Back to the blog
Try siriusly.ai

Which markets to localize first - and what breaks in each

Producing an ad in eight languages used to be an eight-shoot budget question. Now the production cost is close to flat, which moves the hard part from whether you can afford it to which markets deserve it and in what order.

When localizing meant re-shooting, the decision made itself: you localized into the one or two markets big enough to justify a second production, and everything else waited. Generated video removes most of that cost, and a lot of teams have responded by localizing into everything at once. That is usually a worse mistake than localizing into nothing, because each additional market adds real fixed cost - support in that language, a landing page, payment methods, returns, claims review - none of which got cheaper just because the video did.

So the useful question is a ranking question. Here is a method that uses signals most teams already have, and then the specific things that break language by language.

One brief, every market

siriusly.ai generates each language natively - cast, written and spoken in-market - rather than dubbing one master, so a new market costs a render instead of a shoot.

Try it free

Rank on Evidence You Already Have

Nearly every company has more evidence about foreign demand than it realises, and it is better evidence than a market-size report, because it is about your product specifically.

SignalWhere to find itWhat it tells you
Organic traffic by countryAnalytics, unpaid sources onlyWhere people already look for you without being asked
Conversion rate by country on an English pageAnalytics segmented by localeDemand strong enough to overcome a language barrier
Existing customers by countryBilling and shipping dataThe only fully proven signal in the list
Support requests in other languagesInbox and helpdeskDemand you are currently converting badly
Cost per acquisition on English ads run abroadAd platform, geo breakdownA cheap pre-test that needs no localization at all

That last one deserves emphasis, because it is the closest thing to a free experiment available: run your existing English creative into a candidate market and look at what it costs to acquire someone despite the language mismatch. A market that converts acceptably in the wrong language will usually convert well in the right one. A market that does not convert in English may still be a good market, but you have learned much less.

A market where English ads already half-work is the best localization bet you can make. You are removing a known friction, not testing a hypothesis.

The Non-Creative Costs That Decide the Order

Once the demand signals give you a shortlist, the ordering is usually settled by operational readiness rather than by upside. Before committing to a market, check that you can actually serve it:

  • Payment methods. Card-first checkout underperforms badly in markets where local methods dominate. This alone can invert the ranking.
  • Delivery and returns for physical goods, including who pays duties. An ad that converts into a shipping surprise generates refunds, not revenue.
  • Support in the language. Advertising in a language you cannot answer email in is a support debt with a marketing budget attached.
  • Claims and advertising rules. These vary by country as well as by category, and a claim that is fine in one market can be prohibited in another - comparative claims, health language and pricing displays are the usual culprits.
  • Landing page and checkout language. A localized ad pointing at an English page loses much of what localizing bought. This is often the largest hidden cost of a new market.
A workable first wave

Pick the top three markets by existing-customer count, filter out any where you cannot yet take local payment or answer support, and localize those. Run the English creative into the next five as a paid pre-test while you do it.

Wave two is then chosen with real data rather than by guessing, and you have not paid landing-page and support costs for markets that were never going to work.

What Breaks, Language by Language

Translation quality is rarely the failure. The failures are structural, and they repeat:

Text expansion

The same sentence gets longer or shorter depending on the language, and the differences are large enough to matter. German and French commonly run substantially longer than English; some languages run shorter. Two consequences follow immediately: a script written to fit a fixed video duration no longer fits, and a caption line that fitted two lines now fits three - which pushes it down into the platform chrome. Both are avoidable by writing per language rather than translating and hoping, and by keeping captions to a fixed band and line cap, as covered in the safe-zone piece.

Register

Languages that distinguish formal from informal address force a choice English never makes. For consumer short-form the informal form is almost always right - du rather than Sie, tu rather than vous, tu rather than usted - because the entire genre is one person talking to another person. Getting this wrong does not read as a translation error; it reads as a corporation pretending to be a friend, which is worse.

Number and currency formats

Not just how they are written - how they are said. A decimal comma is read differently, currency position moves, and a price read with English number habits in a German sentence is instantly foreign even when every word is correct. This is a speech problem as much as a text one, and it is the single most common tell in a poorly localized ad. The mechanics are in the piece on AI voice.

Caption density and script

Languages written in denser scripts fit more meaning per line and are read at a different speed, which changes how long a caption should stay on screen. Right-to-left languages need the layout mirrored, not just the text swapped. Neither is hard, but both are silent failures - the video plays, and only a native reader can see that it is wrong.

Casting and setting

The creator, the room behind them and the way they speak all carry market signals. A creator who reads as American speaking fluent German still reads as an American speaking German. Where the whole format depends on looking like a real person from the viewer's world, this matters more than it does in any other ad format - which is the core of the argument for generating each market rather than dubbing one master.

Language Is Not Country

Two mistakes hide in treating a language as a market:

One language, many markets: Spanish for Spain and for Latin America differ in vocabulary, accent and pricing context. Portuguese for Brazil and Portugal likewise. Shipping one Spanish ad everywhere reads as foreign in most of the places it lands.

One market, many languages: plenty of countries have several widely spoken languages, and the one your customers use may not be the official first one. Segment on the language your data shows, not on the map.

The practical rule: pick the language variant per target country, and let cost decide how granular to get. Since a variant is now a render rather than a shoot, the granularity can be finer than it used to be - but only where you have the operational side to support it.

How to Test a New Market

  1. Pre-test in English to establish whether there is demand at all, before spending anything on localization.
  2. Localize the winning creative only. Do not localize a creative that has not proven itself in its home market - you will be testing two variables at once.
  3. Have a native speaker watch it once. Not a full review, not a translation audit - one native speaker watching the finished video end to end catches register, number reading and casting problems in under a minute. This step is worth more than any automated check.
  4. Point it at a localized page. If the page is not ready, the market is not ready.
  5. Judge on cost per acquisition, not on engagement. Foreign-language creative frequently gets cheap engagement from outside the target country, which flatters every metric except the one that matters.
  6. Then localize the rest of the matrix. Only after one creative has proven the market is worth supplying variety to.

Common Mistakes

Localizing into everything at once because production is cheap. Production is the small cost. Support, pages, payments and claims review are not, and they arrive per market whether the market works or not.

Fix: rank markets on evidence you already have, and gate each one on operational readiness before it gets a budget.

Translating a script written for a duration. The translated line runs long, the pacing compresses to fit, and the read sounds rushed in exactly the market you were trying to impress.

Fix: write to the duration in each language rather than translating to it. Length is a constraint on the writing, not something to fix afterwards.

Using the formal register by default. It is the safe choice in business correspondence and the wrong one in a format built entirely on sounding like a person.

Fix: informal by default for consumer short-form, and make it an explicit decision rather than a translator's guess.

Sending localized traffic to an English page. Most of the gain from localizing the ad is given back at the click.

Fix: treat the landing page as part of the market launch, not as a later optimization.

FAQs

How many markets should a first wave have?

Three is a good number: enough to learn whether localization moves your numbers at all, few enough that the operational work is real rather than theoretical. Expand once you have a result rather than in parallel with getting one.

Is machine translation good enough now?

For meaning, largely yes. For advertising, meaning is not the bar - register, idiom, number reading and cultural fit are, and those are exactly what a translation step is not designed to handle. Generating the script natively in the target language avoids the problem rather than correcting for it.

Do I need a different creator per market?

Usually yes, and it is one of the cheapest levers available now that casting is a parameter rather than a contract. Where a brand deliberately runs one global spokesperson, keep the person and change the language - but be aware you are trading local fit for brand consistency.

Can I reuse the same hooks across markets?

The mechanism travels; the wording rarely does. A cold open works everywhere; the specific sentence that opens it is culturally loaded. Retest hooks per market rather than assuming the home-market winner wins abroad - the taxonomy to test against is in the hooks piece.

What if my product is not available in a market yet?

Then it is not a market, it is a waiting list. Running ads into it produces demand you cannot serve, which is expensive in refunds and worse in reviews. Wait, or run a genuine waiting-list campaign that says so.

§ End · July 3, 2026
Was this useful?

Keep reading.

All posts
Localization

Why Native Generation Beats Lip-Sync Dubbing

Once you have picked the markets, this is how to produce them.

AI Video

Making an AI Voice Sound Like a Person

Number reading and register are where localized audio gives itself away.

Skip the pipeline.
We already built it.

siriusly.ai chains Gemini Omni, Wan and Seedance chunks automatically - last-frame continuity, prompt drift-control and all - so you just write a brief.

Start free