Perplexity

What it actually is

Perplexity is built around one core idea: instead of answering purely from memory or returning a list of links, it runs a live retrieval pass against the web for nearly every query and attaches a numbered citation to each individual claim in its answer, so you can click through and check the source yourself. That’s not a bolted-on feature, it’s the entire product — the company’s own claimed hallucination rate (3–5%) and citation accuracy (94%) are both framed around exactly this promise, which is part of why the gap between those claims and independent audits matters as much as it does.

You’re also not locked to one model. Perplexity’s own Sonar models handle most queries by default — Sonar for fast answers, Sonar Pro for more complex synthesis, Sonar Reasoning Pro for multi-step problems — but Pro and above let you route a question to GPT, Claude, or Gemini instead, or let Research mode pick automatically. Six tiers sit on top of that core product: Free, Pro ($20/month), Max ($200/month), Education Pro ($10/month for verified students), and two Enterprise tiers. The underlying answer engine is the same across all of them; what changes is how many Pro Searches and Research reports you get per day, and whether you can reach the heavier tools — Model Council, Labs, the full Comet browser — at all.

Where it’s genuinely good

The citation-first design is a genuinely different way to use an AI tool, when it works. Ask a question and you get an answer with sources numbered inline, and focus modes let you narrow those sources before you even ask — restrict to academic papers, Reddit threads, or YouTube transcripts depending on what kind of answer you actually want. Spaces extend that into something persistent: a project workspace with its own custom instructions and uploaded files that Perplexity treats as context for everything you ask inside it, rather than starting from zero every conversation.

Model Council, exclusive to Max, is the most genuinely novel feature here — running the same query through three frontier models at once and surfacing where they agree and disagree is a fundamentally different kind of answer than any single-model chatbot gives you, and it’s well suited to exactly the high-stakes research and decision-making Perplexity markets itself for. Labs, for anyone willing to wait ten minutes instead of three, goes further still, producing actual spreadsheets, dashboards, and small working apps instead of just a written report.

Where it falls short

The accuracy gap is the real issue, and it’s worth taking seriously precisely because sourcing is the whole pitch. Perplexity cites real URLs, but a widely cited external audit found the claim attached to that URL didn’t match what the source actually said in roughly a third of cases tested — and, notably, the higher-tier model variant in that same study didn’t score better than the standard one. A citation that looks verifiable but isn’t is a harder failure mode to catch than an answer with no citation at all, because the presence of a link creates a false sense of having already checked.

The publisher lawsuits are the other thing worth knowing about before you rely on this for anything you’ll publish or cite onward. Multiple major outlets have alleged in court that Perplexity’s crawlers ignored robots.txt and other explicit blocks; separate research from Cloudflare in 2025 raised similar concerns about undeclared crawler behavior. Perplexity disputes parts of this and has struck revenue-sharing deals with some publishers as an alternative path, but the litigation is real and ongoing, not resolved.

Comet is the sharpest edge. Since mid-2025, independent security researchers — Brave, LayerX, and others — have repeatedly demonstrated that a malicious webpage can embed hidden instructions Comet will follow as if they came from the user, potentially exposing email, saved credentials, or even local files. Perplexity has patched specific reported issues, but the pattern of repeated, serious disclosures from multiple independent groups over nearly a year is the kind of track record that should make you think carefully about what you connect it to.

Which plan to actually get

Free is genuinely enough if you’re using this occasionally for quick, sourced answers — it’s not a crippled trial. Move to Pro once you’re doing real research regularly; unlimited Pro Search and full model selection are what most people are actually paying for, and Education Pro at $10/month is the better deal if you’re a verified student and don’t need more than that. Max only makes sense if Model Council or Labs are things you’ll use often enough to notice — for occasional heavy research days, staying on Pro is the better value. Enterprise is about governance and admin controls, not a materially better core product, so it’s worth it specifically for the SSO and data-isolation requirements, not for extra capability.

Ease of use

The basic search experience is as easy as typing into Google and genuinely easier to act on, since the sources are already attached. Where the learning curve actually shows up is in choosing the right mode for the task — Quick Search versus Pro Search versus Research versus Labs each trade speed for depth differently, and picking wrong means either waiting too long for a simple question or getting a shallow answer to a complex one. Comet adds a separate layer entirely: it’s a full browser asking for real permissions, and given its security history, that’s not a decision to make on autopilot.

Who it’s actually for

Researchers, analysts, and students who want sourced, citable answers and intend to actually verify the sources rather than take the synthesis on faith — that verification step is what the product is genuinely built for, not an optional extra. It’s also a strong pick for finance and market research workflows where Spaces and Deep Research can meaningfully cut prep time. It’s a weaker choice if you want to trust an AI’s synthesized answer without checking sources yourself, or if you need a browser-based agent and aren’t ready to weigh Comet’s security disclosures against the convenience it offers.