Arabic vs English AI prompts: what changes
We compared how ChatGPT, Gemini and AI Overviews answer the same buying question in Arabic and in English, and what the gap means for your content.
Asking an AI assistant the same buying question in Arabic and in English frequently returns two different sets of businesses, because the assistant answers from the sources that exist in that language rather than translating one answer into the other, and most businesses have far more sources in one language than the other.
This comparison is the language dimension of the wider question covered in how Saudis use AI search to find businesses; here we go deep on exactly what differs and why.
On this page
Why would the same question return different answers in different languages?
Because an assistant is not translating an answer from one language into another; it is separately finding, reading and citing sources written in the language it was asked in. If your business has a strong English page and a thin or missing Arabic one, the English question surfaces you and the Arabic question does not, even though the underlying business is identical.
This matters more in Saudi Arabia than in a fully monolingual market because so many buyers move between the two languages depending on context — English at work, Arabic at home, or a mix within the same conversation. DataReportal's connectivity numbers show a population that is almost entirely online, which means both language pools are large enough to matter commercially, not a rounding error you can safely ignore [2].
It also matters because the two language communities are not simply the same people switching a setting. A professional audience raised partly on English-language education, international brands and English-language workplaces reads and searches English content fluently and often by default. A broader, larger audience is far more comfortable in Arabic, and for some buying categories — family healthcare, religious travel, government-adjacent services — the Arabic-first share of the audience is decisive. Treating "our English site is strong" as evidence that the business is visible to AI search overall skips over which half of the market that strength actually covers.
Should you write in Modern Standard Arabic or a local dialect?
Modern Standard Arabic, for anything you want an assistant to cite. Spoken Gulf dialect is how people talk to each other on social media and in reviews, and it has its own value there, but the overwhelming majority of published, indexed, citable content — news, documentation, official pages — is written in Modern Standard Arabic. That is simply where the depth of source material sits, and it is the register an assistant has the most training and retrieval material in.
This does not mean your Arabic pages should sound stiff or bureaucratic. Plain, direct Modern Standard Arabic, written the way a competent local copywriter would write it for a real reader, reads naturally and performs well. The line to avoid is at the other extreme too: a heavily formal, legalistic register that no one actually enjoys reading is also a weak source, for the same reason overly formal English marketing copy underperforms plain English.
What did we actually compare, and how?
We ran a fixed set of realistic buying questions — the kind a real customer types, not a keyword — twice each: once in Arabic, once in English, across the same assistants on the same day. Questions covered clinics, real estate, hotels and online stores, the categories where AI recommendations most often influence a purchase.
| English version | Arabic version (transliterated sense) | What it tests |
|---|---|---|
| Best paediatric dental clinic in north Riyadh that takes Bupa | Best clinic for children's teeth in north Riyadh with Bupa | Whether insurance and location facts transfer |
| Is this online store legit and does it ship fast? | Is this store trustworthy and how fast is delivery? | Whether trust signals transfer |
| Family hotel near the Boulevard with a pool | Family hotel near the Boulevard with a swimming pool | Whether amenity facts transfer |
The differences in wording are intentional; a native speaker rarely asks the literal translation of an English sentence, and testing with natural Arabic phrasing is part of what makes this a fair comparison rather than a translation exercise.
Where did the two languages agree, and where did they diverge?
They agreed most often on well-known national brands — large retailers, banks, telecoms — that are covered heavily by news and reference sources in both languages. Coverage that thick survives a language switch because there is enough material in both pools to draw from.
They diverged most on small and mid-sized local businesses, exactly the category most of our clients operate in. A clinic or a boutique store with a strong English website and social presence but a weak or missing Arabic footprint tended to appear in the English answers and disappear in the Arabic ones, and vice versa for Arabic-first businesses with no real English content.
The hotel and real estate questions produced a third pattern worth naming separately: both languages named the same handful of large listing portals as sources far more than they named individual businesses directly. In categories dominated by aggregators, the language gap shows up less in "which business gets named" and more in "which portal's Arabic or English listing for that business is more complete" — a subtler problem, and one that points at your listing pages rather than your own site as the thing to fix first.
| Category | Gap size | Likely cause |
|---|---|---|
| National retail and telecom brands | Small | Heavy news and reference coverage exists in both languages |
| Clinics and local services | Large | Content quality depends entirely on the business's own bilingual effort |
| Hotels and real estate | Medium | Portals carry much of the weight; the gap sits inside portal listings |
| Online stores | Large | Product and policy pages are rarely translated properly, if at all |
99.0 percent online penetration
Does a literal or machine translation close the gap?
Partly, and unevenly. A machine-translated page carries the same facts as the original, so an assistant reading it does gain some information it did not have before. But two things limit how much it helps.
First, mechanical translations frequently produce phrasing a native speaker would not use, using formal or awkward constructions for what should be a plain, common phrase. Second, and more speculatively, our own comparisons suggest machine-translated pages get treated as a weaker source than pages written directly in the target language — consistent with the general pattern that assistants favour clear, fluent, well-structured content, though we cannot point to a controlled study isolating translation quality as the cause.
There is also a quieter, structural reason machine translation underperforms: it tends to translate a page written for an English search habit, keyword phrasing and all, rather than starting from the question an Arabic speaker actually types. A page optimised for "best dental clinic Riyadh" translated mechanically into Arabic still carries the shape of an English keyword page. A page written from scratch for the Arabic question a parent actually asks reads, and performs, differently — the words are similar but the structure a real question needs is not the structure a translated keyword page has.
Start with your highest-value pages
Home, service and location pages first — the pages a buying decision actually depends on — not your entire blog archive at once.
Write for the question, not the English sentence
Have a native Arabic speaker write the page from the same facts, phrased the way an Arabic speaker actually asks the question, rather than translating sentence by sentence.
Keep the facts identical across languages
Same name, address, phone number, prices and services in both versions. A mismatch reads as two different businesses to an assistant piecing together facts from listings.
Mark up your entity consistently
Organization structured data that names your business the same way in both language versions of the page helps tie the two together as one entity rather than two [4].
Does this apply the same way to ChatGPT, Gemini and Google's AI Overviews?
Broadly yes, though the mechanics differ slightly. ChatGPT can search the live web through a dedicated crawler, which OpenAI documents separately from its training crawler, so a page that is reachable and current in a given language can be found and cited in that language regardless of when the underlying model was trained [1]. Gemini and Google's AI Overviews draw more directly on Google's own index and Google Business Profile data, so your Arabic listing completeness matters at least as much as your Arabic page content.
We cover the AI Overviews side of this specifically, including how Arabic search results behave on Google's own feature, in AI Overviews in Arabic. And the worked clinic example in what ChatGPT recommends for Riyadh clinics shows this exact language gap playing out on one real category, using a hypothetical business built from the pattern we see.
What should a bilingual business actually do with this?
Publish real, complete pages in both languages for anything a buyer would compare before choosing you, keep every fact identical across languages, and measure both languages separately rather than assuming a win in one means a win in the other.
Budget for this as one workstream with two outputs, not two separate projects. The research — what does a buyer actually ask, what facts does the answer need, what does a competitor's page look like — is largely shared across both languages. The writing is where the real extra cost sits, and it scales with how many pages actually influence a purchase decision, not with your total page count. Most businesses find the list of pages that truly matter for this is shorter than they expect: home, three or four service or category pages, and your location or contact page usually cover most of what an assistant would ever need to cite.
If you sell online, the language gap shows up in a slightly different form — see getting a Salla store into ChatGPT for how product and store pages need to handle this. And because this whole comparison depends on there being enough Arabic-speaking, AI-using buyers to matter commercially in the first place, the underlying numbers are in AI adoption in Saudi Arabia explained — worth reading if you are still deciding whether the Arabic side of this is worth the investment at all.
The honest summary: this is not a trick or a hack. It is the same content-quality work you would do for English, done a second time, in a language a real share of your buyers actually use.
Questions people also ask
Should a Saudi business write content in Arabic, English or both?
Both, if buyers ask in both. Most bilingual markets have some buyers who default to one language and some who switch by context, and an assistant answering in a language you have not published in cannot cite you in that language.
Is Modern Standard Arabic or a spoken dialect better for AI prompts?
Write your pages in clear Modern Standard Arabic; it is what most published sources use, so it is what assistants have the most material to draw on. Dialect matters more for social content than for the pages an assistant would cite.
Does translating my English page word for word fix the gap?
Rarely on its own. A literal or machine translation often reads stiffly and can miss terms a native Arabic searcher would actually use, which weakens both how people respond to it and, it appears, how assistants treat it as a source.
How do I know if my business has an Arabic AI-visibility gap?
Ask the same buying question in both languages and compare who gets named. A consistent gap in one direction is your answer; a single run is not enough because answers vary day to day.
Do I need separate Arabic and English domains or just separate pages?
Either can work technically. What matters for AI visibility is that each language has real, complete, indexable content and that your business name, address and phone number match exactly across both.
Sources
- Overview of OpenAI crawlers — OpenAI (checked 23 September 2026)
- Digital 2026: Saudi Arabia — DataReportal (checked 23 September 2026)
- AI features and your website — Google Search Central (checked 23 September 2026)
- Organization structured data — Google Search Central (checked 23 September 2026)
See your Arabic vs English visibility gap
Book a free 30-minute consultation with Ibrahem and Jeremy. We look at your site while we talk, and you leave with a prioritised plan.
Who wrote this
Jeremy ZavaCo-founder
Leads growth and client partnerships.
Works on: CRO, SEO
Reviewed before publishing by Ibrahem Helal, Co-founder.
- How Saudis use AI search to find businessesHow people in Saudi Arabia use ChatGPT, Gemini and AI Overviews to pick businesses, and what a Saudi company must fix to be the name they get back.
- AI Overviews in Arabic search resultsHow Google's AI Overviews behave on Arabic search results in Saudi Arabia, where they differ from English ones, and what it means for your Arabic pages.
- What ChatGPT recommends for Riyadh clinicsA worked, anonymised look at which kind of clinic ChatGPT names when asked for a recommendation in Riyadh, and the traits the named clinics share.
- Getting a Salla store into ChatGPTA step-by-step guide to making a Salla storefront readable and citable by ChatGPT and other AI assistants, from crawler access to product data.