Perplexity review

What Perplexity is for

Typing a question into Perplexity returns a written answer built from live web results, with numbered citations linking to every source it used. That’s the pitch in one line: skip the ten open tabs and the sifting, and get a synthesized answer you can verify in two clicks instead of twenty. It sits somewhere between a search engine and a chatbot, but it’s built around sourcing rather than a frozen training cutoff, so it can answer a question about this morning’s news or an hour-old stock price and show its work.

Under the hood, Perplexity runs its own Sonar search models plus a menu of outside models from OpenAI, Anthropic and Google, and lets you pick which one answers a given query. Past one-off search, Spaces (formerly Collections) hold a project’s files and instructions in one place, and Labs turns a research question into an actual deliverable: a report, a dashboard, a spreadsheet or a small working app, not just another paragraph of text.

Where it shines

Sourcing is the differentiator, and it holds up under outside scrutiny. Columbia’s Tow Center for Digital Journalism tested eight AI search tools on 200 identical queries each in a widely cited 2025 study, and Perplexity came out with the lowest failure rate of the group at 37%, well ahead of ChatGPT Search at 67% and Grok-3 at 94%. That’s not a clean bill of health (more on that below), but among AI answer engines, it’s the one to beat.

The free tier holds up on its own: unlimited basic search plus a handful of Pro searches a day, no card required. Pay $20/mo for Pro and you get Spaces for organizing a project’s files, plus 50 monthly Labs runs for turning research into a report or dashboard instead of a wall of text. And as of late 2025, Comet, Perplexity’s AI browser, dropped its $200/mo paywall entirely. It’s free now on macOS, Windows, Android and iOS, with tab summarization and agentic actions built into ordinary browsing rather than gated behind the top plan.

Reviewers back this up. Perplexity holds a 4.4 out of 5 across 326 reviews on G2, with ease of setup and ease of use both landing in the mid-90s percentile and reviewers citing speed and source transparency.

Where it frays

Best of the AI search engines still means wrong more than a third of the time by that same Tow Center measure, and the specific failure mode is the tricky part. Perplexity tends to cite a real, correctly formatted source and attribute the wrong claim to it, rather than inventing a fake link outright. A citation that looks legitimate is harder to catch than one that obviously isn’t, so anything with real stakes, a client deliverable, a legal argument, a number going into a deck, needs a manual click-through before you repeat it.

Perplexity’s relationship with the publishers it depends on is adversarial right now. The New York Times sued it for copyright infringement in December 2025, joining earlier suits from News Corp (over The Wall Street Journal and New York Post), Encyclopaedia Britannica and Merriam-Webster, and Japan’s Nikkei and Asahi Shimbun. Amazon sued separately in late 2025 over Comet’s shopping agent acting on Amazon’s site and over Comet allegedly summarizing paywalled Tribune content. Cloudflare added its own accusation in August 2025, saying Perplexity’s crawlers disguised themselves as ordinary Chrome browsers and rotated IPs to dodge robots.txt blocks. Cloudflare delisted it as a verified bot in response; Perplexity’s public defense was that a bot acting on a user’s behalf deserves the same access a person browsing manually would get.

Privacy defaults lean toward Perplexity, not the user. AI training on your queries is on by default for Free, Pro and Max accounts, and you have to find the toggle in account settings yourself to turn it off. If you’re not logged in, there’s no opt-out at all. Enterprise plans are the exception, with a contractual guarantee that customer data never trains the models.

Pricing gets confusing once Pro isn’t enough. Max is $200/mo, a 10x jump from Pro, for unlimited Pro search and Labs plus priority model access. Enterprise starts at $40 per seat a month for Enterprise Pro and climbs to $325 per seat a month for Enterprise Max. There’s real distance between those tiers, and no obvious middle step for a team that’s outgrown Pro but doesn’t need the full enterprise stack.

Who should buy it

Buy it if your work runs on fast, sourced answers, market research, competitive scans, first-pass due diligence, and you’re disciplined about clicking through anything before it goes into something you’ll publish or bill for. Free or Pro covers most individuals; Labs and Spaces are the reason to pay for Pro specifically if you build recurring reports rather than one-off answers. Max only earns its $200/mo if you outgrow Pro’s search or Labs caps, or need longer autonomous runs through Perplexity Computer.

Skip it, or at least verify everything twice, if your output has zero tolerance for an uncaught bad citation: legal filings, published journalism, medical guidance. And if what you actually need is an agent that executes multi-step tasks across your own apps rather than an assistant that researches and answers, Lindy is a closer fit than a search-first tool like this one.