How to test a proxy provider before you pay

A trial tells you a provider works. The refund window tells you how long you have to find out that it does not, and it is the term almost nobody prints beside the trial they are advertising. Those windows run from 24 hours to 30 days. One of them stops counting at a gigabyte, whether or not the days are up.

Why the refund window matters more than the trial

Most providers offer a trial of some kind, a minority say plainly that they do not, and which ones publish one is a live list that moves, so it is not reproduced here.

The refund window is the part that does not move, and it is where the real terms live. Read off each provider's own pages and re-derived 27 August 2026, the windows run from 24 hours to 30 days. Where a provider offers no trial at all, the refund window is the trial.

The ends of that range, and the shapes worth knowing:

Three shapes in that list are not windows at all. Case by case is the absence of a published policy, which is reasonable for an enterprise vendor and risky for a small buyer. No refunds is at least honest. And two providers stop the clock on data rather than days, which is the one to read twice.

Twenty-three providers on the roster publish nothing either way, and that is recorded as unknown rather than as a no.

Three of the trials are free in the ordinary sense: Webshare, Decodo and NodeMaven. Everything else is a small purchase or sits behind a business account, and a business account means a verification step before you can test anything. If you go looking for other people's experience of Decodo first, search the old name too: it used to be Smartproxy, and the reviews are still filed there.

Read the condition, not the number

The refund column is where the real terms live, and one clause is worth calling out because it is easy to miss.

SOAX: within 3 days, under 1 GB used. ProxyScrape: within 3 days, under 500 MB used. A refund window with a usage cap is a different product from a refund window without one. If your test needs 5 GB to mean anything, both of those windows closed before you started.

Two more patterns in that column:

• "Case by case" appears three times, on Bright Data, Oxylabs and Froxy. That is not a refund policy, it is the absence of one written down, and it is a reasonable thing for an enterprise vendor to say and a risky thing for a small buyer to rely on.
• Infatica's 24 hours is the shortest window we found. It is enough to test connectivity and not enough to test a scrape job across a working day.

What to actually measure

A trial is not for confirming the provider works. It is for finding out whether it works on your targets, which is the one thing no comparison table can tell you, including ours.

1. Success rate, per target site, not averaged. A provider can look fine across ten sites and fail on the one you care about. Averages hide exactly the failure that would have changed your decision.
2. Response time on your real pages, not on an IP-echo endpoint. Those two numbers are unrelated.
3. Whether sessions hold. If your work needs the same address for ten minutes, test ten minutes, not one request.
4. How much data the run consumed. This is the sleeper. The number you record here decides which pricing tier you should be buying on, and it is easy to be out by 50x.

Point 4 is why a 500 MB trial is more useful than it sounds and a 1-day trial is less. Traffic buys you a measurement. Time only buys you a measurement if you are ready to spend it the day it starts.

Spending a small allowance well

If the trial is 500 MB and your pages are heavy, you have a few hundred requests. Do not spread them.

Pick your three hardest targets, send a hundred requests to each at the concurrency you plan to run, and keep the rest in reserve for the one that surprises you.

Block images and other assets unless you need them. On the five pages we measured, that one setting decides whether a 500 MB allowance is a real test or a rounding error:

The median page gives you about 19,000 document fetches or about 1,000 rendered ones. The heaviest gives you a hundred, which is not enough to test three targets at all.

And start the clock deliberately. A 3-day trial that begins when you click signup, on a Friday, is a 1-day trial.

Each provider's refund terms, with the date we read them, are on its own page: SOAX, Webshare and Oxylabs are three worth comparing, because they are the usage cap, the longest window and the case-by-case answer respectively. How we check covers what we do when a provider publishes nothing at all. Twenty-three of the sixty-three do not, and that is recorded as unknown rather than as a no.