Perplexity

Verdict

This is a documentation review, not a walkthrough: no free account was opened for this page and no screenshots were captured, so nothing here is a first-hand claim about what the signup flow looks like today. On the documented design, Perplexity is the most honestly structured product in consumer AI - it searches, then answers, then shows you the pages it used, which makes the answer checkable in a way a plain chat reply is not. The free tier is unusually substantial because the core behaviour is the free behaviour; the subscription mainly buys more of the stronger models and heavier research modes. The limits to understand before relying on it are that citations prove where a claim came from and not that the claim is true, and that the per-tier allowances are the part of this product that changes fastest.

Visit Perplexity
We have no affiliate relationship with this vendor. Link goes to the official site.

Read this first: what kind of page this is

AI Souk normally opens a free account, follows the vendor's own onboarding and captures dated screenshots of what a new user sees. That did not happen for this page.

So this is labelled a documentation review, which on this site means something specific: every statement below comes from the vendor's published material and from the documented behaviour of the product, and nothing below is a first-hand claim about the signup flow, the current free allowance, or what any screen looks like today. There are no screenshots because there is nothing to show that was actually captured, and inserting a decorative image here would misrepresent the page.

That matters most for one category of fact: the numbers. How many of the stronger searches a free account gets, which model sits behind which mode, and what the paid tier includes this month are the fastest-moving facts in consumer AI, and they are exactly the facts a documentation review cannot pin down. Where this page would otherwise quote one, it tells you to check instead.

What a documentation review can do well is explain what the product is for, how it differs in kind from the things it is compared to, and what to watch out for. That is the rest of this page.

What it actually does, in one paragraph

You ask a question. It runs searches, reads the results, writes an answer, and attaches numbered references to the pages that answer came from. You can then ask a follow-up and it keeps the thread.

That sequence - search first, then answer, then cite - is the whole product and the reason it belongs in a different mental category from a chatbot. A general assistant answers from what the model learned during training, which means it is fluent about the past and unreliable about the present, and in both cases there is nothing to click. An answer engine goes and looks, which means it can handle a question about this week, and it can show you where it got each sentence.

The practical consequence is a different relationship with the output. With a chat reply you are deciding whether to trust an assertion. With a cited answer you are deciding whether to trust a source, which is a question people are far better equipped to answer.

It is also why this is the AI product most often described as having replaced a search habit rather than added a new one.

Who it is for

Four people get obvious value, and they want it for the same underlying reason.

Anyone who needs an answer they will have to defend. If you are going to repeat a figure in a meeting, a cited answer with a link you can open beats a fluent paragraph you cannot verify. This is the strongest use and the one worth adopting a habit around.

Anyone researching something current. Prices, product changes, rules that came into effect this year - the questions where a training-data answer is most confidently wrong.

Anyone evaluating or comparing things. The question of what distinguishes two products is the kind of question that benefits from reading several pages, which is work this does well and quickly.

Students and anyone writing with references, with a serious caveat below about the difference between a citation and a check.

Who it is not for: anyone whose main need is drafting long prose, brainstorming at length, or working inside a document. That is a general assistant's job and this product is not trying to take it. People who do both tend to keep both open, which is a sensible arrangement rather than a failure of either.

How the free tier is shaped, and why that shape is unusual

The important structural fact is that the free tier contains the product's actual behaviour. You can ask a question, get a researched answer, see the sources and ask a follow-up without paying. In a category where free tiers usually exist to demonstrate a capability you cannot use for real work, that is genuinely unusual and it is most of the value score on this page.

What the subscription buys, structurally, is depth rather than a different product: more use of the stronger models, higher allowances on the heavier research modes, and the extras a vendor accumulates around a paid tier over time.

The allowances themselves are deliberately not stated here. Counts, model names and tier contents in this category change on a timescale of weeks, and this site's rule is that a number without a verified date is worse than no number - it looks authoritative and ages badly. Open the pricing page.

The practical advice that does not go stale: use the free tier properly for a fortnight on your real questions before considering a subscription. Because the core behaviour is free, you can evaluate the thing you would actually be paying for, which is unusual and worth exploiting.

The one limitation that matters most: a citation is not a check

This deserves its own section because it is the failure mode of the entire answer-engine category, and the better the product gets the easier it is to fall into.

A citation tells you where a sentence came from. It does not tell you that the source is correct, current, representative, or that the answer summarised it faithfully. An answer engine can cite a marketing page, a years-old forum post, a content-farm article or a press release with exactly the same confident formatting it uses for a regulator's own document. The numbers look identical in the text.

So the skill of using this well is one habit: on any claim that matters, open the source. Not all of them - that would defeat the purpose - but the one you are going to repeat, act on, or spend money because of.

The secondary habit worth forming is constraining the search. Asking it to restrict itself to a specific domain, or to recent material, changes the quality of what it has to work with, and it is the single most effective thing a user can do to improve the output.

The product is more honest than a chatbot precisely because it shows its working. That is an invitation to check, not a substitute for checking.

Where it sits against a general AI assistant

The comparison people actually want is against a general-purpose chat assistant, and framing it as better or worse misses the difference.

A general assistant is strongest when the value is in the generation: drafting, rewriting, structuring an argument, working through code, producing something that did not exist. Its weakness is provenance - it will not reliably tell you where a claim came from, and when it is wrong it is wrong fluently.

An answer engine is strongest when the value is in the retrieval: finding out, comparing, checking what is true now. Its weakness is composition - it is not the tool for a thousand words of your voice.

The way this plays out in practice is that people stop using one for the other's job and the complaints disappear. Questions beginning what, who, when, how much, and is it still true go to the answer engine. Requests beginning write, rewrite, and help me think go to the assistant.

The second comparison worth noting is against a plain search engine, and here the honest answer is that for a quick navigational query - a login page, an address - search is still faster. The answer engine wins when the question would otherwise have cost you four tabs.

What the documentation suggests about getting good results

Four things that follow from how the product is built rather than from anyone's personal habit.

Ask a specific question rather than a keyword. The product is reading pages and composing an answer, so a full question tells it what the answer should address. Keywords tell it nothing a search box would not already know.

Constrain the sources when the topic has a lot of noise. Telling it to look only at a particular site, or only at recent material, is the difference between an answer drawn from documentation and an answer drawn from whatever ranked.

Use the follow-up rather than starting again. The thread keeps context, and narrowing is how research actually works - the second and third questions are usually the valuable ones.

Ask it what it did not find. A question about absence - whether anything contradicts this, whether a figure is disputed - produces more useful material than one more confirming question, and it exposes the cases where sources disagree.

None of this is a trick. It all follows from the fact that the quality of the answer is bounded by the quality of the pages it read.

Privacy, data and what to check on the tier you use

Three things worth reading in the vendor's own policy before putting anything sensitive into any AI product, and this one is no exception just because it is mostly a search tool.

Whether your queries are used to train or improve models, and whether that differs by tier. This is the setting people most often assume rather than check, and it is usually stated plainly in the privacy policy.

Whether there is a control to turn that off, and whether history can be deleted. Both are commonly available and commonly unused.

What happens to uploaded files, if you upload any - retention is a separate question from training, and the answer is often different.

The general rule that applies regardless of what any policy says today: a query is a disclosure. Client names, unreleased plans, personal medical details and anything under an agreement do not belong in a consumer-tier AI search box, not because the vendor is untrustworthy but because the question itself has now left your building. For work covered by a confidentiality obligation, that is a contractual issue rather than a preference.

This page is a documentation review and cannot verify current settings or policy wording for you. Read the policy at the tier you are on.

Support, and the realistic expectation

The support score here is the lowest of the four, and it is not a criticism of this vendor specifically - it is the honest grade for consumer AI at this price point generally.

What exists is self-service: a help centre, documentation and a community. For a product with few moving parts and no configuration to speak of, that is mostly adequate, because the questions that arise are usage questions rather than failure questions.

What does not exist at a consumer subscription price is a support relationship. If an answer is wrong, the route is not a ticket, it is checking the source yourself. If billing goes wrong, expect email and a wait. Anyone needing accountable support for an AI tool is looking at a business or enterprise arrangement, which is a different purchase.

The mitigation is the product's own transparency: because it shows its sources, most problems are diagnosable by the user without help, which is more than can be said for a tool that answers from nowhere.

Actual response times cannot be graded from documentation, and this page will not invent them.

Frequently asked

Is the free tier actually usable? Yes - unusually so for this category, because the core behaviour is free. That is the main reason the value score is high.

Does it replace a search engine? For questions, largely. For navigation - finding a login page, an address, a specific site - plain search is still quicker.

Are the citations reliable? They reliably show where a claim came from. They do not establish that the source is right. Open the ones that matter.

Can I use it for academic work? It is useful for finding material and dangerous as a substitute for reading it. Your institution's rules on AI assistance are yours to follow, and a reference you have not opened is not a reference you can cite.

How many searches do I get? Deliberately not answered here. Allowances change frequently and a stale number is worse than none - check the current pricing page.

Why no screenshots? Because none were captured for this page. It is labelled a documentation review for that reason, and the method page explains the distinction.

Does this page earn a commission? No. There is no affiliate relationship with this vendor; the link goes to the official site.

Verdict, and what this page cannot tell you

Perplexity is the AI product easiest to recommend without caveats about hype, because what it does is modest and legible: it looks things up and shows you where it looked. The free tier contains that behaviour rather than teasing it, which makes trying it a genuinely free decision and makes the subscription an informed one.

Adopt it for questions you will have to defend, for anything where currency matters, and for comparisons that would otherwise cost you a morning of tabs. Keep a general assistant for writing, and stop expecting either to do the other's job.

The one habit that separates good use from bad: open the source on any claim you intend to repeat. The citations are an invitation to verify, and a reader who treats them as a guarantee has reintroduced exactly the problem the product was built to solve.

What this page cannot tell you, stated plainly because the site's method requires it: what the signup flow looks like today, what the current free allowance is, which model sits behind which mode this month, or what the paid tier includes. No account was opened and no screenshots were captured. Those facts live on the vendor's pricing page, and they change more often than this review will.

Visit Perplexity
We have no affiliate relationship with this vendor. Link goes to the official site.