Best multilingual SEO agency, and the per locale AI search answer nobody is checking

The right multilingual SEO agency is the one that refuses to quote until you have decided which locales you are actually committing to. That sounds like a dodge and it is the opposite: the price, the timeline and the whole technical design depend on the locale list, and a supplier who accepts "we want to go international" as a brief is going to translate your existing site and invoice you for it. A translated site is not a multilingual site and it rarely earns anything.

This page covers how to settle the locale list, the four kinds of supplier who answer this search, the criterion to rank them on, and what belongs in the contract. If what you need is the technical architecture rather than a supplier, the hreflang and localisation design is covered in multilingual SEO architecture that holds up, and the URL structure decision in URL structure, hreflang and local signals.

Settle the locale list first, on three tests

A locale earns a page if it passes all three. Most candidate locales fail the second or the third, and discovering that after you have paid for translation is the most common expensive mistake in this category.

• Demand. There is search volume in that language for what you sell, researched in the target language rather than translated from your English keyword list. The literal translation of your best English phrase is frequently not what people in that market type.
• Fulfilment. You can actually serve that market: you ship there, you are licensed there, the price works there, and somebody can answer the phone in that language. Ranking in a market you cannot serve is a cost, not a win.
• Maintenance. Somebody will keep that locale current for at least two years. A stale locale is worse than no locale, because it ages in public with your name on it.

Language and locale are also not the same choice, and conflating them is the error that produces the most rework. Spanish for Mexico and Spanish for Spain are different briefs, as are Portuguese for Brazil and Portuguese for Portugal. Naming the locale rather than the language is free at the brief stage and expensive afterwards.

Four kinds of supplier, and what each is genuinely good at

Supplier	Strength	Weakness	Right when
International SEO specialist agency	The architecture, the hreflang, the market prioritisation, the technical QA at scale	Rarely has native writers in every locale, so it subcontracts the language work	The site is large, or multiple locales are launching at once
Translation or localisation agency with an SEO practice	Language quality, transcreation, terminology and translation memory	Often weaker on the technical side and on in market search research	The architecture is already right and the content is the job
In market freelancer or small agency, per locale	Genuine local search knowledge, which is the thing nobody else can fake	Coordination cost multiplies by locale, and consistency drifts	One or two priority locales that matter commercially
Your existing SEO agency, extended	Already knows your site, your product and your reporting	Usually has no native capability and will subcontract the parts that matter most	You want one throat to choke and accept a markup on the language work

The common and sensible shape for a serious programme is the first supplier for the architecture and the third for the language, with you owning the seam. The thing to avoid is buying the second and expecting the first, which is what happens when a translation quote is accepted as an SEO plan. What a translation agency actually sells at each service level, including transcreation, is set out in the four service levels and how to brief one.

The criterion to rank suppliers on, applied to everybody including us

The criterion is: will the supplier do keyword research in the target language, by a native speaker, and show you the output before you commit to the locale? It is checkable, it is cheap, and it is the single step that separates a multilingual programme from a translation project. An agency that will run that research as a small paid first piece of work is telling you it expects some of your candidate locales to fail the demand test. An agency that goes straight to a page count has not looked.

Ranked on that criterion, in market specialists come first and your incumbent agency usually comes last, for a structural reason rather than a competence one: the incumbent has to subcontract the research and the markup sits on the most important deliverable in the engagement.

What belongs in the contract

• The locale list, with the three tests answered per locale, as an appendix. This is the document the whole engagement is priced on.
• Who owns the translation memory and the glossary. You, and get them as files. On a catalogue this is the largest recurring saving available and it only works if the memory is kept.
• Named native reviewers per locale, with authority to approve. An in market reviewer who can sign off is the difference between a two week cycle and a two month one.
• Per locale reporting, separately. One blended international report hides a locale that is failing, which is the thing you most need to know.
• A retirement plan. How a locale is wound down if it fails, including what happens to its URLs. Nobody writes this and everybody eventually needs it.
• What is translated beyond body copy: titles, descriptions, alt text, structured data, forms, transactional email, currency, units, phone and address formats, and legal notices. These are missed on most multilingual sites.

The reading that is per locale and almost nobody takes

Everything above is measured in rankings and traffic per market. There is a newer surface where your locale decisions are being tested and where nothing in your reporting shows it. When somebody asks an assistant in German, Spanish or Japanese which supplier to use for what you sell, ChatGPT, Gemini, Perplexity, Claude and Google's AI Overviews answer with named companies, and the answer is not a translation of the English one. Different sources, different companies named, sometimes a local competitor you have never heard of. A brand that is named confidently in English can be absent in every other locale it has just paid to launch in.

That makes this a genuinely useful acceptance test for a multilingual programme, and it is a better one than most: it is per locale, it is reproducible, and it reflects what a buyer in that market is actually told. Ask the buying question in the target language, record which companies get named and which sources the answer leaned on, and compare locale by locale. AI Knows Us reads those five surfaces and reports both, and the free scan on your own domain needs no account. On the published plans, Growth at 299 dollars a month watches 100 buyer questions across all five surfaces, which spreads across a few locales rather than many, and the larger plans exist for programmes with more markets than that. Two honest limits before you buy for a multilingual programme specifically: our own pricing page carries the sentence "India is charged in rupees; every other region is charged in US dollars and shown in its own currency", so the commercial terms are not themselves per market in the way a global team might expect; and our rendered pricing text carries no SSO, SOC 2, ISO or DPA, while Profound publishes "SSO/SAML + SOC 2 compliance" on its enterprise tier and tracks "Up to 9 Answer Engines" against our five, read on tryprofound.com/pricing on 21 October 2026. For a global enterprise with a security review, that is the stronger purchase.

Questions buyers ask about hiring for multilingual SEO

Can we just translate the English site and see what happens?

You can, and the usual result is pages that rank for nothing because they target phrases nobody in that market searches. Do the in market keyword research for one locale first, at small cost, and let the result decide how much translation to buy.

Subdirectories, subdomains or country domains?

Subdirectories for most businesses, because they keep the authority on one domain and are the cheapest to run. Country domains earn their cost when you have genuinely separate operations per market, and they are a commitment rather than a setting.

How many locales should we launch at once?

One, properly, then the second. A programme that launches six at once discovers its process problems six times simultaneously, and nobody has the reviewer capacity for that in month one.

Do we need native speakers, or is machine translation with editing enough?

Machine translation with light editing is reasonable for documentation and support content. For anything whose job is to persuade, and for anything carrying a claim, a native writer is the difference between a page that reads as local and one that reads as imported.

How do we judge quality in a language nobody internally speaks?

Buy a paid test translation of one real page and have somebody you trust in that market review it. This settles a supplier choice faster and cheaper than any proposal process.

What is the most common cause of a failed multilingual launch?

A locale chosen because the market sounded large, with nobody to maintain it and nobody to answer an enquiry in that language. It is a fulfilment failure that gets reported as an SEO failure.