How we test and compare tools
Every rating published on Comparateur-Tech is based on a weighted grid of criteria, public and identical for all tools in the same category. This page explains what we measure, how we measure it, and what guarantees the independence of our rankings.
Last updated:
Our four-step protocol
Selection
We choose which tools to test based on their relevance to our readers: market share, reader requests, notable newcomers. No vendor can buy a spot in a comparison.
Testing
Each tool is used in real conditions: account creation, configuration, repeated measurements over several days. The criteria measured depend on the category (see below).
Weighted rating
Results are converted into a score out of 10 using a weighted grid that is public and identical for all tools in the same category. Scores are never adjusted for commercial reasons.
Re-evaluation
Prices, plans and performance change. We re-evaluate major comparisons at least once a quarter, and every page displays its last-updated date.
Criteria and weightings by category
A VPN cannot be rated like a web host. Each category has its own grid; the weightings below are the ones actually used to calculate the scores displayed on the site.
VPN
A VPN is judged first on what it promises: protecting your connection without slowing it down excessively. We measure speed with and without the VPN, verify the absence of leaks, and analyze what the provider can actually know about its users.
| Criterion | Weight | What we measure |
|---|---|---|
| Speed | 25% | Download and upload throughput with/without VPN, latency, stability across several servers (France, Europe, United States), measurements repeated at different times of day |
| DNS / IP / WebRTC leaks | 25% | DNS, IPv4/IPv6 and WebRTC leak tests on every app, kill-switch behavior when the connection is forcibly dropped |
| Jurisdiction and no-log policy | 20% | Country of incorporation and applicable laws, actual content of the privacy policy, published independent audits, track record (legal requests, incidents) |
| Features | 15% | Kill switch, split tunneling, available protocols (WireGuard, OpenVPN), multi-hop, simultaneous devices, quality of the apps |
| Pricing | 15% | Real price after the first commitment period, refund conditions, renewal transparency |
Web hosting
For a web host, the two numbers that matter are availability and response time. We measure them over time with a test site, rather than repeating the figures advertised by the host.
| Criterion | Weight | What we measure |
|---|---|---|
| Uptime | 25% | Availability measured on a test site through an external monitoring service over at least 30 days, compared to the contractual guarantee (SLA) |
| TTFB and performance | 25% | Time To First Byte measured from France, full load time of a typical page, behavior under light load, presence of caching and a CDN |
| Support | 20% | Observed first-response time (chat, ticket), availability in French, technical relevance of answers to real questions |
| Features and limits | 15% | Automatic backups, SSL certificates, number of sites and databases, real limits of “unlimited” plans, ease of migration |
| Pricing | 15% | First-year price versus renewal price (often very different), hidden fees, refund conditions |
Antivirus
We do not run a virology lab: for detection rates, we rely on results published by the leading independent laboratories (AV-TEST, AV-Comparatives), which we cross-check and complement with our own measurements of system impact and usability.
| Criterion | Weight | What we measure |
|---|---|---|
| Detection rate | 30% | Results from the latest AV-TEST and AV-Comparatives tests (real-world protection and detection of widespread malware), consistency of scores over time |
| System impact | 20% | Measured slowdown at startup, when opening applications and during a full scan, idle memory usage |
| False positives | 15% | Number of false alerts reported by independent laboratories, observed behavior with common legitimate software |
| Features | 20% | Firewall, web and anti-phishing protection, ransomware protection, real scope of advertised modules (bundled VPN, password manager) |
| Pricing | 15% | First-year and renewal prices, number of devices covered, commercial practices (auto-renewal, cancellation) |
AI tools
Generative AI tools are harder to rate objectively: quality depends on the use case. We run an identical series of tasks for all tools in the same subcategory, and we give significant weight to data privacy, often neglected in this sector.
| Criterion | Weight | What we measure |
|---|---|---|
| Output quality | 30% | A series of standard tasks identical for all compared tools (writing, code, images… depending on the subcategory), assessed on reliability, relevance and consistency of results |
| Pricing and limits | 20% | Plan prices, real quotas (requests, tokens, generations), concrete differences between free and paid tiers |
| Data privacy | 20% | Whether user data is used for training, opt-out availability, data location, stated GDPR compliance |
| Integrations | 15% | Available API, native connectors, data export, compatibility with common work tools |
| Ease of use | 15% | Interface clarity, learning curve, documentation quality, availability in French |
Our independence
Comparateur-Tech is funded through affiliate links: some links to vendor websites earn us a commission when a reader signs up. This model keeps the site free and without intrusive ads, but it comes with strict rules.
- Rankings are never influenced by affiliation. Scores are calculated from the weighted grid before any commercial consideration, and a tool with no affiliate program can rank first in a comparison.
- No vendor can buy a score, a ranking position or the removal of criticism. We refuse sponsored articles presented as reviews.
- Affiliate links are disclosed and tagged rel="sponsored" in line with Google's guidelines.
Details about our funding are available on the About page.
Hardware and testing conditions
So that measurements are comparable from one tool to the next, all tests are run under the same conditions:
- Machine : [TO BE COMPLETED: model, CPU, RAM, OS, e.g. Windows 11 laptop, Ryzen 7, 16 GB]
- Connection : [TO BE COMPLETED: type and speed, e.g. symmetric 1 Gb/s fiber, ISP X]
- Location : [TO BE COMPLETED: test city or region, e.g. tests run from France (Auvergne-Rhône-Alpes region)]
- Measurement tools : [TO BE COMPLETED: e.g. Speedtest/nPerf for throughput, ipleak.net and dnsleaktest.com for leaks, UptimeRobot for monitoring, WebPageTest for TTFB]
- Protocol : every measurement is repeated at least three times at different times of day; we keep the median, not the best result.
The limits of our method
No methodology is perfect, and we prefer to be upfront about the limits of ours. Our measurements reflect our testing conditions: your VPN speeds will depend on your connection, and a host's TTFB varies with your visitors' location. A good score does not mean a tool suits everyone, it means it keeps its promises within its category.
Major comparisons are re-evaluated at least quarterly, and systematically whenever something significant changes (new pricing, security incident, published audit). Every page displays its last-updated date.
Spotted an error or outdated information? Let us know: we correct and credit relevant reports.