Every large language model has a knowledge cutoff, and the practical cost of that is getting an answer that was true two years ago. You ask ChatGPT about a funding round that closed last month and it tells you it does not know. Perplexity takes the opposite approach: it searches the live web and answers with the sources attached, which turns fact-checking from a chore into part of the output. We spent our testing window on the questions where that difference shows up, from current prices to papers published weeks earlier, and also on the places where Perplexity quietly loses to its more famous rivals. It costs $20/month, the same as the other two major AI subscriptions, and there is a free tier if you want to test the search quality before paying.
Pros and cons at a glance
Everything below comes from our own testing, including the source-verification checks and the image analysis described further down.
What works well
- Real-time web search answers questions the competition cannot, because it is not relying on a training cutoff
- Every answer carries cited sources, so you can verify a claim in seconds instead of trusting it
- Factual lookups are faster than ChatGPT, which spends time hedging about what it does not know
- Vision analysis works, and cites what it found rather than describing an image in general terms
- An API is available if you want the same search layer inside your own product
- Pro voice mode handles hands-free queries when you are not at a keyboard
What to watch out for
- Creative writing is noticeably weaker than ChatGPT or Claude, with more generic output
- Code generation lags both rivals, so it is not a primary development tool
- The context window is smaller than Claude’s, which matters for whole-document work
- Interface polish trails ChatGPT, and search features sit in odd places
- Source quality is inconsistent, with occasional answers leaning on marginal blogs
Our score breakdown
Five categories, scored out of five, averaged for the headline number. Perplexity reaches 4.7, with value as its strongest result and support as its weakest.
| Category | Score | What drove the score |
|---|---|---|
| Performance | 4.8/5 | Fast, sourced answers on anything current |
| Value | 4.9/5 | A real free tier, and $20 buys a capability the others lack |
| Features | 4.7/5 | Search, vision, API and voice, with a thinner creative toolkit |
| Support | 4.5/5 | Adequate documentation, slower human help |
| Ease of use | 4.8/5 | Simple to start, less tidy to navigate than ChatGPT |
Why live search changes the answer
The clearest way to explain Perplexity is to run the same question through it and through a model with a training cutoff. We did that three times during testing, and the pattern held.
The first test was a price check. Asked for the current price of Bitcoin, ChatGPT could not answer at all, since the number moves by the hour and its knowledge ends months before the question. Perplexity returned the price with three exchange sources attached. We checked it against Coinbase in about ten seconds and the figure matched. That is the whole argument for this tool in one interaction: the answer arrived with its evidence, and verification cost seconds rather than a separate research pass.
The second test was a news question, asking about the status of a funding round. ChatGPT replied that as of April 2024 it did not know. Perplexity came back with current coverage from TechCrunch and the Wall Street Journal. For anyone writing about markets, technology or policy on a deadline, that capability is the difference between publishing and guessing.
The third test was academic, asking about the effectiveness of GLP-1 drugs for weight loss based on research published during the year. Perplexity returned peer-reviewed papers from January through March, while ChatGPT’s knowledge stopped in April of the previous year. If your work depends on the last twelve months of a fast-moving literature, search-based retrieval is not a convenience, it is the difference between a current and a stale review.
Source quality is the part you still have to check
Citations solve the trust problem in theory and only partly in practice. Some Perplexity answers cite peer-reviewed journals, government data or established outlets, and those you can accept quickly. Others lean on blogs that are thin on credentials, or on aggregator pages that themselves summarize a primary source you would rather read. The tool gives you the links; it does not audit them for you.
In practice that changes how you work rather than whether you can trust the output. Instead of wondering whether a claim is current, you spend your attention on whether the source deserves the citation. For a research workflow that is the right allocation of effort. For someone who wants to copy an answer into a report without reading the footnotes, it is a trap, and it is the same trap that makes models with training cutoffs feel safer than they are.
Vision analysis, tested on real screenshots
The vision features are more useful than the feature list suggests. We uploaded a screenshot of a pull request and asked which issues it contained. Perplexity identified security concerns in the diff and referenced the exact line numbers, which is the sort of output that saves a reviewer a first pass over a large changeset. Image analysis with citations attached is a small distinction on a spec sheet and a meaningful one when the image is code, a chart or a dashboard, since you can follow the reasoning rather than accepting a description.
Where this does not help is anything requiring sustained context across files. The vision feature reads what is in front of it, and for multi-document reasoning a model with a larger context window is the better tool.
Where Perplexity is weaker
Creative writing is the clearest gap. Asked to draft a product marketing email, the output was competent and generic: correct structure, acceptable claims, no voice. For templates and first drafts that is fine. For copy where tone carries the message, both ChatGPT and Claude produce better material, and the difference is obvious when you read the two side by side.
Code generation is weaker still, and this is not a close call. Both ChatGPT and Claude outperformed Perplexity in our testing on development tasks, so it should not be your primary coding assistant. Where it earns a place in a developer’s workflow is research-adjacent work: library comparisons, checking whether a dependency has been deprecated, finding what changed in a release. Those are search questions wearing a developer’s clothes.
The interface is functional without being refined. Menu organisation feels scattered and the search controls are not where you would expect them, which we scored as roughly 7/10 against ChatGPT’s 9/10. None of this makes the tool hard to use for its core job. It does mean the transition from a ChatGPT habit takes a week of adjustment.
Pricing and who should pay
Perplexity Pro is priced at $20/month at the time of writing, which puts it in a straight line with two other subscriptions most buyers are weighing: ChatGPT Plus at $20/month and Claude Pro at $20/month on the monthly plan, or $17/month if you pay for a year. The free tier is genuinely usable for search, so the honest advice is to test your own most common questions on the free version first. If the answers come back with the sources and dates you need, you may never need to pay.
| Tool | Monthly price | Best at |
|---|---|---|
| Perplexity Pro | $20 | Current facts, cited research, verification |
| ChatGPT Plus | $20 | General work, code, files, broad capability |
| Claude Pro | $20 monthly, $17 monthly on annual | Long documents, nuanced writing, research synthesis |
The decision comes down to what share of your AI use is time-sensitive. If most of it is, Perplexity is the one subscription worth paying for. We reviewed the other two assistants in the same depth, so you can compare the detail directly: ChatGPT Plus for general work, code and files, and Claude Pro for long documents and nuanced writing. If your interest is how AI-generated material reaches social platforms rather than how it is produced, our piece on Meta AI and public Instagram photos covers that ground. If you mostly write, code or work with documents you already own, a general-purpose assistant earns its fee more often, and the search capability matters only occasionally. Plenty of people end up paying for two, which is a reasonable outcome as long as the second is the one you actually open.
Who should buy it
Buy Perplexity Pro if your work depends on what changed recently: journalists checking facts before publication, researchers tracking a fast-moving literature, analysts comparing current prices or announcements, and anyone who has to defend a claim with a source. The verification habit it encourages improves the work even when the tool is wrong, because you are always one click from the primary source.
Skip the paid tier if your usage is mostly drafting, coding or working through documents you control, and keep the free tier for the occasional search. Writers and developers will get more per dollar from a general-purpose assistant, and the researchers who need long-document analysis will prefer a model with a bigger context window.
Verdict
Perplexity Pro solves the one problem that makes AI assistants unreliable for current work, and it solves it with citations rather than confidence. The free tier alone is a better research tool than a chatbot with a stale training set, and the paid tier buys speed and headroom for people who ask current questions all day. It is not the assistant for creative writing or code, and it is not the most polished interface in the category. If your questions are about now, it is the subscription that earns its $20.
Frequently asked questions
Is Perplexity Pro worth $20 a month?
If a meaningful share of your questions are about current events, prices, or research published in the last year, yes. The citation-first approach saves a verification step on every answer. If you rarely need current information, the free tier covers most of the value.
How is Perplexity different from ChatGPT?
ChatGPT answers from a trained model with a knowledge cutoff, so questions about recent events often fail. Perplexity searches the live web during the query and returns answers with sources attached, which makes it more accurate on anything time-sensitive and weaker on creative writing and code.
Can I trust the sources it cites?
They are real and checkable, but quality varies. Some answers cite journals and established outlets, others cite blogs. Treat the citations as a starting point and read the primary source before repeating the claim.
Does it handle images?
Yes. Vision analysis recognises charts, screenshots and diagrams, and cites what it found. In our test it read a pull request screenshot and pointed to specific line numbers, which is useful for a first pass on code review.
Is it good for coding?
No, not as a primary tool. Both ChatGPT and Claude produced better code in our testing. Perplexity is more useful for developer research, such as comparing libraries or checking whether a package is still maintained.
Is there a free version?
Yes, and it is usable for search. Test your own recurring questions on the free tier before subscribing, because search quality is the whole product.
Ask about something that happened this week
Perplexity Pro costs $20/month with a free tier available, so you can test the search quality before deciding.
King
Independent editors covering practical technology, online services and better buying decisions.
