Health tourism SEO: building a 4-language demand map
A method for building the demand map in TR, EN, AR and FA from your own data: seed sets, intent layers, verification sources and their limits.
By Roozbeh Nazari · CEO
The most expensive mistake a multilingual clinic site can make is translating the Turkish keyword list and building the same set of pages in four languages. What changes between languages is not the words but the patient’s search behaviour: the decision stage, the question asked, the source trusted and the terminology used all differ. A demand map exists to turn that difference into page architecture.
This article explains how to build a demand map; it does not hand you one. The method is written so that it can be repeated with the clinic’s own data. For the architecture side, our multilingual SEO architecture playbook for clinics is the companion read.
Why there are no volume figures in this article
The thing that circulates most in health tourism content is tables of search volume with no traceable source. For Arabic and Persian queries, reliable, publicly available and verifiable volume data is close to nonexistent in practice, and third-party tools’ estimates for those languages diverge considerably from one another. Publishing a figure we cannot verify gives the reader a false sense of precision instead of helping them.
For that reason no monthly search volume or market share figure for any language appears below. What is given instead is the procedure that lets a clinic produce those numbers from its own accounts, together with a plain statement at each step of how reliable that data is. A rough map you built from your own data is more useful than a precise table whose source is unknown.
Step 1: a seed set in four languages
The seed set is built from inside each language, not by translation. In practice three sources are used together: the phrasings patient coordinators see in incoming messages, the search box’s autocomplete suggestions, and the queries already appearing for that language in the clinic’s own search performance report. The third is the most valuable, because it is recorded impressions rather than an estimate.
Three typical divergences show up between languages. The first is term preference: the ratio between the clinical name and the everyday name for the same procedure varies from language to language. The second is transliteration: in Arabic and Persian queries, brand and city names are searched both in the local script and in Latin characters, and the two are separate queries. The third is geographic framing: in some languages the search is made with the country name, in others with the city name. All three variants should sit as separate rows in the seed set.
Step 2: separating into intent layers
A raw seed set is useless; without being split into intent layers it cannot be connected to page architecture. In health tourism, four layers are enough in practice:
- Information layer: what the procedure is, how it is done, the recovery process. This layer usually holds the largest volume and does not produce bookings directly; it builds trust and introduces the brand at an early stage.
- Comparison layer: comparisons of country, city, clinic or method. This is the middle of the decision stage and the layer that diverges most between languages.
- Practical layer: price range, duration, accommodation, visa and travel logistics. Commercial intent is highest here, and this is also the layer most sensitive to regulation.
- Brand layer: clinic name, doctor name, reviews and contact. Volume is low but conversion rate is the highest of any layer; when it is neglected, somebody else answers those queries.
Each seed row is assigned to a single layer. If a row fits two layers at once, the row is probably too broad and needs splitting. Once the layer assignment is done, it becomes visible which layer is empty in which language — the map’s real output is that table of gaps.
Step 3: verification sources and their limits
A seed set and its layering remain assumptions until verified. Three sources are used, in this order of reliability. The most reliable is the clinic’s own search performance report: it is recorded impression and click data, but it covers only the queries the site already appears for, which means it does not show the gaps. The second source is the advertising platforms’ keyword planners: broad coverage, but the figures they give are range estimates and are grouped for advertising purposes. The third is trend tools: they give normalised relative interest rather than absolute volume, and sampling noise is high in low-volume languages.
The rule for using all three together is simple: take the absolute figure from none of them, and treat as priority the rows where all three point the same way. Mark rows where the direction disagrees as "uncertain" on the map, and make the investment decision on those only after seeing how the first published pages perform.
Step 4: connecting the map to page architecture
The map is a table of language-and-layer intersections. Each cell either points to an existing page, is empty, or is a cluster where several pages overlap. The three states have three separate actions: strengthen the existing page, open a new page, merge the overlapping pages. The order in which those decisions are applied is the same working order we describe on our health tourism service page: overlap is cleaned up first, then gaps are filled.
A common mistake is targeting the same page count in every language. Layer depth differs between languages, and a layer that needs ten pages in one language may be served by two in another. The map exists to show that asymmetry; forcing symmetry means not reading it. Publishing order, too, comes not from volume but from the size of the gap multiplied by commercial value.
One final caution: a demand map is not a document you produce once and put on a shelf. The layer assignment and the gap table should be updated quarterly with the clinic’s own search performance data. Every page published produces the input for the next update; that is the loop in which the map becomes useful.
Three common ways the map breaks
The first mistake is building the seed set around one person’s idiom. The phrasings used by a single Arabic-speaking coordinator inside the clinic are close to the language of the country they came from and can diverge from the terms a patient in another market uses. Increasing the number of sources — several coordinators, the raw text of incoming messages, questions patients wrote themselves — reduces that skew noticeably.
The second mistake is skipping the comparison layer. The information and practical layers fill up quickly because they are easy to write; the comparison layer is both more laborious and requires care from a regulatory standpoint. Yet the international patient’s moment of decision usually happens precisely in that layer. When it is left empty, somebody else’s content decides the patient’s choice.
The third mistake is using the map only to produce new pages. The map’s most valuable output is often a merge decision: reducing three weak pages that serve the same intent into one strong page delivers results faster than opening three new pages, and it simplifies the internal link structure as well.
Keeping the map alive
A demand map does not live as a document separate from the clinic’s publishing plan. The arrangement that works in practice is to add a status label to every cell of the map: live, being written, planned, out of scope. The out-of-scope label matters especially — recording that a cell was deliberately left empty prevents having the same argument again next quarter.
The input to the update cycle is the performance data of the published pages themselves. Three months on, the queries the published pages have started appearing for reveal phrasings the seed set did not anticipate; those new phrasings are added to the next version of the seed set. The map thus shifts from estimate towards measurement, and rests on slightly fewer assumptions each quarter.