How we test
We don’t write reviews from spec sheets and marketing claims. We order the service or use access provided by the provider, install it, put a test project on it and measure whatever can be measured ourselves. Some links on our site are affiliate links, and we always clearly label paid partnerships. Commissions don’t decide how we rate a service, though. You’ll find the details in our affiliate disclosure.
This page explains how we test each type of service. In every review we also say which devices, app versions and configuration we tested with.
How we test VPNs
You can see the process in our reviews of Proton VPN, NordVPN and Surfshark. We take every VPN through these steps:
- Installing the official apps. We use the VPN in its official apps; in our current reviews that’s mainly the macOS app (for Proton VPN also Android). We note the app version and, where it matters, the build: NordVPN’s Mac app downloaded from its website offers different features than the Mac App Store version. Features on platforms we didn’t try ourselves we check against the provider’s documentation, and we say so in the review.
- Connecting and everyday use. We try quick connect, choosing a country, city and server, search, profiles and pausing the VPN, and we look at what the app shows once connected: the server, the new IP address, the server load and the protocol in use.
- Security features. We turn on the kill switch and check what it really does: with Proton VPN, devices on your local network such as a printer or NAS may stop working, while NordVPN’s Mac app quits selected apps instead of blocking all traffic. We go through the protocols (WireGuard and protocols built on it such as NordLynx, OpenVPN, and obfuscation protocols such as Stealth or NordWhisper), post-quantum encryption, DNS leak protection, malware and ad blocking, routing through more than one server (Secure Core, Double VPN, MultiHop), split tunneling and port forwarding.
- Extra services. When the VPN comes in a bundle, we try the extras too. With NordVPN we tested the antivirus with the EICAR test files and turned on Dark Web Monitor for one email address.
- Differences between platforms. The same VPN doesn’t do the same things on Windows, a Mac and a phone. We put the differences, for example in split tunneling or the kill switch, into a clear table by operating system.
- Our own speed test. We measure on a home connection in Czechia, first without the VPN to get a baseline. Then we connect to servers on five continents (Frankfurt, New York, São Paulo, Dubai, Sydney and Auckland; for NordVPN also Prague) and run the same test on each.
- Privacy and trust. We look at who is behind the service, where it is based and which law it falls under. We check whether the no-logs policy has been independently audited (by whom, when and in what scope), whether the apps are open source, whether the servers run from RAM only and how the provider handled past security incidents, such as the breach of a NordVPN server in Finland.
- Pricing. We take prices from the official pricing page and give the date we checked them. We separate the introductory price from the renewal price and state the commitment length, the money-back guarantee and the limits of the free plan.
What we measure in speed tests
- Download and upload in Mbps, measured through Cloudflare’s speed test servers with four parallel streams and two runs per location.
- Latency (ping) in milliseconds, measured as a ping to 1.1.1.1 through the VPN tunnel.
- Test conditions: the app version, the protocol (WireGuard or a protocol built on it, such as NordLynx) and the speed of the connection without the VPN, so the results can be put in context.
- Deviations: when something outside the VPN affects the test, we say so. In our NordVPN and Proton VPN tests, for example, Cloudflare served the Dubai test from Marseille, which is why the ping was higher.
Treat the results as a guide. The speed on your connection depends on the distance to the server, your network and the server’s current load. Results from different reviews aren’t always directly comparable either: we tested Surfshark on a connection of about 50 Mbps and NordVPN on one of about 80 Mbps. That’s also why we rely on our own numbers rather than on slogans like “the fastest VPN”. We don’t guarantee that a VPN will unblock streaming services, because that changes often. We recommend trying the service with the account you actually use, ideally while the money-back guarantee applies.
How we test web hosting and VPS servers
We test hosting and VPS servers on a real WordPress installation, not just with synthetic benchmarks. You can see the whole process in our benchmark of the Coolhousing VPS START.
- Ordering and setup. We order the plan and go through the whole process as a customer. On a VPS we install the operating system (for example Ubuntu via the noVNC console in the browser), log in over SSH and run our configuration script, which sets up Apache, MariaDB, PHP, Redis, Adminer and a Let’s Encrypt certificate. The script also creates a virtual host and deploys a clean WordPress.
- Administration. We go through the control panel: reinstalling the system, snapshots, monitoring, network settings, upgrading to a higher plan and the provider’s documentation.
- Test scenarios. We measure on a clean WordPress with the default Twenty Twenty-Five theme, no plugins and no cache, and on a WooCommerce store (Kadence starter theme, 1,000 products, 30 pages and posts, Redis and WP Fastest Cache). Commonly used plugins such as Rank Math and Contact Form 7 simulate a typical business website, and we generate test content with FakerPress and WP Dummy Content Generator.
- Page load speed. Google PageSpeed Insights (performance score, FCP, LCP, CLS and TTFB for desktop and mobile) and GTmetrix (TTFB, Fully Loaded Time, Speed Index, page size and number of requests).
- Server performance in WordPress. The Hosting Benchmark Tool and Speedtest Pro plugins (CPU and PHP, file system, database, WordPress core, network) and Query Monitor (page generation time, number and speed of SQL queries, peak memory use, Redis cache hit rate).
- Load tests. In k6 we run three modes: a sudden spike, a gradual ramp-up and sustained load (100 virtual users). We track the average response time, the 95th percentile, the maximum, requests per second and the error rate.
- Benchmarks of the VPS itself. Without WordPress we measure the network (Speedtest CLI: download, upload, ping), the CPU, memory and file operations (sysbench), random disk reads (fio, 4K blocks, four parallel jobs: IOPS, throughput, 99th percentile latency), disk latency (ioping), internal network throughput (iperf3) and static requests per second (ApacheBench).
- Uptime. We monitor the site with UptimeRobot (uptime percentage, number of outages, total downtime).
- Backups, support and price. We check snapshots, backup options and how restoring works. For support we note the available channels (for example 24/7 phone and email) and how fast and to the point the answers were during our testing. On price, we separate promotional and regular prices, prices with and without VAT, and the trial period with a refund.
We leave the server running for several days to see how it behaves beyond a one-off test. In the review we list the configuration (PHP version and limits, database version, web server, kernel), so the results can be compared with other providers. We also look at who runs the service and where the data centre is.
Other services and tools
We go through password managers, website builders, AI assistants and other online tools the way a new user would. We create an account, import data and try the main features on real tasks. In our Proton Pass review we built the whole demo vault with a single CSV import, generated passwords and created email aliases. In the Proton Lumo review we had the AI write an email and summarise a quote in a PDF, then checked that the summary matched the document. For the Bluehost AI Builder review we built a website with online booking from a single sentence, changed it through the chat and checked the page title, meta description, headings and image alt text in the source of the live site. These are the main criteria we look at:
- Real features vs. price: what the service actually does on each plan and whether it’s worth the money.
- Ease of use: how quickly someone who sees the tool for the first time finds their way around.
- Language and support: whether the interface, help and support are available in your language, and how the service handles content in other languages. The Bluehost AI Builder interface, for example, is in English only.
- Privacy and security: encryption, independent audits, open-source code, where your data is stored and, for AI tools, whether your data is used to train models.
- Platforms and compatibility: which systems, browsers and devices the apps are available for, and how the versions differ.
- Free plan, trial and refunds: what you get without paying and how long you can try the service without risk.
- Import, export and lock-in: how easily you can bring your data into the tool, and whether you can get it back out if you want to leave.
- Subscription terms: the commitment length, the price after the introductory discount ends and limits that the pricing page doesn’t show at first glance, such as the credits in Bluehost AI Builder.
When we haven’t tested something ourselves, we say so. Our guide to local AI hardware states plainly that we didn’t test the computers ourselves: every number comes from a cited source, and we compare speeds in a single program (llama.cpp) so the figures stay comparable.
How we rate
At the end of a review we rate the service in several categories on a scale of 1 to 5 (in half points), and the overall rating is based on them. The categories depend on the type of service:
- VPN: price and value, speed, features and apps, privacy and audits, ease of use, extra services.
- Web hosting and VPS: we sum up the results by area (CPU, memory, disk IOPS and latency, web server performance, external and internal network) and list the strengths and weaknesses.
- Other tools: for example security and encryption, features, ease of use, platforms and compatibility, and price and value. For AI assistants we also rate answer quality and models.
We always explain the score: below the rating we sum up what the service earned its points for and what to keep in mind, and sections such as “What we like” and “Things to watch out for” list the pros and cons, along with a note on who the service suits. The reader star rating below the article is separate and may differ from our editorial rating.
Review updates
Services change, so we keep our reviews up to date, especially when prices, plans or important features change. Every article shows the date it was published and last updated, and for prices we give the date we checked them.
We correct mistakes openly. In the NordVPN review, for example, we removed an earlier claim that AV-TEST had named NordVPN the fastest VPN, because it relied on an old speed test, and the review says so. When a service gets worse or shuts down, we update the review even if it’s a partner service.
Found outdated information in a review, or had a different experience with a service? Write to us.
