Keyword Research in Estonian Without Speaking It

Short answer

You can build an Estonian keyword list without speaking Estonian, on one condition: you take the phrasings from places where native speakers have already written them, not from a translator. Those places are the title tags and H1s of Estonian sites in the top 10, google.ee autocomplete, the query report in Search Console, and Google Ads Keyword Planner set to Estonian.

Translating an existing list fails for the same reason translating pages fails: the translator gives you a grammatically correct phrase while people type something else. On top of that, Estonian glues words together and puts the first part into the genitive, so a mechanically translated list is often full of forms that do not exist in the SERP at all.

Why a translated keyword list does not work

I have received the “here are our keywords, please translate them into Estonian” spreadsheet many times. The problem is not translation quality — modern engines handle the grammar. The problem is that a search query is not a sentence, it is a typing habit, and habits differ by language.

Three mismatches I see over and over:

• A different level of abstraction. Where an English speaker searches for “dentist services”, an Estonian is more likely to search for the specific procedure or the price: hambaravi hinnad, implantaadi hind. The translation gives you a signboard while demand goes to an action.
• Loanword instead of the official term. The formal Estonian term for search optimisation is otsingumootori optimeerimine, yet a significant share of demand goes to the short SEO teenus. Build the list from a dictionary and the short form never enters it.
• Different word order and compounding. “Car repair in Tallinn” lives in Estonian as autoremont Tallinnas — one word instead of two, plus a case ending on the city name.

Hence the principle I apply on every multilingual project: each language gets its own keyword research, not a translation of someone else's. It is the same rule I covered in the article on which language to target for SEO in Estonia: a language version is an independent site, not a folder of translations.

What Estonian breaks in standard keyword research

For Estonian queries, Keyword Planner almost always shows ranges instead of exact numbers, and in a narrow B2B topic the difference between “10–100” and “100–1000” decides everything. So the number is not a criterion for me, it is a sorting tool: it helps prioritise inside an already collected list, while the decision to build a page comes from the SERP and the commercial value of the query.

Five sources of native Estonian phrasings

1. Title tags, H1s and menus of Estonian competitors. The most honest source there is: these are the phrases native speakers chose themselves to sell the service. I crawl the top 10 for a handful of seed queries and dump every title, H1 and menu item into one list. After twenty sites you can see which phrasings repeat everywhere — those are the head terms of the cluster.

2. Autocomplete and “Related searches” on google.ee. Open the SERP with the Estonian interface and Estonia as the location. Autocomplete shows what people actually type, and the related searches block at the bottom surfaces neighbouring phrasings no dictionary will give you.

3. The Performance report in Search Console. If the site already has even one Estonian page, that report is gold: it lists the queries you are already shown for, including forms you would never have guessed. On an existing project I always start there.

4. Google Ads Keyword Planner. Language Estonian, location Estonia. It expands the list and gives you the order of magnitude for volume. The search terms report of a running campaign is even more useful: it shows what people really typed rather than what the planner assumed. That is, incidentally, one argument for starting with ads before SEO on a market you do not know yet.

5. Checking the word form in a dictionary. If you are unsure whether a compound exists and how it is spelled, check Sõnaveeb by the Institute of the Estonian Language: it shows the standard spelling and the case forms. It is not a source of queries — it is a filter against your own inventions.

How to check that a phrase is alive

Every collected list contains junk: forms nobody types and words that mean something else in Estonian. Validation takes about a minute per phrase and happens directly in the SERP.

1. Type the query into google.ee with the Estonian interface language.
2. Look at the language of the results. If the top 10 is English or Finnish pages, either the phrase is not Estonian or demand is too small for Google to have assembled a local result set.
3. Check whether Google corrected you. A “Did you mean” line, or a silent query substitution, means you are holding a form that does not work.
4. Assess the page types ranking. Catalogues and product pages mean a commercial query where an article will not rank. Articles and forums mean the opposite.
5. Note who is ranking. If the top is held by large portals and marketplaces, you are competing with media rather than with specialist sites, and the cost of entry is entirely different.

The fifth point matters most on a small market: in Estonia one phrase can have a laughable search volume and still bring a client worth several thousand euros. And the other way round — a nice-looking generic query can turn out to be navigational towards one specific brand.

Handling cases during clustering

The question I get most often: do hambaravi hind and hambaravi hinnad need separate pages? They do not. For Estonian, as for other inflected languages, Google reduces word forms to a lemma, and both phrases almost always return the same SERP. Only keyword tools show them as different, because they count strings rather than meanings.

The working rule: if two phrases share 5 or more URLs out of Google's top 10, that is one cluster and one page. Exactly the criterion I described in the article on keyword clustering — language changes nothing here, it only changes how many forms you have to merge.

What genuinely deserves separate pages is different intent hiding behind similar words. veebilehe tegemine (building a website) and veebilehe korrashoid (website maintenance) look like neighbouring phrases but are two different services and two different landing pages. Once the clusters exist, they turn into site structure — and on a multilingual project the Estonian structure may differ from the Russian one, which is fine.

How I work on Estonian projects

I have been running projects in three languages since 2017, always in the same order.

1. Estonian version first, the rest afterwards. The Estonian SERP holds the bulk of commercial demand in the country, and the structure is usually built from it.
2. Seed phrases from the SERP, not from my head. Twenty to thirty competitors, every title and H1 into a spreadsheet.
3. Expansion via autocomplete and the planner. At this step the list grows several times over, mostly through the long tail.
4. Manual cleaning. This is where not speaking the language hurts most, so I validate doubtful phrases in the SERP rather than by intuition.
5. Clustering by SERP overlap and distribution of phrases across pages.
6. Native review at the copy stage, not the keyword stage. The keyword list is made of phrases taken from real search results, so it needs no linguistic review. The page copy written from those phrases does: naturalness of phrasing is exactly where an AI model and a non-native writer are wrong with equal confidence.

An honest limit: this process is tuned for service sites and small catalogues, in the range of dozens to hundreds of pages. For a shop with tens of thousands of products, checking every phrase in the SERP by hand is unrealistic — there you need a different approach, closer to programmatic SEO with templates and sampled validation.

A mistake I made myself: I built the Estonian list by mirroring the Russian structure. The result was pages targeting phrases that had neither demand nor competitors in the Estonian SERP, simply because the service is sold with different words on that market. Now I first look at what kind of pages exist in the Estonian top 10, and only then decide what to build.

What to check before calling the list finished

• No phrase comes from translation alone: each one is confirmed by the SERP, autocomplete or Search Console.
• Compounds are written as one word, the first part in the genitive.
• Same-meaning forms sit in one cluster instead of being spread across pages.
• Both the loanword and the formal term have been tested, and the one with a relevant SERP was chosen.
• Low-volume keywords have not been dropped mechanically by a Keyword Planner number.
• Estonian and Russian clusters do not share a page: language versions are linked with hreflang rather than merged.

If even one point fails, you will be rebuilding the list as soon as the first Search Console data arrives — and that costs more than checking now.