Arabic SEO keyword research fails for a specific reason: teams take an English keyword list, translate it, and treat the output as research. What comes back is a set of grammatically correct phrases that nobody types. Real Arabic research starts in Arabic, deals with the fact that the same query has several legitimate spellings, and separates Modern Standard Arabic from the dialect and Latin-letter forms people actually use on a phone in Deira.
Key takeaways
- Translate the intent, not the keyword. A translated English phrase is a guess with better grammar.
- Several spellings of an Arabic query are the same query once normalised. Deduplicate before you count volumes or you will double count demand.
- Google works out a page’s language from the visible content, and explicitly ignores the
langattribute and the URL, so a half-translated page is an English page with Arabic decoration. - Keyword Planner hides low volume terms by its own admission, which is most of the Arabic long tail for a single emirate. Search Console is the only honest source of your own Arabic queries.
- Keep hreflang out of the research phase. It is a delivery problem, not a discovery one.
Why a translated keyword list is not an Arabic keyword list
Take “web design company Dubai”. Translate it and you get a phrase that reads properly and carries the right meaning. Now consider what somebody in the UAE actually types. They might search for a company, for the service, for a price, or for a nearby area, and the word order and the level of formality shift with each one. The translated phrase sits somewhere in the middle of that and matches none of it exactly.
The working method is to translate the intent into Arabic and then let Arabic speakers and Arabic data generate the phrasings. Start from what the searcher wants, not from the English string. “Someone comparing prices before contacting an agency” is an intent. It produces different Arabic phrasing from “someone who already decided and wants a company in their emirate”, and those two need different pages.
From a practical standpoint, the first hour of Arabic research should be spent in Google itself, typing seed terms into the Arabic interface and reading the autocomplete suggestions, the related searches and the People Also Ask box. That is real query data generated by real behaviour, and it costs nothing. Write down what appears, in the spelling it appears in.
The spelling problem, and the rules underneath it
This is the part that quietly corrupts Arabic keyword spreadsheets. Arabic has several character pairs that writers use interchangeably, so one query reaches your list as three or four rows.
The clearest published specification of what search systems do about this is Apache Lucene’s Arabic normaliser, documented in the Lucene 9.12.1 API. It applies five transformations.
- Alef with madda, alef with hamza above and alef with hamza below all become bare alef, so أ, إ and آ collapse to ا.
- Teh marbuta becomes heh, so شركة and شركه become the same token.
- Dotless yeh becomes yeh, so ى and ي stop being different words.
- The short vowel marks are removed: fathatan, dammatan, kasratan, fatha, damma, kasra, shadda and sukun.
- Tatweel, the character used to stretch a word for typographic effect, is stripped entirely.
Lucene is not Google, and Google does not publish its Arabic tokenisation. Treat these rules as a sound working model rather than as Google’s implementation. The practical instruction holds either way: before you total up volumes, normalise your keyword column with these five rules and merge the rows that collapse together. Expect the list to shrink and the apparent total volume to fall with it. That drop is not a loss, it is the removal of a number that was never real.
The same problem returns at the writing stage. If your page uses one spelling and half your audience types another, normalisation is doing the matching rather than your copy. Pick the spelling that dominates in your own Search Console data and use it consistently, and do not sprinkle variants through the text hoping to catch both. That reads badly to a human, which is the audience that converts.
Standard Arabic, dialect, and Latin letters
Three registers compete in the same search box, and they do not behave the same way.
Modern Standard Arabic is the written register. It dominates formal, informational and research queries, and it is what most published Arabic content is written in, which makes it the safer default for guides, comparisons and service pages.
Gulf dialect appears in conversational and local queries, and it varies by country. A phrase that is natural in Emirati speech may be unusual in Egyptian or Levantine Arabic, and the UAE audience includes large numbers of speakers of both. This matters most for a business serving residents rather than the region, because the dialect mix of your customer base is not the dialect mix of the country.
Then there is Arabizi, Arabic written in Latin letters with digits standing in for sounds that Latin script lacks. It is common in messaging and it does appear in search. Whether it appears in your category is an empirical question rather than a matter of opinion, and Search Console will tell you within a month of having any Arabic visibility at all.
The honest position on register is that you should not decide it from an article, including this one. Pull your own query data, sort by language and script, and let the split you actually have decide where the content budget goes. If you have no Arabic visibility yet, start with Modern Standard Arabic for the substantial pages, because it is the widest net and the easiest to write well.
Where Arabic search volume actually comes from
Four sources, each with a different weakness, and you need at least three of them.
Google Keyword Planner gives you volumes by location and language, and it is the only free source of absolute numbers. Google’s own help on finding new keywords notes that keywords with very low search volumes are not discoverable or forecastable, and that forecasts draw on the last seven to ten days with seasonality applied. For Arabic terms scoped to a single emirate, a great deal of genuine demand sits under that floor and simply will not appear. Widening the location to the UAE or the GCC to make terms visible, then narrowing back for the actual campaign, is a reasonable workaround as long as you remember that you widened it.
Google Trends gives relative interest rather than volume, and it is the right tool for two comparisons: Arabic against English for the same concept, and one Arabic phrasing against another. It will not tell you how many searches there are, and that is fine, because at this stage you are choosing between options rather than forecasting traffic.
Search Console is the only source that tells you the truth about your own site. It shows the queries that actually brought impressions, in the spelling they were typed. It is also the only place Arabizi will surface honestly. The caveat is that it reports what you already rank for, so it cannot find demand you have never touched.
Autocomplete and related searches close the gap. They generate phrasings, and they generate them in the register people use. They give you no numbers at all, which is why they pair with Keyword Planner rather than replacing it.
Paid tools vary considerably in Arabic coverage, and that variation is not always visible in the interface. Before trusting one, take twenty Arabic terms you know from Search Console carry real impressions and check whether the tool reports volumes for them. A tool that returns zero for terms you can prove exist is telling you about its index, not about the market.
Google decides your page is Arabic by reading it
This is the technical fact that undermines most half-finished Arabic sites. Google’s guidance on managing multi-regional and multilingual sites, last updated on 10 December 2025, states that Google uses the visible content of a page to determine its language, and that it does not use code-level language information such as lang attributes or the URL.
So an Arabic lang="ar" on a page whose body copy is English does nothing. A page with an Arabic navigation and an English article is a mixed page, and the same guidance asks for a single language for content and navigation on each page and advises against side-by-side translations. Google also recommends different URLs for each language version rather than switching content with cookies or browser settings, and warns against automatically redirecting users from one language version to another, suggesting visible links between versions instead.
On URL structure, the documentation sets out the options plainly: a country-code domain gives the clearest geotargeting but is expensive and covers one country; a subdomain is easy to set up but the geotargeting intent is less obvious to users; a subdirectory is simple and low maintenance with weaker geotargeting signals; and URL parameters are not recommended because they are hard to segment. For a UAE business running Arabic and English, a subdirectory is usually the right trade-off.
The hreflang annotations that connect those versions are a separate job with its own rules, including the requirement that if page X links to page Y, page Y must link back to page X, and the use of ISO 639-1 language codes with optional ISO 3166-1 Alpha 2 region codes. We covered the implementation and the common failures in our guide to hreflang for Arabic and English UAE websites. Keep it out of the research phase; deciding what to write comes first.
A research workflow you can run this week
- Write your six to ten intents in English, as sentences describing a person and what they want. No keywords yet.
- Have an Arabic speaker who knows the business express each intent as three or four natural search phrases. Not a translator working from a word list, someone who understands what you sell.
- Feed each phrase into Google’s Arabic interface and harvest autocomplete, related searches and People Also Ask. Record the exact spellings returned.
- Normalise the whole list with the five rules above and merge the collapsed rows.
- Pull volumes in Keyword Planner at UAE level, then at emirate level, and keep both numbers. The gap between them is your visibility floor, not a measurement.
- Compare shortlisted phrasings in Google Trends to choose between near-synonyms.
- Check your Search Console for any Arabic queries you already receive and let them override everything above, because they are observed rather than estimated.
- Map the surviving terms to pages, one primary intent per page, and only then decide the URL structure and the hreflang plan.
Step two is the one teams skip, and it is the one that decides whether the rest is worth doing. A briefing session with somebody who knows both the language and the customer produces better seed terms than any tool.
Reporting without double counting
Rank trackers treat spelling variants as separate keywords, because they are separate strings. A dashboard can therefore report more Arabic keywords than you are really tracking queries, and can show a ranking improvement that is one page moving once and being counted several times.
Two habits fix it. Keep a normalised key column alongside the raw keyword in your tracking sheet, and report on the normalised key. And when you report Arabic and English performance together, report them as separate lines rather than a blended total, because they typically have different intent profiles and blending hides which one is working. Our comparison of Arabic and English digital marketing for GCC audiences goes further into that split on the campaign side.
If you are planning Arabic pages for a UAE site, the sequence that saves money is intent first, native phrasing second, normalisation third and tooling last. Our team handles SEO for UAE businesses and the content work behind it, in Arabic and English, and we would rather look at your Search Console data than guess at your market.
Documentation dates and specifications above were checked against primary sources on 18 September 2026. Search engine behaviour and tool coverage change, so re-check the current documentation before acting on any of it. Cover photo: Museum of the Future 8, via Wikimedia Commons (CC0).
