QA testing across languages, time zones, screens and locations
"Works on my machine" has a browser version: works in my browser. Your QA browser has your language, your time zone, your screen, your graphics card and years of cookies. Your users don't. Bugs that only show up for a visitor in another country, on a smaller screen or with a different system language slip straight through.
Separate browser profiles are a cheap way to test those conditions on purpose. This post shows a setup for QA and web teams.
What changes between real visitors
A page can behave differently depending on things the browser reports about itself:
| Signal | What can break |
|---|---|
Language (Accept-Language, navigator.languages) | Wrong translation served, untranslated strings, layout overflow in long languages |
| Time zone | Dates off by a day, "today" in the wrong place, cut-off times for orders |
| Location (IP and geolocation) | Geo-redirects, prices in the wrong currency, region-locked features, consent banners |
| Screen size and pixel ratio | Layouts that only fail at 1366×768, blurry images on high-DPI screens |
| Operating system | Font fallbacks, keyboard shortcuts, download links for the wrong OS |
| Fonts | Text that wraps differently and pushes buttons out of view |
| Storage | First-visit flows, onboarding, cookie banners that never show to you again |
Browser devtools can fake some of these for one tab. A profile changes them consistently: in every tab, frame and worker, and across restarts, the way a real visitor's browser would.
Build a small test matrix
You don't need hundreds of profiles. A dozen well-chosen ones cover most real-world differences:
- Markets: one profile per important market, each with a proxy in that country. Time zone, language and location follow the proxy automatically.
- Devices: a Windows laptop at 1366×768, a large macOS screen, a Linux desktop, and an Android phone profile for mobile layouts (the phone profile is a beta on the anti-detect kernel).
- States: a first visit profile you clear before each run (cookies and site data cleared at start), and a returning customer profile that keeps its logins.
Name them after what they test (QA · de-DE · Win 1366), put them in a QA lab group and tag them by market.
Make runs repeatable
A test is only useful if it can be repeated:
- Keep fingerprints stable. A profile keeps its device until you ask for a new one, so yesterday's run and today's run see the same "computer".
- Clear what the test needs cleared. Per profile, choose what is dropped before each start or after closing: cache, cookies, history, site data.
- Write the checks down as a flow. Open the page, wait for the main element, take a screenshot, read a price or a heading into the results, move on. Run it on all QA profiles at once, with a few in parallel. Screenshots and results are kept per run.
For code-based suites, the Local API starts any profile and returns a DevTools address. Playwright, Puppeteer and Selenium attach to it and drive the real profile:
const res = await fetch("http://127.0.0.1:50380/v1/profiles/PROFILE_ID/start", {
method: "POST",
headers: { Authorization: `Bearer ${process.env.OWLX_API_KEY}` },
});
const { endpoint } = await res.json();
const browser = await chromium.connectOverCDP(endpoint);
Use the profile's existing context and pages. Don't override the user agent, viewport or time zone in your test code — the profile already sets them consistently, and overriding one of them creates exactly the mismatches real users never have.
Check that the environment is what you think it is
Before trusting a run, make sure the profile really looks like the visitor it stands for:
- The check page that opens at start shows the profile's IP, location, time zone and language.
- The leak check confirms that the site sees only the proxy's IP, not the office's — otherwise geo-dependent tests silently test your own location.
- Check fingerprint in the profile's menu shows what a page can read from the profile.
Things to remember
- Test your own sites, or sites you have permission to test. Load tests and scraping of other people's sites aren't QA.
- Third-party services you embed (payment pages, maps, captchas) have their own rules about automated traffic; test them in their sandbox modes.
- Profiles on your computer's own operating system pass the strictest checks; use other operating systems for layout and behaviour tests, not to test anti-bot systems.
Different browsers see different websites. A small, stable matrix of profiles turns those differences from customer reports into test cases.