Technical SEO for Bilingual Sites: hreflang, Canonicals and the Mistakes That Cost Traffic

The core risk

Publishing the same content in two languages creates two URLs that a search engine must understand as alternates of one another rather than as duplicates or competitors.

Get the signals right and each language ranks for its own audience. Get them wrong and the engine may index only one, split authority between both, or serve the wrong language to a user — all of which are quiet failures that show up as unexplained traffic weakness rather than an error message.

URL structure

Choose one pattern and keep it:

• Subdirectories — example.com/en/, example.com/ar/. Simplest to operate, consolidates domain authority, and the usual recommendation.
• Subdomains — en.example.com. Workable but splits some signals and adds configuration.
• Separate domains — strongest geographic signal, most expensive to maintain, and authority must be built independently for each.

For a single organisation serving two languages, subdirectories are almost always correct.

hreflang

The hreflang annotation tells the engine that two URLs are language alternates of the same content. Three rules govern it, and violating any one silently invalidates the set:

It must be reciprocal. If the English page points to the Arabic page, the Arabic page must point back. One-directional annotations are ignored.

It must be self-referential. Each page includes an hreflang entry pointing to itself, in addition to its alternates.

It must use valid codes. ar for Arabic, en for English, optionally with a region — ar-SA, en-GB. Invalid or invented codes are discarded.

Add an x-default entry pointing to the page shown when no language matches the user, typically a language selector or the primary language.

Canonical tags

The most damaging bilingual SEO error is pointing the Arabic page's canonical at the English page. This tells the engine the Arabic version is a duplicate that should not be indexed at all — and it will comply.

Each language version is canonical to itself. The canonical tag deduplicates within a language; hreflang relates across languages. They do different jobs and must not be confused.

Translate more than the body

A page is not properly localised if only the article text changed. All of these need translating:

• Title and meta description — these are what appear in results and drive click-through
• URL slug, where practical, since it is a relevance signal and improves user confidence
• Image alt text
• Structured data fields containing human-readable text
• Breadcrumbs and navigation labels

Avoid machine translation as the published output

Search engines are explicit that unedited machine translation is low-quality content. Beyond ranking risk, it damages credibility with exactly the audience you were trying to reach — a technical reader recognises awkward translation immediately, and it undermines the expertise the content was meant to demonstrate.

Machine translation as a first draft that a fluent speaker then edits is a legitimate and efficient workflow. Machine translation published directly is not.

Do not auto-redirect by IP

Redirecting users to a language based on IP geolocation prevents crawlers from seeing both versions and frustrates users whose location does not match their language preference — a common situation for expatriates and multilingual readers.

Offer a language switch prominently, remember the choice, and let the user decide.

What to verify after launch

Confirm both language versions are indexed. Check that hreflang returns no errors in search console tooling. Search a distinctive phrase from each language version to confirm the right page surfaces. Watch impressions by language over several weeks — a version that never gains impressions is being suppressed by a signal problem, not by a content problem.