The Dubai Ad Campaign We Planned in English and Had to Rebuild in Three Languages

TLDR: A Dubai search campaign built entirely in English was quietly serving three different buyer populations: people searching in English, in Arabic and in French, each with its own intent, price sensitivity and days of the week. A market analysis by query language, not by keyword volume, showed the mismatch, and the campaign was rebuilt as three campaigns with their own landing pages, schedules and bids. The method works for any market where more than one language is typed into the same search box.

The first search terms report I pulled for that account was a mess I recognised immediately, because I had made it myself. The campaign was a dental clinic group in Dubai, six months live, run in English by a team that spoke English, with English ads pointing at an English landing page. The report showed rows in Arabic script, rows in French with the accents stripped, and rows in English that read like a translation. All three were being answered by the same ad.

I was sitting in the Tangier office late on a Thursday, laptop warm, the call to prayer coming through the window, and I remember thinking that the campaign had been treated as one market because the budget line was one number. The market did not know that.

The Problem With Planning a Campaign in the Team's Language

Most campaigns in Dubai are planned in English for a simple reason: the people planning them work in English. The keyword research is done in English, the ad copy is written in English, and the landing page is signed off in English. Everyone in the room reads it, so it feels complete.

Dubai does not work like that on the search side. A large share of residents search in Arabic. French-speaking expats from Morocco, Lebanon, Algeria and France search in French, and a fair number switch between French and English within the same week. Google's own language targeting does not fix this on its own, because it matches on any language Google believes the user understands, and the query language is only one of the signals it weighs alongside the person's settings and history. Google's own help page gives the example of a user typing a query in Spanish and being served an ad in English. A campaign set to English can still serve an English ad to someone whose actual question was written in Arabic, and Google's September 2026 change, which drops the campaign-level language setting for Search and matches on the language of the ad itself, moves the problem rather than removing it. That is precisely what our report was showing.

The result was a campaign that looked healthy on the surface and behaved like three campaigns underneath, each with a different cost per lead, each dragging the others' numbers around.

The Analysis: Split by Language First, Then by Everything Else

The method I used is not sophisticated, and that is why I still use it. Most of the elapsed time on that account was waiting for enough data to fill each bucket.

Step one is to export every search term from the last 90 days and tag each row by language. Arabic is trivial because of the script. French and English need a short list of stop words and a manual pass on the ambiguous rows, which took an afternoon. Step two is to break each language bucket down by the dimensions the platform already gives you: day of week, hour, device, average cost per click, and the conversion action the click led to. Step three is to read the three buckets as if they were three different clients.

That is where it got interesting. The three groups were not using different words for the same thing. They wanted different things.

• English queries were dominated by price comparison and treatment names. They clustered on weekday lunch hours and on desktop, and the click was expensive: the most competitive terms cost well over $5 (AED 18) per click, because every clinic in the city was bidding on the same English words.
• Arabic queries were longer, more descriptive, and often framed as a problem rather than a treatment. They peaked on Sunday and Monday evenings on mobile. Clicks cost a fraction of the English ones because far fewer clinics were bidding in Arabic at all.
• French queries were the smallest bucket but converted at the best rate of the three. They came on Saturday mornings, cost a little less per click than the English terms, and asked, more than the other two, about the clinician, the language spoken at reception and whether the practice worked with a particular insurer.

One campaign was averaging all of that into a single bid, a single ad schedule and a single page that spoke to the first group and confused the other two.

What the Rebuild Looked Like

We did not touch the budget for a month. We rebuilt the structure and let the same money flow through three channels instead of one.

Each language got its own campaign, its own ad schedule and its own bid strategy. The Arabic campaign had its schedule opened on Sunday and Monday evenings and bids raised there, while the English campaign lost its weekend budget entirely. The French campaign got a lower daily cap and a much higher bid on the handful of queries that mattered, because those clicks were rare and worth waiting for.

The landing pages were the bigger job. A translated page is not a localised page. The Arabic page led with the problem the searcher had typed, not the treatment name, and the French page put the clinician's languages and the insurers accepted above the fold, because that was the question French speakers had been asking. Google's guidance on managing multilingual and multi-regional sites makes the point for organic search, and it applies just as well to paid landing pages: one language per page, clearly signalled, with the content actually written for that reader.

The team we run at a digital marketing agency like ours is spread across Tangier, Dubai and California, so Arabic and French copy came from native speakers on staff rather than from a translation tool. That mattered more than it sounds, because it is the difference between a page that reads as a clinic and a page that reads as a machine.

What Changed, and What Did Not

I will be careful with the numbers here because they belong to the client. What I can say is that the blended cost per lead fell within the first six weeks, and that the fall came almost entirely from the Arabic and French campaigns, which had been starved under the old structure. The English campaign barely moved. It had been competing on the most expensive words in the city and still was; the difference was that it stopped carrying the other two.

The other change was in the conversations with the clinic. Before, we reported one campaign with one cost per lead, and the owner asked why it was going up. After, we reported three, and the owner started asking which reception staff should work Sunday evenings. That is the kind of question a market analysis is supposed to produce.

Split the Report by Language Before You Split the Budget

If you run search campaigns in any multilingual market, in the Gulf or anywhere else, the method is the same and it is cheap. Pull 90 days of search terms, tag by language, break each language down by day, hour, device and cost, and read each bucket as its own market. Do the keyword research per language rather than translating a list; Ahrefs has a solid keyword research guide whose first lesson is that the words people type are not the words you would have chosen for them.

Then build one campaign per language, with a page written by a person who lives in that language. Give it a month with the same money before you judge it. The mistake this prevents can run for years without anyone noticing, because a single blended number never looks wrong.