Keyword Research for the Romanian Market: The Bilingual RO/EN Workflow in Semalt

Keyword research for Romania has a specific problem that imported methodologies do not solve: volumes are small, and tools frequently display zero for queries that in reality bring customers every week. If you filter the list by a minimum volume of 100, you cut exactly the part that converts.

On top of that, the same need is expressed in three ways. Someone looking for an ergonomic office chair types "scaun ergonomic", "scaun de birou ergonomic" or simply "office chair" - the last being very common in the Bucharest corporate segment. A site tracking only the first form believes the niche has 300 searches a month when it actually has more than a thousand.

The guide below is the process we use, built around real data rather than estimates.

Stage 1: start from what you already have, not from an empty list

The best source of keywords is not a generator, it is your own Search Console. That is where the queries Google already considers you relevant for live, with real impression, click and position data.

In the Search Console analytics module you extract three separate lists:

1. Positions 4-15 with above-average impressions. These are the immediate opportunities. A query in position 6 with 800 monthly impressions needs three small changes, not a six-month campaign.
2. Positions 1-3 with CTR below 15%. You are up there and people are not clicking. The problem is in the title, in the snippet or in intent - you rank on a query that does not match what you offer.
3. Queries with impressions, zero clicks, growing month over month. The market is moving in that direction. These are candidates for new content.

On a mid-sized site the first list produces between 20 and 60 concrete actions. Most projects need nothing else for the first two months.

Stage 2: expansion through language variants

This is where the Romanian specifics come in. For each core query you build a small set of variants:

Not all of them deserve separate content - most of the time one page covers them all. But all of them deserve separate tracking, because position can differ significantly between variants and that is where opportunities hide.

Stage 3: validation through real positions, not estimated volume

The rank tracking module works on any domain with no ownership verification. That completely changes the order of work: instead of estimating whether a query is worth competing for, you look at who is already there.

For every candidate query note three things: who holds the TOP 3, what page type ranks (category, product, guide, list) and whether results are dominated by aggregators. That last observation is decisive in Romania, where on many commercial queries the top positions are held by large marketplaces. In such a SERP a small shop will not take position 1 with a product page - but it can take position 2-3 with a well-built comparison guide.

The rule we apply

If there is no site of your size anywhere in the TOP 10, the query is not for this year. Note it and move on. A content plan built on unrealistic queries burns budget without moving anything.

Stage 4: the queries people ask AI

Search behaviour has forked. The same person searches "sofa bed price" in Google and types "which sofa bed survives daily use and costs under 4,000 lei" into an AI chat. The second form is longer, more specific and, in many niches, closer to the buying decision.

The AI visibility module shows which queries a domain appears in inside model answers, which pages get cited and which competitors are named. The system tracks over 1.2 billion AI queries and ranks more than 2.8 million websites.

How we use it in keyword research: we check the competitor who appears most often in answers and look at their cited pages. The pattern is almost always the same - they are pages that answer a question directly, with clear structure, concrete numbers and no long preamble. That gives you both the topic list and the format.

Stage 5: grouping by intent and mapping to pages

The final list splits into four groups, and each group gets a page type:

• Transactional ("buy", "price", "company") - category and service pages.
• Commercial investigation ("best", "comparison", "review") - comparison guides that link internally to commercial pages.
• Informational ("how", "what does it mean", "how long does it take") - blog articles with a clear internal link to the next step.
• Navigational / branded - homepage and brand pages; no new content needed here, only clean titles.

The classic mistake is writing a blog article for a transactional query. If the TOP 10 contains only category pages, your article cannot break in no matter how well written it is.

Stage 6: execution - manual or through a campaign

Once the list is ready you have two paths. Manual: you write and optimise yourself, checking positions weekly. Or through a managed campaign, where selection and prioritisation happen automatically on data: AutoSEO at $149 a month picks keywords with AI, prioritising queries the site already touches in Google, and you can add your own queries at any time. FullSEO, from $500 a month, adds manual control over the list and research done by specialists, including high-volume queries.

Our recommendation for a Romanian project at the start: two months of manual work on the Search Console list, then a campaign if you want speed. In that order you know exactly what the campaign delivered, because you have a clean baseline.

A complete example: a dental clinic in Bucharest

To see the workflow applied, here is a real keyword map, simplified, for a clinic with three practices in Bucharest. Content budget: four new pages and eight articles in one quarter.

The important observation: the query "dental implant Bucharest price" showed a modest volume in the estimator. In Search Console, however, it appeared with steady impressions and an average position of 11 - exactly the kind of opportunity from the first list in stage 1. A service page with real prices, published and submitted for indexing, moved the query into the TOP 5 in seven weeks. No link investment was needed for that.

The second observation: the page that brought the most appointments was not the one with the biggest volume, it was the comparison guide. People comparing two solutions are already deciding; those searching for a general price are still exploring.

How to build a topic cluster that actually works

A cluster does not mean "ten articles on the same subject". It means one central page that sells and a set of satellite pages answering the questions that come up before purchase, all linked in a way that makes sense to a reader.

The minimum structure we recommend has four layers. The central commercial page, optimised for the main transactional query. Two or three comparison guides, catching "X or Y" and "best X" queries. Four to six short informational articles, each answering one concrete question. And a frequently asked questions page collecting the real questions clients ask - that last one is often the most cited in AI answers, precisely because it is structured as question and answer.

Internal links run both ways: every satellite links to the central page with descriptive anchor text, and the central page links to three or four relevant satellites. Check monthly, in ranking data, whether any satellite has started ranking for the central query - if so you have cannibalisation and need to narrow the satellite's topic.

What to watch in the first 90 days

Once the list is built and the first pages are live, the checking rhythm is simple and short. In the first two weeks you look only at indexing: have the new pages been discovered and are they receiving impressions? If not, the problem is not the keyword but discovery, and the indexing module fixes that quickly.

Between weeks three and six you track entry into the TOP 50 and then the TOP 20. Movement in that zone is the best early signal that the page is considered relevant. Do not adjust the copy during this period - if you change it every week, you will never know what mattered.

Between weeks seven and twelve you look at CTR and at the queries the page actually attracts. Queries you never considered almost always show up; those go into the tracking list for the next quarter. The cycle closes and research starts again, but with better data than at the beginning.

How often to redo the research

The main list gets reviewed quarterly. Between reviews, the only thing worth checking weekly is what newly entered the impression-generating queries in Search Console - that is where trends appear before they become obvious. In seasonal Romanian niches (travel, gifts, construction) the review has to happen two months before the season, not during it.

How to present the keyword list to a client

A file with 400 rows and volume columns convinces nobody. What convinces is a single page with three columns: the query, the page that serves it, and what we do concretely for it this month.

In the presentation we group queries into three effort categories: "quick wins" (positions 4-15, on-page optimisation only), "new content" (pages that are missing) and "long term" (queries where competition is far above the current level). The client immediately understands where the first two months of budget go and why the rest requires patience.

The appendix with all the data stays available but is not discussed in the meeting. Nobody makes decisions looking at 400 rows.

Three local traps

A displayed volume of zero does not mean zero searches

For long, specific queries tools frequently display zero. If Search Console shows impressions, the query exists, whatever the estimator says.

Directly translating an English list does not work

Query structure differs. In English people search "best running shoes for flat feet"; in Romanian, "pantofi alergare talpa plata" - shorter, without prepositions, often without diacritics. A translated list produces pages that match no real search.

Bucharest is not one single market

On locally-intended queries, results differ between districts. If you sell services that involve travelling to the customer, track positions from several areas rather than only the centre.

The tool does not replace judgement

No platform tells you which query is worth the effort in your business context. A keyword with 50 monthly searches that brings contracts worth thousands of euros beats any generic query in the niche. Data shows you what is possible; the decision about what is valuable stays with you and the client, and the best source for it is the sales team, not the tool.

Frequently asked questions

How many keywords should I track?

At the start, 30-50, all directly tied to revenue. Expand once you have a stable weekly rhythm of reading the data.

Can I see a competitor's keywords?

You can see which queries their domain ranks for and which AI queries it appears in, without any access to their accounts. Search Console data obviously stays private.

Is it worth tracking the English version of my site too?

Yes, as a separate set. Otherwise you will not notice cannibalisation between versions, which is the most common problem on bilingual Romanian sites.