How to Localise a Winning Video Ad for New Markets

Take an ad that has a real read in the source market — 48–72 hours, around 1,000 impressions, and a cost per result you would scale — and write down why it won: the hook angle, the avatar, the offer. That note is the brief for every new market. If you cannot say what won, you will translate noise and call it international expansion, then blame the country when it fails.

Localisation multiplies production work: new voice, new captions, often a new face, a claims pass, a landing page in the right currency. That cost is justified when the structure is proven, not when you are still hunting a hook at home. Keep the source-market 3 × 2 grid in the archive so you know which of the six cells you are cloning rather than 'the vibe of the campaign.' Localise the winning cell, not the whole losing set. If the winner was a complaint hook with avatar B and a bundle, that is the cell you rebuild — not a mash of all six.

Pick the next market for operational reasons as well as media: you can fulfil, the site or storefront exists, currency and shipping are honest, and someone can review claims in that country's language. A cheap CPM in a country you cannot support is not a win; it is a refund queue. One new market done properly beats five dubbed dumps that all share an English checkout. You can generate language variants quickly; you cannot fake local trust in the first second, and you cannot fix fulfilment with a better caption.

Give a native speaker the winning structure and the why-it-won note, not a sentence list to convert in a spreadsheet. The hook must land in under three seconds in that language, which often means fewer words, a different idiom, or a different cultural reference. Keep the four beats — hook, problem, demo, CTA — and throw away English word order without apology.

Translation preserves meaning on paper and fails in a feed, where the first second is a spoken line plus a face. A question hook in English may need to become a complaint in Spanish; a pattern interrupt may need a local product ritual instead of a US bathroom trope. Numbers, units and currency must match the page the click hits. If the English hook used a measurement your market does not use, convert it in the spoken line, not in a caption asterisk nobody reads. The native writer owns the first sentence; you own the job the sentence has to do.

Scripts and voiceovers can run in 30+ languages, which is the easy part of the workflow. The hard part is the first line sounding like a person who lives there. Have the native writer read the hook aloud on a phone timer. If it overruns three seconds, cut words rather than speeding the voice. Do not cram an English-length sentence into a language that needs more syllables — that is how localised ads sound dubbed even when the lip-sync is technically fine. Short, spoken, local, and timed like a feed, not like a brochure.

A US-coded face and accent in a new country is a hook-rate tax you pay before the product appears. Pick from the 50+ avatar library for age, presentation and setting that would plausibly record this recommendation there. Pair a native voice. Do not stretch an English founder clone into a language they do not speak and hope lip-sync hides the mismatch.

Setting matters as much as ethnicity or age. A car-interior unboxing may read as native in one market and as imported influencer content in another. Bathroom-mirror UGC that is ordinary in California can be a problem in markets with stricter dress and gender norms — recast and reframe, do not just swap the voiceover on the same footage. If you sell into several Spanish-speaking countries, Mexico-cast with Spain-Spanish copy still sounds foreign in the first second. Be specific about country, not 'LATAM' as a face, and not 'Europe' as an accent.

Keep B-roll that is culturally neutral or reshoot what is not. A US plug, a US fridge, a highway that is obviously California — viewers notice even when they cannot name why the ad feels imported. Product-in-hand close-ups travel; lifestyle wides often do not. If you cannot reshoot, crop tighter on the product and let the local avatar carry context. Consistency of pairing (one face, one voice per market) still applies; do not test five random international avatars on day one just because the library is large. That is a new test, not localisation of a winner.

Burned-in English captions over a local voice is the fastest way to look lazy on mute, which is how a lot of the market will meet the ad. Regenerate captions in the target language, re-break lines to 4–6 words, and check that longer languages still sit in the centre-safe area. End cards, price chips and 'Shop now' supers need the local language and local currency too.

German, Spanish and Hindi lines often run longer than English, and a layout that was tight in English will overflow or flash unreadably if you keep English timing. Rewrite for the space you have, then time to the new voice, not to the English edit. Right-to-left languages need a separate layout pass so the first word is not sitting under a platform icon on the right edge. Safe zones still eat the bottom fifth and the right edge on TikTok-style placements — do not 'solve' overflow by dropping text into the UI because it looked fine in a desktop player.

Export a 9:16 1080p master for that market and only then derive 1:1 or 16:9. A caption stack that fitted English 9:16 will collide after translation; check each ratio as if it were a new file. If the brand kit has a locked English end card, make a market version rather than shrinking English legal text until it is unreadable on a phone. On-screen claims must match the spoken line word for meaning. A mismatch is both a trust problem and a review problem, and it is one of the first things a native QA pass should catch.

Platform approval in Ads Manager is not legal clearance in the country you are buying. Health, typical-results, earnings and environmental lines that ran in the US often cannot run as-is in the UK, the EU or India. Rebuild the claim from the product fact you can substantiate there, and add the local disclosure the format requires. Do not paste 'results not typical' over a transformation hook and call it done.

Keep a claims sheet per market: allowed, needs qualifier, banned. EU and UK practice is harder on implied typical results and on nutrition language; India has its own overlay on fairness, disease and 'clinically proven' tropes; several LatAm markets treat testimonials differently from each other even when they share a language. If the English winner only works as a medical miracle, it is not a candidate for localisation — retire that hook rather than watering it into nonsense that no longer converts and still is not safe. The structure can travel; the miracle usually cannot.

Disclosures belong in the language of the ad, on screen long enough to read on a phone, not in a landing-page footer a paid click may never reach. An AI avatar is a paid, scripted presenter: do not invent a personal medical outcome for a face that is not a real customer in that market. Safer local ads present the product, use real permitted proof, and stay specific. Have a human who knows that market sign the sheet. The generator will localise language; it will not tell you whether a claim is lawful, and treating a green ad-account status as legal sign-off is how international tests get expensive.

A perfect local video that clicks through to a USD English checkout is a broken ad, no matter how good the transcreation was. Match currency, shipping promise, bundle and language on the destination. If you cannot support the order, do not run the market. The last three seconds should name a next step that is actually true there, in the same language as the voice.

Offer localisation is often the real performance lever once the hook is rewritten. A US 'subscribe and save' may need to become a single-purchase bundle, a local marketplace SKU, or a COD-friendly message depending on how people actually pay. Do not invent a discount that the page does not show. UTM and naming should include market and language codes so you do not mix GB-en and US-en in one report row and then 'learn' the wrong thing. If you use catalogue ads, the feed must match too — stale FX-converted prices are a trust and policy problem as well as a conversion problem.

If the page is only half translated, pause paid until the first screen after the click is in-language, including the primary CTA and the price. Paid social will send people who believed the video. A language snap-back at the cart is how you buy expensive bounces and train the algorithm on the wrong conversion quality. Creative localisation without destination localisation is unfinished work, and it is the usual reason a 'localised' campaign looks like the country 'does not convert' when the country never got a local checkout.

Have a native speaker in the target country watch the cut on a phone: pronunciation, slang, on-screen typos, and 'would a local actually say this.' Fix before spend. Then launch a small test — same discipline as home market, 48–72 hours or about 1,000 impressions per variant — rather than duplicating budget at scale on day one because the English file was a winner.

QA is not 'a bilingual colleague in your HQ who spent a semester abroad.' Regional Spanish, Portuguese versus Spanish, Hindi versus English for urban India, Gulf Arabic versus a Levantine read — the wrong variety is audible immediately and costs you the first second. Ask the reviewer to rewrite the hook if it sounds translated, not only to mark typos in a doc. Check lip-sync on the new language; regenerate rather than stretching an English mouth shape under a longer sentence. If they hesitate on a claim, send it back to the claims sheet, not to the editor for a smaller font.

Klip Kanvas can regenerate the winner in a new language with lip-sync, which is the production shortcut; the market still needs the transcreation and claims pass above or you have only dubbed the file. Start with two local hooks × two local-appropriate avatars if you have the credit, not twenty languages at once. Scale spend only when the local CPA is a number you would accept, not when the English winner 'should work there.' Different auctions, different proof, different first second. Treat the new market as a new test with a head start, not as a copy-paste of home.