Perplexity review: Finally, an AI search tool that shows its work

The first time Perplexity got it wrong on me, I clicked the citation. It was something obscure, the population of a Brazilian municipality circa 2018. Perplexity quoted a figure. I clicked the source link. The cited Wikipedia page actually said something slightly different. The number was off by a few hundred, in a way that wouldn't matter for most things and would matter a great deal for others.

That moment crystallized something. Perplexity isn't a tool that gives you correct answers. It's a tool that gives you answers you can actually verify, and that's a different and more useful thing than ChatGPT or Claude offer most of the time.

After two years of using Perplexity as my default search-replacement-slash-research-tool, I have opinions. Here they are.

What Perplexity does that's different

The product is a search interface that, when you type a question, returns a synthesized answer plus citations to the sources it drew from. The synthesis is good. The citations are the whole point.

You can ask "what's the current FDA stance on semaglutide for adolescents" and get a paragraph that summarizes the latest guidance, with numbered footnotes pointing to FDA.gov, recent JAMA articles, and major news coverage. The same question on ChatGPT gets you a confident answer with no sources, or a request to enable web search that may or may not surface comparable material. Claude does better here with its web search feature, but Perplexity's interface around sources is meaningfully better designed for the task.

Three specific features earn the subscription:

The Focus modes. You can scope your search to Academic (peer-reviewed sources), Reddit (genuine user opinion), YouTube, Writing (no sources, pure generation), or Math. The Academic mode in particular has saved me hours when researching medical, scientific, or policy questions. It pulls from journals, not just web articles.

Collections. You can save queries and follow-ups into a Collection - a sort of mini-research project that persists. Adding pages, asking new questions about old ones, all in one named context. For multi-week research projects, this becomes genuinely valuable.

The Pages feature, added in 2024, lets you turn a research session into a shareable, well-formatted page that compiles your findings. I've used this to write briefing docs for clients more times than I'm willing to admit.

Where it falls down

The biggest issue with Perplexity is the same issue with any AI tool that synthesizes from sources: the synthesis can be subtly wrong even when the citations are correct. The Brazilian municipality I mentioned at the top is one example. I have a dozen others. The model can read a source correctly and then summarize it inaccurately, sometimes inverting a key claim. This is not unique to Perplexity, but the citation-first design creates a false sense of security. You see the footnote and trust the sentence. The footnote is real. The sentence might not be.

The fix is: actually click the citations. Read what they say. Use Perplexity as a research starting point, not a research conclusion. I treat Perplexity like a research assistant whose work I always review, not like a librarian whose answers I take at face value.

The second issue is freshness. Perplexity is faster than competitors at indexing new content (recent news, fresh research papers), but it's not real-time. For breaking news or rapidly developing situations, you'll often get information that's hours out of date. The web search integration in ChatGPT Pro and Claude has been catching up here, and for time-sensitive queries those tools are sometimes faster.

The third: Perplexity occasionally refuses to commit to an answer. For controversial or politically charged topics, the model leans hard on "perspectives differ" framing, which is fine but sometimes frustrating when you want a direct answer to a factual question. This is a model-tuning choice, not a technical limitation, and it shows up in similar ways across major AI products.

Pricing - the cheaper of the major options

Perplexity is free with rate limits, $20/month for Pro, and $40/month per user for Enterprise.

The free tier is genuinely useful - five Pro Searches per day, which is enough for most casual research. Pro Search uses the more capable models and goes deeper on multi-step questions. The paid Pro tier is unlimited Pro Search, access to additional models (Claude Opus, GPT-5, Grok), and image generation thrown in.

Comparing pricing:

• Perplexity Pro: $20/month
• ChatGPT Plus: $20/month (broader features, weaker search)
• Claude Pro: $20/month (better writing, no search citation discipline)
• Google AI Pro (with Gemini): $20/month (tied to Google Workspace)

The sweet spot for Perplexity is the user who specifically needs research with verifiable sources. If you're not that user, the other $20/month tools are arguably better all-around values.

Where it's the right tool

• Research projects where you need to verify claims against sources
• Reading lists for unfamiliar topics - Perplexity gives you the survey and the sources, you read the sources
• Academic or scientific queries where sourcing actually matters
• Quick fact-checks ("did X actually happen this year?") where you want both the answer and the link
• Multi-step research where Collections help you keep context

Where I'd switch:

• For writing, use Claude or ChatGPT. Perplexity's prose is fine but the others are better.
• For coding, neither Perplexity nor general AI tools beat Cursor or Copilot in your editor.
• For long analysis where you're not citing sources, Claude's reasoning is sharper.
• For brainstorming and ideation, ChatGPT and Claude both do this better - Perplexity wants to find sources for everything.

The Comet browser play

Worth mentioning: Perplexity launched a browser called Comet in 2024 that builds search and AI directly into the browsing experience. It's interesting if you want AI-first browsing, but it competes with Arc's AI features, Brave's Leo, and any number of Chrome extensions. The browser strategy hasn't broken out the way Perplexity hoped, and it's not a reason to buy Pro unless you specifically want it. Most users will get more value from the standard web app.

Verdict

Perplexity is the AI tool I'd pay for first if my primary use case is research and fact-finding. The citation-first design is genuinely better than alternatives for that workflow, and the focus modes plus Collections make it more than just "search but AI."

It's not the right tool for everything. For pure writing, coding, image generation, or open-ended conversation, other AI tools are stronger. For specifically the task of finding things out and being able to check the work, Perplexity is the best in class.

The hidden value is the verification habit. Perplexity made me more rigorous about checking AI claims, not less. Other tools could learn from that.