Discussion starter: an international SEO question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.
- International SEO
Right-to-left languages like Arabic and Hebrew: which rendering or template issues hurt SEO for you, and how did you fix them?
Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.
1 reply
-
adminAdmin
The RTL problems that cost search visibility are mostly not about text direction itself. Google says it uses the visible content of a page to determine its language, and doesn’t use code-level language information such as
langattributes, or the URL. A missingdir="rtl"breaks layout and accessibility, but it doesn’t, by itself, make Google read an Arabic page as English.The issues that do cost visibility are template decisions that cluster on bilingual Arabic/English and Hebrew/English sites: both languages served on one URL, direction and language decided in the browser so the first HTML is the wrong version, prices and dates in structured data written with localised digits, slugs that exist in several Unicode spellings, and invalid hreflang codes. Direction bugs in titles, layout and fonts mostly hurt users and click-through. They are still worth fixing, but after the problems that change what Google can crawl, index and parse.
Public case studies that isolate RTL issues are thin, so what follows is documented behaviour, the mechanism behind each problem, and how to check for it.
Sort the problems by what they break
Problem Crawling, indexing or parsing Users and page experience Priority Both languages on one URL, switched by cookie or browser setting One language version never gets its own indexable URL Some visitors see the wrong language Fix first Direction and language set by JavaScript after load The first HTML carries the wrong language, or no text The layout flips after first paint High Localised digits in structured data Prices and dates may not parse None High for ecommerce Slugs in several Unicode spellings Duplicate URLs; canonicals and hreflang pointing at redirects Some broken links High on large catalogues Invalid hreflang codes Annotations ignored or read as the wrong market None Cheap, so fix now Mixed-direction titles and descriptions None Scrambled display, weaker click-through Medium Physical CSS, mirroring bugs, heavy fonts Rarely Overflow, hidden navigation, slow text Medium 1. One URL serving two languages
A pattern that shows up on bilingual sites is a language toggle in the header that sets a cookie and re-renders the same URL in the other language. Google’s guidance is direct about this. It recommends different URLs for each language version rather than cookies or browser settings, notes that Googlebot usually crawls from the USA and sends requests without an
Accept-Languageheader, and advises against automatically redirecting users based on what you think their language is. A site that picks the language from a cookie, that header or the visitor’s IP shows Googlebot one default, and the other language never gets a URL of its own to be indexed under.The fix is structural: one URL per language (subfolders such as
/ar/and/en/are the simplest to run), a switcher that links to the equivalent URL rather than setting state, reciprocal hreflang between each pair, and a suggestion banner instead of a forced redirect if you want to nudge visitors towards their language.2. Direction and language decided in the browser
Some single-page app setups load translations and set
document.dirandlangclient-side once the locale file arrives, sometimes reading the locale from local storage or a cookie. The consequences stack up. The server HTML ships with English strings or empty containers. The page paints left to right and then flips, which shows up as layout shift in field data. Crawlers and link-preview fetchers that don’t run JavaScript see only that first version. And if the locale file fails, the rendered page can fall back to English or show raw translation keys.W3C’s internationalisation guidance is to add
dir="rtl"to thehtmlelement when the document is right to left, never to use CSS to apply the base direction, and not to assume that a language declaration indicates direction: direction has to be declared withdir. The fix is to resolve the locale from the URL on the server and send<html lang="ar" dir="rtl">(orlang="he") with the translated text in the initial response.To check, compare view-source with the rendered DOM. Then run the URL Inspection live test on an RTL URL and look at the rendered HTML and the screenshot: the text should be Arabic or Hebrew, and the layout mirrored.
3. Localised digits in structured data
Number and date formatting libraries set to an Arabic locale can output Arabic-Indic digits and Arabic separators, so a price displays as
٣٬٤٩٩٫٠٠. That is right for readers. The damage happens when the template reuses the same formatted string in JSON-LD, Open Graph price tags or feed files. Schema.org’s definition ofpriceasks for the digits 0 to 9 (U+0030 to U+0039) rather than similar-looking Unicode characters, a full stop as the decimal point, and the currency inpriceCurrencyrather than as a symbol in the value. Merchant listings need a valid price on every Offer, so an unparseable price is not a cosmetic bug.Dates carry the same trap, plus a calendar one. Google’s Organization markup documentation, for example, asks for
foundingDatein ISO 8601 format. Show Hijri or Hebrew calendar dates in the visible text if your readers expect them, but generate machine-readable dates from raw values, in ISO 8601 with ASCII digits.The fix is two formatting paths, one for display and one for data, with JSON-LD built from the raw number and date, never from the display string. A unit test that fails when a price field contains anything other than 0 to 9 and a full stop stops the bug coming back.
Hypothetical example:
A Gulf electronics retailer runs
/ar/and/en/on one template. Its English product pages show merchant listing features and its Arabic ones don’t. The parsed Product markup in the Rich Results Test shows why: the Arabic template fillspricefrom the display formatter. Building the JSON-LD from the raw price fixes the Arabic pages with one template change, and a unit test keeps it fixed.4. Slugs that exist in several Unicode spellings
Arabic and Hebrew slugs are fine for Google. Its guidance is to use UTF-8 in URLs and to escape URLs properly when linking to them. The problems are consistency and length:
- Lookalike letters. Arabic yeh (ي, U+064A) and Persian yeh (ی, U+06CC), or Arabic kaf (ك, U+0643) and keheh (ک, U+06A9), look alike and encode differently. Text typed on a Persian keyboard, or pasted from another source, produces a slug that looks identical and is a different URL.
- Optional marks. Arabic diacritics (tashkeel), the tatweel stretching character (U+0640) and Hebrew vowel points (niqqud) appear in some titles and not others, so the same product can get two slugs.
- Normalisation forms. Alef with hamza can be stored as one code point or as a base letter plus a combining mark (NFC or NFD). The two look the same and percent-encode differently.
- Length. Each Arabic or Hebrew letter is two bytes in UTF-8, so it becomes six characters when percent-encoded. A 40-letter slug turns into roughly 240 characters in logs, analytics and any tool that truncates URLs.
Duplicates alone are manageable. The damage comes when canonicals, hreflang and sitemap entries are generated from a different spelling from the live URL, so they point at redirects or 404s. In Search Console that can surface as “Duplicate, Google chose different canonical than user”, or as language versions missing where hreflang says they should appear. The fix belongs in the CMS: normalise to NFC when the slug is created, map lookalike letters to one set, strip tashkeel, tatweel and niqqud from slugs (keep them in the content), redirect existing variants to the chosen form, and generate every reference to a URL from one function.
Native-script slugs or transliterated Latin slugs? Native slugs read naturally when shared and in displayed URLs; Latin slugs are shorter and safer in tooling. Public evidence that either ranks better is thin, so choose on operations and users, then stay consistent.
5. Invalid hreflang codes
hreflang values use ISO 639-1 language codes, optionally followed by an ISO 3166-1 alpha-2 region code. Two errors are specific to these languages. The first is
ar-AR, copied from thear_ARlocale some platforms use: as an hreflang value it reads as Arabic for Argentina. Usear, or a real region such asar-SAorar-AE. The second isiw, Hebrew’s old code, which some systems still emit (older Java-based platforms historically returnediwfor Hebrew). The current code ishe. This is the cheapest item on the list to audit: crawl the hreflang values and compare them with the two standards.6. Titles and snippets with mixed-direction text
A title that starts with a Latin brand or model name and continues in Arabic or Hebrew, or mixes in numbers and separators such as | and -, can display in a scrambled order, because the bidirectional algorithm reorders runs of text around neutral characters. W3C says browsers should apply the document’s base direction to the
titleelement, and that when right-to-left text begins with a Latin acronym you can prepend a right-to-left mark (‏). It also recommendsdir="auto"on forms and inserted text whose direction you don’t know in advance. Thetitleelement can’t contain markup, so invisible marks are the only tool there; in body content,<bdi>does the same isolating job for user names, reviews and product names from feeds.Check how affected titles display in a browser tab and in real results before and after the change. Google may rewrite titles, and how it treats directional marks isn’t documented, so treat this as a click-through fix and measure it on a small set of pages first.
7. Mirroring, hidden navigation and fonts
Physical CSS (left and right margins,
text-align: left, off-canvas menus positioned withleftoffsets) breaks when the page flips: drawers slide in from the wrong side, off-canvas elements cause horizontal overflow on mobile, and carousels start at the wrong end. W3C recommends start and end rather than left and right, which in CSS means logical properties such asmargin-inline-start. Indexing rarely suffers, because the text is still in the DOM, but users see a broken page. Adding RTL screenshots at mobile width to visual regression tests catches mirroring bugs before release.Arabic web fonts in particular carry many glyph forms and can be large, and swapping from a fallback font with different metrics moves text around. Subset carefully (Arabic needs its contextual letter forms and shaping features kept), preload the main weight, and tune fallback metrics with
size-adjustand related descriptors. Then segment real-user Core Web Vitals by language folder, because an RTL section can have worse field data than its LTR twin on the same template.One content-level check belongs here too. Because Google reads language from visible content, English navigation, footers and widgets make up a large share of what it reads on pages with little main text, such as a product page with a two-line description. Localise template text on RTL pages, and treat thin main content as a bigger risk there than on the English version.
What to check first, and what would change the order
- View source on an RTL URL:
langanddiron thehtmlelement, translated text in the initial HTML - One URL per language, a switcher that links to the equivalent URL, no redirects by IP or
Accept-Language - hreflang: valid codes (
arorar-SA,heorhe-IL), reciprocal pairs, URLs in the same encoded form as the live URL - JSON-LD, Open Graph tags and feeds: ASCII digits, full-stop decimals, ISO 8601 dates
- Slugs: NFC, one letter set, no tashkeel, tatweel or niqqud, with variants redirected
- Titles that start with Latin text: display checked in a tab and in results
- Logical CSS, RTL screenshots at mobile width, no horizontal overflow
- Fonts subset and preloaded, fallback metrics tuned, field data compared with the LTR version
Two Search Console patterns would change the order. If the RTL pages are indexed and earn impressions per page in line with the LTR version, but click-through or conversion lags, spend the effort on titles, layout and fonts. If RTL URLs sit in “Crawled - currently not indexed”, or show “Duplicate, Google chose different canonical than user” or “Alternate page with proper canonical tag” where you didn’t intend it, go back to URLs, canonicals, hreflang and main-content depth before touching presentation.
Need help with this on your own site?
Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.
Add a reply Cancel reply
Replies are open to approved members. Log in or apply to join.