Search for “best VPN” and most results are leaderboards, and most leaderboards repeat the same specs: number of servers, countries covered, streaming unblock lists. Those specs aren't useless, but anything that can be written into marketing copy can also be exaggerated. What really decides whether a purchase turns into a bad experience is four things you can check one by one: whether capacity is oversold, whether the server count is inflated, whether the service might disappear mid-term, and whether anyone answers when something goes wrong.

Below, each of the four is broken out with “what the problem looks like” and “how to verify before ordering”, then pulled together into a checklist you can tick off. Every price, billing period and refund window mentioned here matches what this site publishes; no unsupported numbers.

110+Countries covered
150+International routes
30day money-back guarantee
Unlimiteddevices

These four numbers are the reference points used throughout this article: the route count to check against the detail list, the refund window to shape how you pay, and the device count to judge whether adding one more device costs extra. They are not a recommendation on their own, but each one has to match the detail behind it.

Four traps, sorted out first

Treating “getting burned” as one vague feeling leaves you nothing to verify. Split into four categories, each has a matching check — and most of those checks can be done before you pay.

Four common problems and the checks that match them
Problem How it shows up What you can check before ordering
Overselling Fine during the day, noticeably slower at peak hours; the same route is nearly unusable when everyone is online Check whether a data allowance and reset rule are stated; test once in each of two time windows before deciding
Inflated server counts Advertised server count far exceeds what you can actually select; one machine listed under several names; only countries listed, never cities Ask for routes down to the city; check whether the route type is labelled; check whether each one can be selected individually in the client
Provider disappears Only long billing periods are sold; after payment, support and the site both vanish Buy the shortest period first; confirm where the refund terms are written and that the number of days is explicit
Support goes silent Tickets go unanswered for days; the reply is just “try restarting” Send one ticket before you order, asking a specific question

Of these four, overselling and inflated counts are the “invisible before purchase, painful in use” kind, while a vanishing provider and silent support only surface when something breaks. So verification comes in two steps: read the terms and route list before paying, then run a real test in the first few days after.

Overselling: a stated allowance beats “unlimited”

Overselling isn't jargon; it's business as usual in the bandwidth trade: a chunk of transit capacity is sold at peak rates, the total sold exceeds actual capacity, and nobody notices when few people are online — everyone just crowds in together when they are. Whether a provider shows restraint isn't about how fast it claims to be; it's about whether it dares to put a hard number on the allowance.

Three things you can check

  1. Is the data allowance and reset rule explicit? A service that states “X GB per month, reset on your activation date” at least gives you a boundary you can calculate; one that only says “unlimited” and never mentions any throttling condition usually keeps the limits out of sight. This site's monthly plans meter by the month: ¥9.9 includes 60GB, ¥18 includes 250GB, ¥28 includes 500GB, reset monthly on your activation date.
  2. Is there a use-it-up, never-expiring option? Low-volume users are better served by data packs: ¥158/300GB, ¥358/1000GB, ¥658/3000GB, valid until used up, not zeroed at the end of each calendar month. This also tells you something in reverse: the provider's revenue doesn't have to depend on you forgetting to cancel every month.
  3. Are route types distinguished? Two routes both labelled “Japan” can behave very differently at peak hours depending on whether they are IEPL dedicated, relayed or direct; a provider willing to spell out route types is usually willing to spell out capacity too.

How to test it yourself

No specialist tools needed. On the same device, against the same website, spend ten minutes in the morning and ten minutes around 9pm; if the gap is big enough to affect your use, that route's capacity doesn't match demand in your time window — switching routes or switching providers both cost less than putting up with it.

When testing, compare only the same thing: same website, same resolution, same device. Cross-device or cross-time comparisons tell you nothing and are easy to skew with your own impressions.

Inflated counts: break the server number into three parts

Server count is the easiest metric to water down. Counting an entry point as a server, counting several ports on one machine as several routes, or listing one city under several different names all inflate the number. Don't ask “how many are there in total”; ask “where exactly are they, what type of route is it, and can I select it individually”.

Three checks on server count
Check What to ask What a good answer looks like
City Which city does this route sit in Named down to the city, not just the country
Type Dedicated, relayed or direct Clearly labelled, with different route types listed separately
Selectability Can I pick it individually in the client Every route switchable on its own, not randomly assigned

This site publishes 110+ countries / 150+ routes, with the route table grouped by region and labelled by city and type, so you can check it directly. The number itself says nothing about quality, but it has to match the detail behind it: if a provider advertises 150+ routes and can't produce a table that names cities, the number is just copy.

Two smaller things worth a look

  • Do the streaming claims map to specific routes? Saying “supports Netflix” without saying which route is the same as saying nothing; “this route in this city works with this platform” is something you can actually verify.
  • How is route status displayed? Listing routes and their status is normal, but live numbers you can't verify yourself shouldn't be there. When you see metrics like “users online” or “99.9% availability”, treat them as decoration first.

Be cautious with words like “lifetime” or “permanent”. A one-off buyout promise has no channel for delivery once the service stops; by contrast, paying monthly, refund terms written as a number of days, and data packs that don't expire are how you keep the risk on your side.

Subscription and protocols: whether a route is real shows up in the client

Once you have the subscription, whether the routes actually exist is visible at a glance in the client. A subscription link is a personal address that belongs only to you; the client fetches it and gets the server list. The proper flow is to log in to your user panel and copy it there — not to grab a configuration of unknown origin from a so-called sharing site.

The usual order for importing a subscription

  1. Log in to the user panel, find the subscription entry and copy the full link.
  2. In the client, choose “Import from link” or “Add subscription”, paste it and update.
  3. After updating, check that the number of servers matches the route table, then switch through them one by one.
  4. Once the first connection succeeds, open a page that only shows your exit address, such as My IP, and confirm the exit matches the region of the route you selected.

Newer protocols aren't automatically better

Common protocols each trade off differently; there's no need to swear by one:

  • Shadowsocks: lightweight with a simple handshake, well suited to mobile networks; whether UDP works depends on the server-side implementation.
  • VMess: more parameters (UUID, transport), sensitive to system clock drift — if the time is out of sync, the handshake simply fails.
  • VLESS: lighter still, often paired with TLS-based transports, fewer configuration items but more dependent on the server's transport settings.
  • Trojan: runs over standard TLS, indistinguishable from ordinary HTTPS traffic at the port level; a typical misconfiguration shows up as “connects but nothing goes through”.
  • Hysteria2 / TUIC: built on QUIC (over UDP), usually steadier on lossy, long-distance links; the trade-off is a clear drop on networks that restrict UDP, where switching back to a TCP-based protocol is actually more reliable.

A practical test: if a provider offers both TCP-based and QUIC-based protocols and lets you switch freely in the client, it treats “networks differ” as a normal condition; one that offers a single protocol and answers every problem with “try again later” usually has limited troubleshooting depth.

DNS leaks and split routing

Two common problems remain after you connect. The first is DNS leaks: domain lookups don't go through the tunnel but are resolved by the local network, so the exit address changes while the query records stay local. To check, use a public DNS leak test page and look at which resolver is shown; if it belongs to your local ISP, turn on remote DNS in the client or point DNS at an address inside the tunnel. The second is split routing: in rule mode, local domains connect directly and everything else goes through the proxy, and an outdated rule list causes misjudgements that look like “some sites are fast, some won't open at all” — updating the rules usually fixes it.

Platform differences cluster here too. On macOS the client needs network extension permission, and when that isn't granted correctly the symptom is “shows connected but pages won't load”; on Windows the usual problem is the system proxy stacking on top of the client proxy; Android supports per-app proxying, so you can send only the apps you need through the route; iOS works through a system-level network extension and asks you to confirm on first install. None of these are route problems — check permissions and proxy stacking first, and suspect the server afterwards.

Payment and refunds: split one payment into two

You can't defend against a provider disappearing with technical means, only with payment structure. The principle is simple: buy the shortest period first, and only consider a longer period or a bigger data pack once you've confirmed it works.

  • Start monthly. A monthly subscription is the smallest possible trial cost; the ¥9.9 plan with 60GB is enough to run a full round of checks: import the subscription, switch routes, watch peak hours, try split routing.
  • Confirm where the refund terms are written. This site states a 30-day money-back guarantee; the terms should be visible before you pay, not heard from support afterwards.
  • Keep a payment trail. Alipay, WeChat Pay and USDT each suit different situations; what they share is a traceable payment record. Anyone asking for a private transfer outside public payment channels — walk away.
  • Data packs suit low usage. Valid until used up and never expiring means you don't have to buy a long period just to lower the unit price, and don't have to bet on “this provider still being here in a year”.

One detail that's easy to miss: how upgrades mid-term are handled. If you change tiers within a monthly billing cycle, the terms should spell out how the remaining days are calculated. This site's rule converts the difference by remaining days and is written in the plan description, so you don't have to ask support.

Support going silent: send one ticket before you order

Support quality can be tested before you pay, and the cost is one ticket. Pick a specific question — not “is it good?” but “which city is the Japan route in?”, “can I select Hysteria2 in the client?”, “what do I do if the network extension authorisation fails on macOS?”. The speed and specificity of the reply is roughly the level of support you'll get when something goes wrong.

Four questions worth asking

  1. Route detail: the city and type of a given route.
  2. Protocol support: which protocols are available in the client and whether you can switch freely.
  3. Troubleshooting path: when it connects but pages won't load, what's the first thing to check.
  4. Refund process: where to submit it and how long it takes.

If three of the four answers come back as “please wait” or “try again later”, that is your answer. Conversely, replies that contain concrete steps and page locations show the provider deals with these questions routinely, rather than starting to think about them when you ask.

Look at the help centre too. Having a help centre doesn't mean it helps: documentation that actually works says “you see this symptom → check this first → if it still fails, do this”, instead of repeating “check your network connection” three times. You can read the help centre before deciding whether to send a ticket, or the other way round — the point is not to discover after paying that nobody answers.

Pre-order checklist

Compress all of the above into one table and tick each item. If any item can't be confirmed, don't pay yet.

  • ✅ The route table names cities, not just countries
  • ✅ Every route is labelled by type: dedicated, relayed or direct
  • ✅ The data allowance and reset rule are stated openly
  • ✅ Refund terms are visible before payment, with the number of days stated explicitly
  • ✅ Monthly billing is available, and the monthly allowance is enough to run one full round of checks
  • ✅ Payment goes through public channels with a traceable record
  • ✅ A ticket sent before ordering gets a specific reply within a reasonable time
  • ✅ The client lets you switch protocols and routes freely, rather than assigning them at random
  • ❌ Only “unlimited” is written, with no conditions at all
  • ❌ The server count is huge, but no detail table naming cities can be produced
  • ❌ Only long periods are sold, or only private transfers are accepted
  • ❌ Tickets go unanswered for days, or the reply is a single “try restarting”

Bottom line: there's no universal answer to “which VPN is best”, but there is a universal process of elimination: first check whether the allowance is a hard number, then whether routes map to cities, then whether you can pay for just one month, and finally whether someone answers a specific question seriously before you pay. Pass all four and what's left is comparing price and experience; fail any one and it isn't worth trying even at a lower price.