English-Arabic Localization QA for AI Search

An Arabic page that answer engines cite has to stand on its own sources, not on a translation of the English one. That finding decides the scope here: I audit and specify the Arabic surface as a first-class page, with its own schema, its own Arabic-language citations and its own FAQ written from Arabic search behaviour. Projects that consist of pushing an English site through an engine and asking me to sign off the output, I decline. Engine choice: Decided by a blind test on 200 segments of your own content, scored independently by two Arabic reviewers on accuracy, register and terminology. Vendor claims are beside the point; what matters is how a given engine handles your product nouns, your legal sentences and your returns language. The test takes two days and has changed the answer on most engagements. Termbase: A TBX glossary covering product names, feature names, regulatory terms and the transliteration of every brand and person name, decided once and enforced everywhere. Three spellings of a founder's name in Arabic script is three entities to a model, and none of them are you. Register: Modern Standard Arabic on published pages for UAE and Saudi readers; Gulf dialect only where your team already writes it and can sustain it. Mixed register inside one page is the most common failure I find and the fastest tell of machine output to a reader in Dubai or Riyadh. Rendering: Bidirectional text, so IBANs, SKUs, currency codes, URLs and English product names sitting inside Arabic sentences keep their direction and their punctuation. Arabic-Indic against Western numerals decided per market and applied consistently across prices, dates and phone numbers. Logical CSS properties instead of left and right. Line breaking and truncation in components built for English string lengths, where your buttons break before your paragraphs do. Search and answer surface: Hreflang covering ar, ar-AE and ar-SA with a working x-default and self-referencing tags on both sides. Arabic queries are frequently answered from English sources, which means an Arabic page that only mirrors the English has no reason to be selected. It needs facts and sources a reader in that market recognises. Schema is written in Arabic rather than inherited, author entities included. What I hand over: Findings ranked by severity, the corrected termbase, fixed templates or a written spec your developers can implement, and a release checklist your team runs on every future Arabic deployment without me. Not included: I am not a translation vendor. No per-word work, no linguist staffing, no TMS administration. Languages beyond English and Arabic are out of scope; I refer them rather than supervise a language I cannot read. No Arabic market content strategy as a separate deliverable, though the audit tells you which pages have no business existing in Arabic. Who this is not for: Teams launching Arabic next month who need a sign-off rather than findings. A first pass on a site that has never had Arabic QA usually finds enough to move a launch date, and I would rather have that conversation before the engagement than during it. Also not for teams whose Arabic is handled entirely by an agency they cannot instruct, since every finding here lands as work somebody has to accept.