Machine vs human translation for websites: where each fits
Machine translation suits drafts and support content; pages that sell need a human. Costs, Google's position, and the Arabic errors machines still make.
By Sohail Hussain21 November 20266 min read
Use machine translation to produce drafts, to translate internal and support content, and to handle user-generated text. Use a professional human translator, or machine output edited by one, for every page that sells, makes a promise or carries legal weight. For Arabic the line sits further toward human work than for most European languages, because gender, register and regional vocabulary are exactly where machine output goes wrong, and Arabic readers notice within a sentence.
This matters beyond quality. A machine-translated homepage tells a Gulf buyer the company did not think the Arabic audience was worth paying for. Translation is part of every Arabic and multilingual website we build, and we arrange professional translators or work with the client's own. The wider build is covered in the guide to building an Arabic website.
What are the options, and what does each cost?
There are four levels of translation, not two.
| Option | What it is | Cost, as of October 2026 | Fits |
|---|---|---|---|
| Raw machine translation | DeepL, Google Translate or an AI model, published as is | API fees of cents to a few dollars for a whole company site | Internal use, getting the gist |
| Machine translation with post-editing | A translator corrects the machine draft | Usually quoted below the full translation rate | Support articles, product catalogues, long-tail pages |
| Human translation | A professional translator, checked by a second reader | AED 0.30 to 0.80 a word for Arabic | Service pages, product pages, legal text |
| Copywriting in Arabic | Written in Arabic from a brief, not translated | SAR 300 to 1,000 a page in Saudi Arabia | Homepage, campaigns, slogans, Arabic-first sites |
The translation and copywriting rates are the ones in our guides to website costs in the UAE and in Saudi Arabia. Raw machine translation is close to free. Google's Cloud Translation pricing charges per million characters, and a 20-page company site is roughly 100,000 characters. The cost that matters is not the API bill; it is what an unreviewed page costs in trust.
Post-editing rates vary more than full translation rates, because the effort depends on how good the draft is. Ask a translator to quote both ways on a sample page. To compare the per-task cost of machine and human work at your volume, try our AI vs human cost comparison, and count the words on your key pages with the word counter before you ask for quotes.
What does Google say about machine-translated pages?
Google does not ban machine translation. Its concern is low-value pages produced at scale. The spam policies list scaled content abuse, and the examples include generating many pages from other content "through automated transformations like synonymizing, translating, or other obfuscation techniques" where little value is provided to users. For content like that, Google's instruction is to exclude it from Search.
In practice the line is clear enough:
- Your own service pages, translated and reviewed for real Arabic customers: fine, however the first draft was produced.
- Thousands of auto-translated pages published to catch search traffic in another language: the pattern the policy describes.
The safe working rule is to keep unreviewed machine translations out of the index. Publish them with noindex, or not at all, until a person has checked them.
Where does machine translation go wrong in Arabic?
The errors are rarely nonsense. They are plausible sentences that are subtly wrong, which is worse, because a non-Arabic reader signing off the page cannot see them.
Gender. Arabic verbs, imperatives and "you" change with the reader's gender. "Register" is سجّل addressed to a man and سجّلي addressed to a woman. Machine output defaults to the masculine form, sometimes switching mid-page. A translator decides deliberately: masculine by convention, plural, or a neutral construction such as a verbal noun (التسجيل, "registration") on buttons.
Register and dialect. Business sites are written in Modern Standard Arabic. Machine output is mostly standard Arabic too, but marketing lines can come out stiff, or occasionally drift into dialect phrasing that reads as out of place.
Terminology. Words with several English meanings get the wrong one. "Quote" as in a price can come back as اقتباس, which means a quotation from a text, instead of عرض سعر, a price quote. "Lead", "booking", "plan" and "support" all have traps of the same kind.
Names and fixed terms. Brand names, product names and job titles get translated when they should stay in English, or transliterated differently on every page.
Punctuation and numbers. Latin commas and question marks where Arabic uses ، and ؟, mixed digit styles, and dates in the wrong order. Numbers, dates and currency in Arabic interfaces covers the formatting rules.
Length. Translated strings that are longer or shorter than the English break buttons and headings. Neither the machine nor a translator working from a spreadsheet can see this.
AI models such as ChatGPT, Claude and Gemini handle some of these better than older engines, because you can give them a glossary and instructions on tone and form of address. They introduce a different risk: fluent output that quietly drops or rewrites a sentence. The review step stays.
When is machine translation good enough?
Use it without hesitation for:
- internal documents, emails from suppliers and anything you only need to understand;
- first drafts handed to a translator for post-editing;
- user reviews and comments, labelled as translated, with the original one click away;
- a large support knowledge base, where speed matters more than polish, provided the most-viewed articles are post-edited and readers can report a bad translation.
Do not rely on it for:
- the homepage, service pages, pricing and contact pages;
- legal pages such as terms and privacy policies, which need a translator and then your lawyer;
- regulated claims in health, finance or education;
- anything aimed at government buyers, or a Saudi-first site where Arabic is the main language.
How do you set up a translation workflow that holds up?
The quality of a translated site depends more on the process than on the translator.
- Write a glossary. Brand terms, product names, which words stay in English, and the agreed Arabic for your key services.
- Write a short style guide. Modern Standard Arabic or not, the form of address, which digits to use, and the tone.
- Give context. Send the translator the staging site, not a spreadsheet of disconnected strings.
- Have a second reader. Ideally a native speaker in the client's own team, who knows the business.
- Review on the page. Check the Arabic in the real layout, where truncation and line breaks show.
- Keep a translation memory. Most professional tools store approved sentences, so later updates are charged only for what changed.
- Tie Arabic to every English change. A new English page creates a translation task before it is published.
If the site is on WordPress, the translation plugin shapes this workflow; WPML vs Polylang vs TranslatePress compares how each handles it.
Was this guide useful?