AI and machine translation for your UAE website: when it is safe, and when it hurts SEO, trust and legal accuracy
By Dr. Hani Elimam Manager & CEO, SmartTech Studio
The pitch is hard to resist. Paste your English website into an AI tool, and a few minutes later you have an Arabic version, ready to publish, for almost nothing. For a UAE business that knows it should have had an Arabic site years ago, that sounds like the problem solved. Sometimes it comes close. More often, what goes live is an Arabic site that reads like a manual nobody edited, states a price or a refund term slightly differently from the English, and ranks for nothing, because the words Arabic speakers actually search for were never on the page.
This article is for owners and marketing managers deciding how to produce the Arabic (or the English) half of a bilingual website in the UAE. It sets out what Google’s own documentation says about automatically translated content, how search engines handle two language versions of one site, where machine translation is perfectly reasonable, where it does damage, and a workflow that uses the speed of AI without publishing its mistakes.
We write from two seats. SmartTech Studio builds bilingual Arabic and English websites and apps, and our sister company, Smart World Legal Translation, is a legal translation firm, so we see both the search side and the side where one wrong word in a contract matters. Our own localization service is done by human linguists. The workflow below is what we recommend to any team that chooses to start from a machine draft.
What does Google actually say about machine-translated pages?
Google does not ban machine translation, and it does not reward it either. What it judges is the result. Its Search Central post on AI-generated content (developers.google.com/search/blog/2023/02/google-search-and-ai-content) describes a “focus on the quality of content, rather than how content is produced”. In the same post it warns that “using automation—including AI—to generate content with the primary purpose of manipulating ranking” is a “violation of our spam policies”.
The spam policies themselves (developers.google.com/search/docs/essentials/spam-policies) are more specific. Scaled content abuse is described as many pages generated “for the primary purpose of manipulating” rankings, typically “large amounts of unoriginal content that provides little to no value to users, no matter how it’s created”. Among the examples, Google lists scraping content to generate many pages, “including through automated transformations like synonymizing, translating, or other obfuscation techniques”, “where little value is provided to users”. Its advice for content of that kind is blunt: “If you’re hosting such content on your site, exclude it from Search.”
Read carefully, that is not a rule against translating your own website. It targets translation used as a shortcut to mass-produce pages: other people’s content machine-translated onto new URLs, or thousands of thin pages pushed into Arabic to catch searches. A company translating its own service pages for its own customers is not in that category. But the test Google applies, whether a page gives its reader value, is exactly where raw machine output tends to fall short in Arabic. A page that is technically Arabic but awkward, inaccurate or unclear is a weak page, and a site full of weak pages is a weak site.
How Google handles two language versions of one site
Before quality comes mechanics. Google’s guide to multi-regional and multilingual sites (developers.google.com/search/docs/specialty/international/managing-multi-regional-sites) recommends a separate URL for each language version of a page, rather than switching the language with cookies or browser settings, and advises against adapting content by IP address. In practice that means /en and /ar paths, subdomains or separate domains, each with its own fully crawlable pages, not one URL that switches language depending on who is looking.
This is why a translation plugin or a browser-side widget that swaps the text after the page loads does little for search. The words are translated in the visitor’s browser. There is no Arabic URL for Google to crawl, index and show to someone searching in Arabic.
The language versions are then connected with hreflang annotations, placed in the page head, in HTTP headers or in the XML sitemap. Google’s rules (developers.google.com/search/docs/specialty/international/localized-versions) are strict: “Each language version must list itself as well as all other language versions,” and “If two pages don’t both point to each other, the tags will be ignored.” The reserved x-default value covers visitors whose language matches none of your versions.
Two details from the same documentation matter for machine translation. First, “Google doesn’t use hreflang or the HTML lang attribute to detect the language of a page”; it reads the visible content. A page tagged as Arabic whose body is half English, because the translation skipped the product descriptions, sends mixed signals. Second, “Localized versions of a page are only considered duplicates if the main content of the page remains untranslated.” Translating only the navigation and footer does not produce a real Arabic page. The Arabic visitor gets an English page in Arabic clothing.
For the layout side of the same problem, from mirrored navigation to Arabic fonts and form fields, see our guide to bilingual RTL web design.
Where machine translation is fine, and where it is not
Machine translation is not one level of risk. It depends on who reads the text, what they do with it, and what happens if a sentence is wrong. The table is how we sort content when a client asks which parts can start from a machine draft.
| Content | Machine draft? | What it needs before use |
|---|---|---|
| Internal documents, emails, research | Yes | A reader who knows it is a machine draft; no review while it stays internal |
| First drafts for a translator | Yes | Full human post-editing against your glossary |
| User reviews, comments, support tickets | Yes, to understand them | Show the original; offer the translation as an option labelled automatic |
| Blog articles and guides | Draft only | Native rewrite, Arabic keyword research, fact check |
| Home page, service pages, Arabic marketing copy | No | Written or transcreated by a native copywriter |
| Prices, offers, refund and delivery terms | No | Human translation, checked figure by figure against the source |
| Terms and conditions, privacy policy, contracts | No | A legal translator, and your lawyer to decide which version governs |
| Product claims in health, finance or safety | No | A specialist translator, plus whoever approves claims in the source language |
| Interface strings: buttons, menus, errors, forms | No, not out of context | Translation with screenshots or in the running build, then right-to-left QA |
The pattern is simple. The closer a text sits to money, obligations or the first impression a customer forms of you, the less a machine draft is worth on its own.
Why Arabic is where raw machine output breaks
Legal and commercial terms have exact equivalents that general-purpose tools often miss. “Deposit” may be a refundable security amount or a non-refundable down payment, and Arabic uses a different word for each, with different expectations attached. “Shall” in a contract is an obligation, not a future tense. “Warranty”, “guarantee” and “liability” have settled renderings in Gulf legal Arabic, and a model that picks a plausible synonym can change what you promised. Onshore courts in the UAE work in Arabic, so the Arabic text of your terms is not a courtesy copy: it may be the version a judge or a regulator reads. Our sister company’s daily work in legal translation is a constant reminder of how much weight one term can carry.
Prices and numbers fail more quietly. A tool may turn “from AED 1,500” into a sentence that reads as a fixed price, drop “excl. VAT”, render “per month” as “per year” inside a long sentence, or mix Arabic-Indic and Western digits on the same page. Each slip is small. Each one is also a written promise to a customer, and in a dispute the customer will quote the Arabic.
Product claims drift away from their qualifiers. “Up to twice as fast” becomes “twice as fast”; “helps reduce” becomes “eliminates”. In regulated categories such as health products, financial services or education, a claim that was carefully qualified in English can become a claim you are not allowed to make once the qualifier disappears in Arabic.
Arabic marketing copy is where machine output is most visible. Literal translation produces grammatical sentences that no Emirati or Gulf copywriter would write: English word order, idioms carried over word for word, an unsure mix of formal and colloquial register, and the wrong form of address. Gender is a real problem too. English “you” has no gender, Arabic verbs and adjectives do, so a tool picks one, usually the masculine, and sometimes switches halfway down the page. Readers notice in the first line, and the impression is not “international company” but “they did not think we were worth a translator”. That gap is the difference between translation and localization, which we cover in translation vs localization in the UAE.
Interface strings are the hardest case, because they are short and arrive without context. “Save” can mean حفظ, saving a file, or توفير, saving money. “Book” is a reservation or a volume, and “Order” is a purchase or a sort order. A tool translating a strings file sees one word at a time and guesses. Then the right-to-left layout adds failures of its own: punctuation and phone numbers jumping to the wrong end of the line, English brand names breaking the text direction, labels longer than their buttons, icons and arrows pointing the wrong way. None of this can be checked in a spreadsheet. It only shows in the running site.
The search cost: a translated keyword is not an Arabic search
Even a perfect translation can rank for nothing. Arabic speakers in the UAE do not search with the translation of your English keyword. They use different terms: formal and colloquial variants, English loanwords written in Arabic script, and sometimes English itself. A machine renders your English heading faithfully, and the term people actually type never appears on the page. The method for finding those terms is in our guide to Arabic keyword research in the UAE.
Metadata suffers in the same way: titles that run too long once translated, meta descriptions cut off mid-word, image alt text left in English, and URL slugs that are either English or a transliteration nobody searches for. AI answer engines, which now summarise pages for people who ask their questions in Arabic, quote clear and accurate statements. A muddled machine translation gives them little worth quoting, as we explain in AI search visibility in the UAE.
A workflow that uses AI without publishing its mistakes
The answer is not to refuse AI. It is to decide where it sits in the process, and to put people at the points where errors are expensive. This is the workflow we recommend to any team starting from a machine draft.
1. Build a terminology glossary first. Before a single page is translated, list your brand name and how it is written in Arabic, your product and service names, your legal and pricing terms, and the words you never want used. Decide the register (Modern Standard Arabic for most business sites, with a light Gulf tone where it fits) and the form of address. Every later step checks against this list.
2. Sort the content by risk. Use the table above. Low-risk content can start from a machine draft. Legal text, prices, claims, the home page and the service pages go to human translators or native copywriters from the start.
3. Machine draft, then full human post-editing. A post-editor is a professional translator who compares the draft with the source sentence by sentence and fixes meaning first, style second. Light post-editing, which fixes only obvious errors, is acceptable for internal text. Anything public needs full post-editing.
4. Native review of the finished Arabic. A second native speaker reads the Arabic on its own, without the English, as a customer would. This is where unnatural phrasing and register problems surface, because the reader is no longer anchored to the source.
5. In-context QA on the real build. Review the Arabic pages on the staging site, on desktop and on mobile: line breaks, truncated buttons, mixed-direction text, numbers, forms, error messages and transactional emails. Fix strings in the source files, not in screenshots.
6. Legal sign-off where it matters. Terms, the privacy policy and anything contractual are checked by a legal translator, and your lawyer decides which language version governs.
7. Keep it maintained. Every change to the English goes through the same chain. Most bad Arabic sites degraded because later updates were machine-translated straight to production.
What it costs to fix a bad Arabic site
Fixing an Arabic site that went live as raw machine output usually costs more than doing it properly the first time, because the work doubles: someone has to audit what exists before anyone can rewrite it. A typical repair includes a content audit against the English, a terminology glossary built after the fact, retranslation or native rewriting of the key pages, a check of every price, term and claim, right-to-left fixes in templates that were never tested in Arabic, and new metadata and keyword targeting.
There are search costs as well. If the Arabic URLs change, every old URL needs a 301 redirect to its new equivalent, and the hreflang pairs have to be updated on both sides. Google then has to recrawl and reassess the pages, which does not happen overnight. The cost that never shows on an invoice is trust: customers who met the broken version have already formed their impression.
For reference, our multilingual content service prices human localization from AED 0.35 per word and complete bilingual website copy from AED 1,500, both excluding VAT. Neither involves machine translation: our linguists write or translate the text themselves, and the final price depends on volume and subject. If you already have an Arabic site and are not sure how bad it is, a short review of the key pages will show whether it needs a repair or a rewrite. Send us the URL.
Checklist before you publish the Arabic version
Run through these before the Arabic pages go live, and again after any large update.
Each language has its own URL, crawlable without cookies, scripts or IP-based redirects.
Hreflang pairs are reciprocal, every page lists itself, and an x-default is set.
The main content is translated, not only the menus and footer.
Prices, VAT wording, refund, delivery and warranty terms match the source, figure by figure.
Terms and the privacy policy were translated by a legal translator, and your lawyer has decided which version governs.
Every page has an Arabic title, meta description, headings and image alt text, written around real Arabic search terms.
A native speaker has read the Arabic without looking at the English.
Buttons, forms, error messages and emails were checked in the live right-to-left layout, on mobile.
There is a process for updates, so the next change is not machine-translated straight to production.
Sources
- Google Search Central: Spam policies for Google web search (scaled content abuse)
- Google Search Central: Tell Google about localized versions of your page (hreflang)
- Google Search Central: Managing multi-regional and multilingual sites
- Google Search Central Blog: Google Search’s guidance about AI-generated content
FAQ
Will Google penalise my website for using machine translation?
Not for the act of translating. Google’s spam policies target scaled content abuse, which includes automated translating used to mass-produce pages that add little value. A reviewed translation of your own pages is fine. Raw machine output is judged like any other page: on whether it actually helps the reader.
Is Google Translate or ChatGPT good enough for my Arabic website?
For a first draft, often yes. For publishing, not on its own. Prices, terms, product claims, marketing copy and interface strings need full human post-editing, a native review and testing in the right-to-left layout before they go live.
Can I use a translation plugin that switches language on the same URL?
It helps visitors read the page but does little for search. Google recommends a different URL for each language version, connected with hreflang, rather than cookies or browser settings. A widget that translates in the browser leaves no Arabic URL for Google to index.
How much does human localization of a website cost?
Our human localization starts from AED 0.35 per word, and complete bilingual website copy from AED 1,500, both excluding VAT. The final price depends on the volume and the subject, and no machine translation is involved.