Define the expected destination for each entry
Choose a pair of equivalent public pages, such as an Arabic service page and its English counterpart. Record their exact addresses and expected content. Distinguish an entry to the site's neutral homepage from an explicit language URL: those entry points may be governed by different product decisions and should not be tested as though they were identical.
Google recommends separate URLs for language versions and advises against automatically moving visitors between them based on an assumed language. Use that guidance when reviewing the intended behavior with the owner. A visitor following a shared Arabic page should have a predictable way to read it and choose another language.
Build a small session-condition matrix
Test each explicit URL in a fresh session, then in sessions with an Arabic or English preference already selected. Record the browser language setting and the sequence of addresses reached. Include a request without an Accept-Language header in the technical test where practical; Google's guidance notes that Googlebot requests do not set that header.
Change one condition at a time. Clearing a preference cookie, switching browser language and changing the entry URL together may stop the symptom without revealing its cause. Preserve a failing combination and its trace before trying a repair. Do not include the full cookie value or private session credentials in shared evidence.
Find which rule sends the visitor back
Record the first transition and the transition that returns to an earlier address. Identify whether each occurs through an HTTP response or after the page loads. Compare CDN, server and application rules against that sequence. A language loop often looks like one browser error, but the useful finding is the pair of decisions that disagree.
Check the preference update when a visitor deliberately switches language. Determine whether the next request reads an old value or whether another layer overrides the selected URL. Treat these as hypotheses until the request trace and configuration agree. Do not remove all redirects just to make one failing session stop looping.
Worked example: cookie preference versus browser preference
In an illustrative case, the server sends /en/service/ to /ar/service/ because a stored preference says Arabic. A browser script then sends the Arabic page back to English based on the browser setting. The repeated pair of addresses explains the loop; neither the page translation nor its heading is the source of this particular conflict.
The team agrees that an explicit language URL and deliberate language choice should remain usable, then removes the conflicting automatic override in the appropriate layer. It tests both pages with old preference states as well as a fresh session. The example is not a recommendation to delete preference storage indiscriminately or to alter unrelated authentication cookies.
Verify stable access after the change
Repeat every row in the session matrix, including direct links and a switch in both directions. Confirm that the destination contains the corresponding page rather than only a language homepage. Reload and revisit the address to check that the result stays stable. If the translation does not exist, document that exception instead of inventing a supposedly equivalent destination.
Save the traces, agreed precedence and changed configuration with your website-audit action list. Keep the language-selector interface and hreflang review as related but separate checks. A stable navigation path establishes access under the tested conditions; it does not by itself establish that all language versions have been crawled or indexed.