Check this page with an assistantOpens a chat asking it to summarise this article and name the evidence behind each claim.
Claude opens with the prompt on your clipboard: Anthropic does not support prefilled prompts on the web, and we would rather copy it than ship a button that drops it.
The Decision You Cannot Cheaply Undo
Most international SEO guidance opens with hreflang. Hreflang is a tagging exercise with a right answer, and it can be fixed in an afternoon. The domain structure you commit to when entering a market is a migration to change, and it determines how much of your existing standing the new market inherits.
Three options, and the difference between them is entirely about consolidation. A subfolder is part of the domain you already have, so everything that domain has accumulated applies to it. A country domain is a separate site that happens to belong to you. A subdomain is treated as separate for most practical purposes, which means it takes on the isolation of a country domain without the geographic signal that made the isolation worth accepting.
Framed that way the default becomes obvious for most businesses. Unless you have a genuine local operation, local staff, and the budget to run each market as its own programme, splitting your effort across domains means starting several times from nothing while a consolidated competitor compounds once.
The Three Structures, Honestly Compared
Each option is right for someone. The question is which circumstances you are actually in, rather than which structure the largest companies in your category use, since their constraints are not yours.
| Structure | Strength | Cost, and who should choose it |
|---|---|---|
| Subfolder, example.com/fr/ | One domain compounding; new markets inherit everything | Weaker geographic signal. The default for almost everyone |
| Country domain, example.fr | Unambiguous local association and local trust | Starts from zero per market; only with a real local operation |
| Subdomain, fr.example.com | Operational separation, easy to delegate to a local team | Isolation without the local signal; rarely the best answer |
| Parameters, example.com?lang=fr | None worth listing | Ambiguous, hard to target, avoid entirely |
One genuine exception favours country domains beyond size: markets where a local domain is a trust requirement rather than a preference. In some countries buyers hesitate at a foreign domain in a way that no amount of localisation offsets, and where that is true the conversion difference can outweigh the SEO cost of starting fresh. That is a market research question, not an SEO one, and it should be answered by talking to people in the market.
Language and Country Are Different Questions
Serving Spanish is not the same as serving Spain. One is a language decision, the other is a market decision involving currency, shipping, legal terms, and expectations. Conflating them produces sites that are technically translated and commercially useless.
The practical test is whether anything other than the words changes. If a Spanish speaker in Mexico and one in Spain would see the same prices, the same delivery terms, and the same guarantees, you have one language version and should not pretend otherwise. If they would not, you have two markets, and the differences are the reason each page exists.
Get this settled before implementation, because it determines the shape of everything downstream, including how many locale versions you are committing to maintain forever. Once decided, the tagging that connects equivalent pages is a solved problem, and the mechanics, reciprocal clusters, x-default, and the common failures, are in the hreflang guide.
Entering a Market Without Betting the Quarter
The cheap test is a small set of genuinely localised pages in a subfolder, launched before any structural commitment. It produces real demand data from the market rather than a projection, and if the market works you have already built on the structure you would have chosen.
- Translate the decision pages first.
Product or service pages, pricing, and the handful of questions every buyer asks. Not the blog. Fifteen pages that let someone buy beat two hundred that let someone read.
- Localise the commercial detail, not just the words.
Currency, units, date formats, payment methods, delivery expectations, and any legal requirement in that market. A perfectly translated page quoting the wrong currency converts nobody.
- Have a fluent human review everything.
Machine translation as a first draft is fine and increasingly good. Published unreviewed, it reads as foreign to natives in a way that undermines trust precisely where you have none to spare.
- Research demand in the market, not in translation.
People search using their own idioms, not translated versions of your English keywords. Query research has to be done in the target language, ideally by someone who speaks it, or you will build pages for phrases nobody uses.
- Do not auto-redirect by location.
Sending visitors to a locale based on IP prevents crawlers from seeing other versions and frustrates people who wanted the one they asked for. Offer a suggestion they can decline instead.
What Actually Determines Whether a Market Works
Rarely the structure. Markets fail because nobody local was involved, because the offer did not fit, or because the content was translated rather than written for the market. The domain decision gets the attention because it is the part that feels technical.
Local corroboration matters more than it does at home, because you arrive with no reputation. Presence on the review platforms and directories people in that market actually use is what makes you credible there, and it is the same mechanism described in earning third-party citations, simply harder because you are starting from nothing.
Give it a realistic horizon too. A new locale is a new site as far as accumulated evidence is concerned, even in a subfolder, so the timescales resemble those in the first 90 days more than they resemble adding pages to an established market. Judging a market entry at month three is judging it during the part where nothing is supposed to have happened yet.
Questions People Ask About International SEO
- Should I use a ccTLD, a subdomain, or a subfolder?
Subfolders for most businesses, because they keep everything on one domain that has already accumulated signals, and a new market starts from that base rather than from zero. Country domains give the strongest local signal and require you to build authority separately for each one. Subdomains sit awkwardly between the two, taking on much of the separation without the local trust a country domain buys.
- Does a country domain rank better in that country?
It carries a clear geographic association, which helps, and it starts with no accumulated history, which hurts. For a large brand with a real local operation and the budget to run each market properly, the trade is usually worth it. For a growing business testing a market, starting from zero in exchange for a signal you can approximate other ways is a poor bargain.
- Do I need to translate everything before launching a market?
No, and trying to is how expansions stall. Translate the pages that carry the decision first: the product or service pages, pricing, and the two or three questions every buyer asks. A market entry with fifteen genuinely localised pages outperforms one with two hundred machine-translated pages nobody checked.
- Is machine translation acceptable?
As a starting point that a fluent human then edits, yes. Published without review it produces text that is technically correct and reads as foreign, which damages trust in exactly the market you are trying to enter. The cost of review is the cost of the market entry; if you cannot afford it, the market is not ready.
- What is the cheapest way to test a new market?
A small set of localised pages in a subfolder, using the language and currency that market expects, before committing to any structure change. That gives you real demand data from the market itself rather than a projection, and if it works you have already built on the structure you would have chosen anyway.
