Speaker Count Is the Wrong Metric
The most common way to choose target languages is to rank the world's languages by speaker count and work down the list.
This is the worst available basis, and it produces predictable failures: localizing into a large-population language where nobody wants your specific content, into a market that cannot transact, or into a language whose speakers already consume your content comfortably in English.
Speaker count measures how many people could theoretically understand the content. It says nothing about whether any of them want it, whether they can act on it, or whether they would have understood the original anyway.
The useful inputs are demand evidence, competitive density, English proficiency, market viability, and cost — and they frequently point somewhere quite different from a population ranking.
Start With Evidence You Already Have
Before any analysis, look at what your existing data says.
Traffic by country and language. Markets already producing organic interest despite having no localized content are demonstrating demand. This is the strongest available signal and it costs nothing to obtain.
Signups, enquiries, and conversions by market, which is a stronger signal than traffic because it indicates intent rather than curiosity.
Browser and account language settings among existing users, which frequently reveals language communities that country-level data obscures — diaspora populations, multilingual countries, and second-language users.
Support enquiries arriving in other languages, which indicate users already engaging despite a language barrier.
Existing content performance in markets where you have any presence.
A market appearing in this data with meaningful volume is a much better bet than one selected from a population table, because the demand is already demonstrated rather than assumed.
Test Search Demand
The second input is whether people search for your category in that language.
This is directly measurable and frequently reveals that a large-population language has negligible demand for your specific topic, or conversely that a smaller language has substantial unmet demand.
Search demand is particularly important because it is the mechanism that localization most reliably serves. English content does not surface for queries in other languages regardless of how well the audience reads English, so search reach is a structural gain that localization captures and nothing else does.
Check the actual terms in the target language rather than translating your English keywords, since the translation frequently produces a phrase nobody uses.
Assess Competitive Density
For a given topic, the volume of existing content differs by orders of magnitude between languages.
English content is abundant in nearly every category. Many other languages have substantial gaps, particularly in specialized, technical, and professional topics where content production has concentrated in English.
This creates an asymmetry worth exploiting. A localized version of proven content entering a market with thin coverage can establish position quickly, where the same content in English would compete against thousands of established alternatives.
Assess this directly: search the target term in the target language and look at what ranks. Sparse results, low-quality results, or results that are themselves poor machine translations indicate opportunity. Dense results from established local sources indicate a harder market.
This assessment frequently reorders priorities. A language with a smaller audience and negligible competition in your topic may return more than a larger language where the topic is saturated.
Weigh English Proficiency Carefully
High English proficiency in a market narrows the localization case but does not eliminate it.
In markets like the Netherlands, Denmark, Sweden, and Finland, most of the audience can consume English content comfortably. Localization there has to be justified by a specific mechanism: search reach, consumer conversion, regulatory expectation, reaching audiences outside the urban professional demographic, or completion in long-form content.
In markets with limited English proficiency — much of Southeast Asia, South Asia, Latin America, the Middle East, and parts of Eastern Europe — localization is a straightforward reach decision. English content reaches a narrow slice, and localizing multiplies the addressable audience.
Proficiency is not uniform within a market. It skews younger, more urban, and more educated everywhere. Content aimed at general audiences, older audiences, or vocational sectors benefits from localization even in high-proficiency markets.
The practical test: can you name the specific mechanism by which localization will produce a result in this market? If the answer is a general appeal to preference, the return is likely small.
Check Market Viability
For revenue-driven programs, reach without the ability to transact is not useful.
Consider whether the market can actually buy: payment methods available, pricing tolerance relative to your price point, currency and billing support, and delivery or service capability.
Consider whether you can support the market: can anyone respond to an enquiry in that language, is documentation available, does the product itself work there.
Consider regulatory and operational constraints: data handling requirements, content restrictions, tax and invoicing obligations.
A market that generates interest and cannot convert produces frustration rather than revenue, and the effort would have returned more elsewhere.
Factor In Cost Asymmetry
Languages differ in what they cost to localize into, and the differences are large enough to affect prioritization.
Voice availability and quality varies considerably. Major European and East Asian languages have extensive high-quality voice options; many others have fewer, and some have none suitable for professional content. This can determine whether dubbed audio is viable at all.
Reviewer availability. Finding a qualified native reviewer with domain knowledge is straightforward for widely spoken languages and genuinely difficult for others.
Technical complexity. Right-to-left scripts, complex shaping, word segmentation without spaces, and extensive inflection all add verification and process cost. Arabic, Thai, Urdu, Bengali, Finnish, and Hungarian each carry real technical overhead that Spanish or Indonesian does not.
Expansion severity. Languages that expand substantially require more editing at the timing stage.
Production overhead. Layout mirroring and rendering verification for non-Latin scripts is real work.
None of these should veto a language with strong demand, but they should inform sequencing and budgeting, and they explain why two languages with similar audience size can differ substantially in cost.
Sequencing
Order matters as much as selection.
Start with one market and finish it. One market with content, metadata, landing pages, and support capability converts. Five markets with content alone leak most of the value.
Choose the first market for learning as much as for return. The first localization surfaces process problems, and doing it in a language where you have reviewer access and reasonable technical simplicity makes those lessons cheaper.
Deepen before broadening. More content in a working market almost always returns better than the first content in an untested one. This is the discipline most programs lack.
Cluster by process similarity where possible. Languages that share technical characteristics — Latin script, moderate expansion, available voices — batch efficiently once the process is established.
Defer the technically hard languages until the process is stable, unless demand evidence is strong enough to justify solving both problems at once.
Testing Rather Than Deciding
Where evidence is ambiguous, testing beats analysis.
Localize a small batch — five to ten assets — into two candidate languages, publish with localized metadata, and measure over a reasonable window.
This produces a real answer within weeks for algorithmic platforms and within a few months for search-driven discovery, and it costs a fraction of committing to a full programme.
Read the results carefully. Low discovery indicates a metadata problem rather than a market problem. Discovery without engagement indicates a content or quality problem. Engagement without conversion indicates a funnel problem. Each points somewhere different.
Give search-driven results enough time. Evaluating after two weeks reliably produces a false negative on content that would have worked.
A Practical Framework
For each candidate language, score:
Demand evidence from your existing analytics — traffic, signups, enquiries.
Search volume for your topic in that language.
Competitive density in that language for your topic.
English proficiency and whether a specific localization mechanism applies.
Market viability — can they transact, can you support them.
Cost factors — voice availability, reviewer access, technical complexity, expansion.
Rank on the first three, filter on viability, and use cost to sequence rather than to select.
Then pick one, do it completely, measure, and decide whether to deepen or move on.
The Common Mistakes
Ranking by population. The most common and the least informative basis available.
Localizing into a market you cannot support. Interest generated and wasted.
Spreading across many markets simultaneously. None done completely, results disappointing everywhere, and no way to attribute the failure.
Ignoring competitive density. Entering a saturated market when an open one was available.
Assuming high-proficiency markets do not need localization. Search reach and consumer conversion are real mechanisms that proficiency does not address.
Assuming low-proficiency markets automatically justify it. Reach is necessary and not sufficient; demand for your specific content still has to exist.
Evaluating too early, concluding a market does not work when the content had not yet been discovered.
The programmes that choose well are not the ones with the most sophisticated analysis. They are the ones that looked at their own data first, tested rather than assumed, and finished one market before starting the next.
Revisiting the Decision
Language selection is not a one-time decision, and the inputs change.
Demand shifts. A market that showed no organic interest two years ago may now produce meaningful traffic, and your analytics will show it before any analysis would.
Competitive density changes. Content gaps close. A market that was open may now be saturated, and one that was saturated may have thinned as competitors withdrew.
Cost factors improve. Voice availability and quality for less-served languages improves over time, and a language that was not viable for dubbed content may become so.
Your content changes. A programme that expands into new subject areas may find different markets relevant than its original content did.
Market conditions change. Economic, regulatory, and platform conditions all shift, and a market that could not transact may now be able to.
Schedule a review — annually is reasonable for most programmes — checking current traffic, current search demand, and current competitive density against the languages both selected and rejected.
Equally, review the languages you are already serving. A market that has not performed after a fair period should be dropped rather than maintained out of inertia, and the effort redirected to one that responds.
The Sequencing Question in Practice
Most programmes have more candidate languages than capacity, and the practical question is what to do first rather than what to do eventually.
A workable sequence for a programme starting from nothing:
First, the market with the strongest existing demand evidence and manageable technical requirements. This produces a working process and a real result.
Second, deepen that market with more content rather than adding a language, unless the first market saturates.
Third, add a language that shares process characteristics with the first — similar script, similar technical requirements — so that the established workflow transfers.
Fourth, add a technically harder language once the process is stable enough to absorb the additional verification requirements.
At each stage, measure and be willing to stop. A programme that reaches four markets, three of which are performing, is in a far better position than one that reached twelve and cannot tell which are working.
Getting Started
For a programme with no localized content, the first decision is smaller than it appears.
Pull your existing analytics and identify the two or three markets already producing organic interest. This takes an hour and is more informative than any amount of market research.
Check search demand in those languages for your specific topic, and check what already ranks. This takes another hour and frequently eliminates one of the candidates.
Consider whether you can support the survivors — can someone respond in that language, can the market transact, does the rest of the experience work.
Pick one. Do it completely: content, metadata, landing page, support capability.
Measure over a fair window, understanding that search-driven discovery builds over months.
Then decide whether to deepen or move on, and repeat.
The framework in this article exists to be applied once at the start and revisited annually. It does not need to be run exhaustively for every candidate language, and programmes that spend months on selection analysis before producing anything have generally chosen worse than those that tested two markets and read the results.
The selection decision matters. It matters less than finishing the market you selected.
