Pro is the funnel — but it does make the Free tier less useful for evaluation than competitors' free tiers.

Who should use it

Researchers, journalists, analysts, and knowledge workers are the obvious fit. For anyone whose day-to-day involves synthesizing information from multiple sources with source-verification as a baseline, Perplexity Pro at $20/mo is a genuine productivity multiplier. We know journalists who have moved their entire research workflow into Perplexity because Deep Research and the citation format are that useful.

Developers building apps that need grounded web search should look at Sonar API. Building the same capability on top of Claude or GPT with a separate search API requires wiring up Programmable Search Engine or Bing Search, parsing results, feeding them back into the model, formatting citations. Sonar does all of this in one API call. The math works: Sonar is meaningfully cheaper than DIY-grounded-search at any real volume.

Teams in research-heavy industries — legal, finance, consulting, due diligence — should evaluate Enterprise Pro. The combination of grounded search, Spaces for team context, SSO, and zero-retention addresses the specific concerns those industries have about AI-powered research. At $40/user/mo, it's cheaper than most enterprise AI subscriptions.

Students and educators should look at Education Pro at $30/seat/mo — meaningful discount from Enterprise Pro. NotebookLM (Gemini) is the adjacent product to consider for pure research; Perplexity wins on breadth of sources, Gemini wins on document grounding.

Who should not use Perplexity as their primary AI: developer teams shipping code daily (Claude wins), consumer chat users who want multimodal features (ChatGPT wins), anyone whose work is mostly creative writing rather than fact-retrieval. For those, Perplexity is a useful secondary — a "research mode" tool — but not the default.

Verdict

Perplexity has carved out a real product category — grounded AI search — and dominates it. For research-heavy workflows, it's not a competitor to Claude or ChatGPT; it's a different kind of tool that fills a different need. The Pro tier at $20/mo is one of the highest-value AI subscriptions in market, and the Sonar API is the cleanest production grounded-search API available.

We rate it 8.5 / 10. Points lost on the general-purpose-LLM use case (it's not trying to be one), on citation quality variability, and on the narrow audience for the Max and Enterprise Max tiers. Points gained on Deep Research, the model picker, Sonar API polish, and the overall product clarity — Perplexity knows what it is in a way most AI companies don't.

If you're on the fence, use Perplexity Pro for a week as your default search engine instead of Google. By the end of the week, you'll either have replaced a chunk of your research workflow or you'll know the tool isn't for you.