Safe, Actually is a solo publication. I run the site, perform the hands-on testing, verify the evidence, make the editorial decisions, and publish the content myself.
I also use AI tools to assist with parts of the research and production process. AI helps me organize information, investigate sources, structure data, draft and edit content, create charts from real data, and assist with video production. It does not replace genuine hands-on testing or make the final recommendation for me.
When I say I tested something, it means I actually tested it. If I have not tested a particular device, location, feature, or use case myself, I will say so and explain what evidence I used instead.
My goal is simple: show you enough evidence to understand what worked, what did not, what it really costs, what other users report experiencing, and whether a product makes sense for your situation.
My Testing Approach
For hands-on VPN reviews and comparisons, I do not decide the winner first and then look for evidence to support it.
My general process is:
Research the question → decide what needs to be tested → get legitimate product access → run the tests → capture and record the results → verify current product information → research broader independent and user-reported evidence → compare everything → then reach the verdict.
My own testing is important, but it is not the only evidence I use. One test on one device, internet connection, location, or day cannot represent every user’s experience.
VPN Performance
When performance matters, I first test my normal internet connection without the VPN. That gives me a baseline.
I then connect to the VPN and repeat the relevant tests under conditions that are as consistent as practical.
Depending on the comparison, I may measure:
- download speed
- upload speed
- ping or latency
- nearby-server performance
- longer-distance performance
- connection stability
I record relevant testing information such as the device or operating system, connection conditions, VPN protocol where relevant, server or test location, and test date.
The results show what happened in my testing environment. They should not be treated as a guarantee that someone using a different device, internet provider, country, server, or connection will receive exactly the same result.
Leak and Kill-Switch Testing
When a VPN’s security features are relevant to the review, I check whether the connection exposes information it is supposed to protect.
Depending on the test, this can include:
- IP address exposure
- DNS leaks
- WebRTC leaks
- kill-switch behavior
For a kill-switch test, the important question is not simply whether the app has a button labelled “kill switch.” I want to know what actually happens to internet traffic when the VPN connection unexpectedly fails.
I report what happened in the test rather than repeating the provider’s claim that a feature exists.
Apps and Features
A feature listed on a VPN provider’s website does not automatically tell me how well it works or whether it is available on every operating system.
When I have access to the relevant device, I check the actual app.
Depending on the review, this may include:
- installation and setup
- server selection
- VPN protocols
- kill switch
- split tunneling
- auto-connect
- settings
- important device-specific features
- general ease of use
If I have not personally tested a feature on a particular operating system, I will not present it as firsthand testing.
Real-World Use
Benchmark numbers are only part of the story.
When a review is specifically about a real-world use case, I test that use case when I have legitimate access and can reproduce it properly.
Depending on the topic, that may include:
- streaming
- gaming performance
- website access
- connection stability
- travel-related use
- device-specific behavior
I record what worked, what failed, and the relevant conditions of the test.
I will not claim that I personally tested a VPN in a country, on a device, or with a service that I did not actually have access to.
Price and Renewal
VPN pricing can be confusing because the advertised monthly equivalent may be very different from what is actually charged at checkout or what the subscription costs when it renews.
Before publishing or materially updating a commercial comparison, I check current provider information and look for:
- the amount charged today
- subscription term
- effective monthly cost
- renewal price or renewal terms
- refund period
- device limits
- important plan restrictions
- whether an offer is limited to new customers
Where the information is available, I show the introductory price separately from the renewal price.
I normally use current text or tables rather than relying on a permanent pricing screenshot because prices and promotions can change quickly.
Privacy and Trust
I do not treat a provider saying “no logs,” “private,” or “secure” as proof by itself.
Depending on the review, I may examine:
- the provider’s privacy policy
- independent audits
- what information the provider says it collects
- ownership
- jurisdiction
- relevant security or privacy history
- public evidence that supports or challenges important claims
I distinguish between different kinds of evidence.
Provider claim — what the company says about itself.
Independent evidence — audits, independent testing, research, legal records, or other credible third-party evidence.
User-reported evidence — experiences reported through sources such as forums, review platforms, app stores, and online communities.
My testing — something I personally tested and recorded.
Those are not the same thing, and I do not present them as though they are.
When I Cannot Test Something Myself
There will be situations I cannot test firsthand.
For example, I may not own the required device, have access to a particular streaming service, or physically be in the country where a VPN’s reliability needs to be assessed.
In those situations, I do not pretend otherwise.
Depending on the question, I may use:
- official product documentation
- independent professional testing
- published audits
- reputable research
- clearly attributed user reports
I identify where the information came from and separate it from my own testing.
If the available evidence is too weak to support a confident conclusion, I would rather say that than invent one.
What Other Users Report
My own test tells me what happened under my testing conditions. It cannot tell me everything that thousands of other users may experience across different devices, countries, internet providers, servers, and app versions.
That is why I also research what other users report when their experiences are relevant to the decision.
Depending on the product and question, I may examine sources such as:
- Reddit discussions
- relevant specialist forums and communities
- Trustpilot
- G2 or similar software-review platforms where relevant
- Apple App Store reviews
- Google Play reviews
- provider community and support forums
- other credible sources where users describe their experiences
This can reveal recurring issues that may not appear during a limited hands-on test, such as:
- connection failures
- app bugs
- problems after software updates
- customer-support problems
- renewal or billing complaints
- streaming services repeatedly blocking certain servers
- country-specific reliability problems
- device-specific problems
I do not treat a single Reddit comment, forum post, Trustpilot review, star rating, or complaint as proof that something is generally true.
User reports can be mistaken, incomplete, outdated, manipulated, or affected by circumstances that are different from mine.
Instead, I look for recurring patterns across multiple reports and sources. Where possible, I then cross-check important claims against my own testing, official information, independent evidence, or other credible sources.
How I Use AI
I use AI tools as assistants throughout parts of the research and production process.
AI may help me:
- find and organize relevant sources for further investigation
- organize research and notes
- compare information across multiple sources
- identify recurring themes or reported user problems for me to investigate
- structure test data and comparisons
- draft and edit written content
- create charts from genuine test data
- plan video scenes and supporting visuals
- assist with narration, captions, and video production
I do not assume information is true merely because an AI system produced or summarized it.
Important claims still need to be checked against the underlying evidence and appropriate sources.
AI does not perform my physical or hands-on VPN tests. It does not turn generated screenshots into genuine product evidence, and it does not independently choose which VPN wins a comparison.
I remain responsible for evaluating the evidence and making the final editorial decision.
How I Reach a Verdict
I do not decide a winner solely from my own test results.
A VPN can perform differently depending on the device, operating system, internet provider, physical location, VPN server, protocol, app version, and use case. A result from my setup is useful evidence, but it is not the entire picture.
When reaching an important recommendation, I consider the relevant evidence together.
Depending on the comparison, that can include:
- my own hands-on testing
- current product features and limitations
- current pricing and renewal terms
- independent professional testing
- privacy policies and independent audits
- credible research and other independent evidence
- recurring experiences reported by users
- important problems and limitations
- the particular device, use case, and priorities the recommendation is intended for
Different people can therefore receive different recommendations.
A VPN that performs best in one benchmark does not automatically become the best VPN for everyone. A slightly slower service could still be the better choice for someone because of its price, privacy evidence, device support, features, reliability, ease of use, or another factor that matters more for that person’s situation.
I look at the overall weight of the relevant evidence and make the final editorial decision myself.
AI can assist me in researching, organizing, and comparing that evidence, but it does not decide the winner.
When Evidence Conflicts
VPN performance and product behavior can change, and different sources do not always reach the same result.
My test may sometimes differ from another independent test or from what other users are reporting.
When that happens, I look at factors that could explain the difference, including:
- test date
- device
- operating system
- app version
- VPN protocol
- server
- physical location
- internet connection
- test methodology
- the particular use case being tested
If the disagreement cannot be reasonably resolved, I report the uncertainty rather than forcing a definite conclusion.
Keeping Information Current
VPN apps, features, prices, streaming compatibility, server performance, policies, and ownership can change.
Where relevant, I record information such as:
- test date
- device or operating system
- VPN app version
- important test conditions
- date pricing was checked
I update important recommendations when new evidence materially changes the decision.
An old test result will not be quietly presented as though it happened today.
Affiliate Relationships
Some links on Safe, Actually may be affiliate links. If you buy something through one of those links, I may earn a commission.
That does not determine which product wins a comparison or which product I recommend.
The recommendation should come from the evidence and the needs of the person the content is intended to help.
If a cheaper option, a free option, another provider, or no VPN at all makes more sense for a particular situation, I will say so.
You can read the full Affiliate Disclosure for more information.
What I Will Not Do
I will not:
- claim I tested something I did not test
- fabricate speed, latency, leak, streaming, or security results
- create fake VPN screenshots and present them as genuine
- pretend I used a device I do not have access to
- pretend I tested a VPN in a country where I did not actually perform the test
- present AI-generated material as genuine test evidence
- repeat provider marketing claims as though they are independent facts
- treat one online comment or review as proof of a widespread problem
- hide an important downside because a company has an affiliate program
- hide relevant renewal costs
- present one test result as a guarantee of everyone’s experience
- let an affiliate commission decide the winner
- pretend there is a testing team, editorial team, or research staff behind Safe, Actually
Safe, Actually is currently run by me, with AI tools assisting parts of my research and production workflow.
If the way I operate or test products materially changes in the future, I will update this page so it continues to reflect how the work is actually done.
