How to choose an antidetect browser: nine checks in an hour

A promise on a landing page is beyond your reach, and a browser answer is not: a platform reads it with the same lines you do.

Those answers get changed along one of two roads: inside the build itself, or by a script laid over a stock Chrome. The second road shows from the outside, and that is what the nine checks below catch.

None of them needs an account on a platform or a purchase. A trial profile, your everyday browser and a console are the whole kit.

1. 1

   Two profiles

   Create profiles that declare two different systems.

2. 2

   Your own window

   Open your everyday browser as a third window.

3. 3

   One comparison

   Whatever matches across all three came from your machine.

Each line below asks the browser for one value and sets it beside a second value that you already know.

A page may ask not for a font name but for a family: serif, sans-serif, monospace, cursive, fantasy and math. A live machine resolves all six into different fonts and returns a different line width for each.

Measured on 18 August on a Mac. A profile of the machine's own system answered with four different widths, and a profile of a foreign system returned all four as one and the same number.

Chromium never tells a platform's scripts the exact heap limit: the answer lands on one of a hundred steps, from 10 million to 3.76 billion bytes.

Read that number in the trial profile and in your everyday browser. A number between the steps, or one exactly equal to your real memory, means nobody was substituting anything here.

The same number also has to agree with the memory the profile declares beside it. A profile that names a modest memory and hands out a generous heap ceiling is two answers from two different machines.

The name of the video card is declared by the profile, while the state of the card is reported by your machine, and those two answers arrive from different places.

On a rented Windows server with no card, a measurement on 4 September returned webgl and webgpu in state disabled_off, and navigator.gpu appeared in none of six runs. The name of an expensive card beside that answer is a pair that does not occur.

Android hands a page three connection fields: type, downlinkMax and ontypechange. Desktop Chrome hands over none of them, which makes this the fastest test of a phone profile.

If the fields are there, ask for their values. Your own machine shows through at once: type reads wifi and downlinkMax reads infinity, while the profile declares a cellular connection.

A blocker's card shows “enabled” and “files found”, and its rule may still do nothing at all.

Measured with one and the same extension carrying one blocking rule: on one build the forbidden request reached the server, on another it did not. Load a blocker into the trial profile and watch whether the block ever reaches the network.

The same goes for any other extension you were promised. Load it, use it once, and judge by the result rather than by the green badge on the card.

A page asks for the number of cores and the platform name not only from its own window: it asks the same question inside an invisible frame and in a separate worker. A script laid over stock Chrome reaches the window and often fails to reach those two.

Open the developer window in the trial profile and ask the window for its core count, then create a worker and ask it the same. Two different numbers out of one profile are an answer about the vendor, not about your machine.

This row did not make the table of nine because it takes three lines of code rather than one. In return it separates substitution inside the build from substitution on top of it better than any other.

An eleventh question is not about substitution but about constancy: close the trial profile, open it tomorrow and read the same rows. A real machine answers the same way day after day, and a profile has to behave the same.

Here a profile's answers are derived from its seed, so yesterday's canvas, fonts and memory match today's to the byte, while a second profile differs. A vendor whose fingerprint changes from launch to launch sells you a new machine at every login.

The long part is not reading the answers, it is creating the profiles and wiring the proxies. The nine lines themselves run down the table in a few minutes.

A sensible order: fonts and memory first, then the video card, then the connection. The first two disagree more often than the rest and cost you one line each.

Nine answers split sellers into two halves: some change the values inside the build, others lay a script over a stock Chrome.

The difference shows in whether a profile answer matches your machine answer. The choice is yours, and you now make it before paying rather than after the first ban.