SEO · 04
International SEO Consulting.
The wrong country's page ranks, sister sites compete, a language redirect hides pages. No multi-market setup is perfect. I find what sends searchers to the wrong version and specify the fix.
You hire me when
- A new country is next.
- Traffic lands on the wrong market.
- Sister shops compete.
- A language redirect or a language switcher was changed.
What you get
- Market and language URLs, hreflang, redirects.
- Traffic analysis per market and language.
- Expansion plans.
- Duplicate-content analysis across sister sites.
Why the wrong version ranks
Google does not see “your German site”. It sees URLs. Each URL is a page with a language, a market and a crawl history, and Google picks one for each searcher.
Hreflang tells Google which URLs are siblings. It works only when Google has crawled both, and both point to each other. Google says: “If two pages don’t both point to each other, the tags will be ignored.” (Google Search Central, Tell Google about localized versions)
So the tags can be correct and the result still wrong. My rule comes from my own whitepaper: “There is no perfect multi-language, multi-market SEO setup. Only an eternal chain of compromises.” (My whitepaper, translated from German.) The goal is to choose the compromise on purpose and to measure it.
What goes wrong in international SEO in 2026
These are the failures I measured in my 2025 and 2026 audits, and what I specify for each:
| What breaks | What I measured | What I specify |
|---|---|---|
| Right tags, wrong market | Dozens of versions, hreflang correct, under half of US impressions on the US version | Fix crawling first, delete versions nobody needs, differ the content |
| Google has not seen all versions | About 5 weeks to crawl the site once; 6 in 10 submitted pages not indexed | Crawl cycle under 14 days, redirects and junk URLs out |
| Two sites, one language, no hreflang | Almost 40 URL variants of one article, identical text on both sites | A full hreflang set, a self-canonical per site, one x-default |
| Language chosen by the request | One URL rendered a different language depending on the request | Language from the URL only |
| Language versions already earn the clicks | About 4 in 10 of about 90k clicks in 90 days on language versions | A migration gate: keep 90 percent of the clicks per wave |
| Canonical and hreflang contradict each other | A variant page: canonical to the main page, hreflang to itself | One rule per page type, tested on every release |
| The market segment comes after the second country | About 30k URLs would be re-mapped, each redirect costing 20 to 50 percent of its traffic | The market segment in the URL from day one |
The tags are right and Google still shows the wrong market
One site, dozens of market and language versions, hreflang spot-checked correct on the important pages. In 2025 I looked at where Google sent people, in two views.
- Site view: the version made for one market gets traffic from other markets.
- Market view: US traffic landed on an English version made for another market.
- US impressions in 2024: under half went to the US version. The rest went to the versions made for other markets.
The mechanism is a loop. Google crawls version A, sees B in the hreflang set, and queues B. B points back to A. If both pages look alike, Google may index only one and show it to everybody. Google’s doc says why: “Localized versions of a page are only considered duplicates if the main content of the page remains untranslated.” (Google Search Central) Same-language market versions are exactly that case.
There is no guarantee this loop completes. The worse the crawling, the less likely it does.
What I specify:
- Crawling and indexing first, see the next finding.
- Delete the versions you do not need: a 410, no links, and the old sitemaps gone after 35 days.
- Differ the content between versions of one language: other wording in the main content, and other phone numbers, addresses and calls to action around it.
- Link the siblings: the start pages link to every other version, and each page has a few visible links to pages of its own market and language.
- Weigh a domain per market. Each domain gets its own crawl budget, but it is much harder to run.
Google has not even seen all the versions
For hreflang to work, Google must crawl every version of a page. On the same site, in 2025:
- About 390k smartphone crawls in 90 days, which is about 4k a day.
- About 150k pages known to Google, so about 5 weeks to crawl the whole site once.
- Only about 30k pages in the sitemaps, and 8 in 10 known URLs in none of them.
- 6 in 10 submitted pages were not indexed. The top reason: “Duplicate, Google chose different canonical”.
- 1 in 7 crawled URLs was a redirect. Several page types were far over the HTML budget below.
My rule of thumb: under 14 days for a full crawl is good, over 30 days is a crawling problem. Here a fix cannot show faster than the crawl.
What I specify:
- Stop linking to redirects and to URLs with parameters. Link only to URLs that are in a sitemap.
- A page either exists (linked, in a sitemap, answers 200) or it does not (410). The sitemap is complete.
- 250 KB of unzipped HTML at most, 120 KB on average. More in Web Performance.
Two sites in one language, with no hreflang at all
Two country sites, same language. In a 2026 workshop audit I probed one article.
- Almost 40 URL variants of that article answered 200. Each carried a canonical to its own site, so Google merged them.
- Both sites carried the article with the same title, the same description and the same text, word for word.
- No hreflang on either site, in the pages I probed.
The parameter variants were handled. The two sites were not. For Google they were two unrelated sites fighting for the same queries.
Google’s advice for this case: “pick a preferred version and use the rel=“canonical” element and hreflang tags to make sure that the correct language or regional URL is served to searchers.” (Google Search Central, Managing multi-regional and multilingual sites)
What I specify:
- Same path on both sites, enforced by shared IDs.
- A full hreflang set on every page that exists on both sites: language plus market, each page listing itself.
- A self-canonical on each site, never a canonical across the two.
- One x-default, and tags only on pages that answer 200.
- A check before launch on a sample of pages, and a re-check after.
Google’s doc: “Each language version must list itself as well as all other language versions.” (Google Search Central)
One URL, a different language for different visitors
In a 2026 audit of a site with several languages, the same URL gave different pages depending on who asked.
- Depending on a cookie or a browser setting, the same URL returned another language. The language attribute and the title did not always agree.
- Googlebot sends neither, so it always sees one default version.
That is a second render path. It is not cloaking, but Google sees one version and visitors see another. Behind a cache, the wrong language can be served to everybody.
Google is clear: “Google recommends using different URLs for each language version of a page rather than using cookies or browser settings to adjust the content language on the page.” (Google Search Central) And on redirects: “Avoid automatically redirecting users from one language version of a site to a different language version of a site.” (Google Search Central) The reason: “These redirections could prevent users (and search engines) from viewing all the versions of your site.” (Google Search Central)
What I specify:
- Language comes from the URL only. No cookie and no browser header changes a byte of the HTML.
- A visible link to the other versions, and a suggestion banner instead of a redirect.
- A CI test that fetches the same URLs with and without a language header and demands a zero-byte diff.
The language versions are already the ranking pages
The same audit. Google’s Search Console, 90 days:
- About 4 in 10 of the clicks, about 40k of about 90k, landed on language versions of pages, not on the main URL.
- One language version alone: about 20k clicks.
- For about 3 in 4 of the pages of one template with impressions, a translated version existed.
I tested what Google does with the canonical. The page declared the main URL as canonical. Google indexed the translated version as its own canonical and ignored the declaration. The cause: the content was translated, so it was not a duplicate. The English alternate in the hreflang set pointed to a URL whose canonical was a different URL.
So the “messy” language URLs were not a defect to clean up. They were assets. Any change to them is a URL migration of tens of thousands of URLs.
What I specify:
- Step 0: language from the URL only, before any caching or path work.
- A canonical rule: a language URL is self-canonical only if the page really has that language. Hreflang lists exactly the published languages, plus an x-default.
- A gate per migration wave: 90 percent of the clicks of the old URLs must still arrive some weeks after the wave. Here that was a floor of about 35k clicks per 90 days. I budget 20 to 50 percent loss per redirected URL, so I move in waves, riskiest last.
Canonical and hreflang that say opposite things
In a 2026 analysis for a site with several languages I compared a main page with its variant page.
- The variant had a canonical to the main page.
- Its hreflang pointed to itself.
- The variant was thin, the main page rich.
A canonical says “index that one instead”. Hreflang says “I am the version for this language”. Both on one page contradict each other, and Google picks.
What I specify:
- One rule per page type: either the variant canonicals to the main page and carries no hreflang of its own, or it is self-canonical with a full set.
- A reciprocity test on every release: every alternate answers 200 and points back.
- The same checks for every new page type, before the first page is built.
Adding the market segment after the second country costs a migration
In one relaunch spec, the CMS vendor asked whether the new URLs needed a language path. On a single-market site I normally say no, because a segment means re-mapping every URL.
Here a full re-map was coming anyway: the URL schema was being rebuilt. The extra cost of a market segment was close to zero. Without it, a second country later forces a change of the URLs that already rank.
- About 30k URLs in the inventory.
- 20 to 50 percent of a URL’s traffic is at risk per redirect, so changing once is better than twice.
What I specify:
- Decide the market segment in the same step as the URL schema, before the design is signed off.
- Keep the CMS folder and the URL path separate. The CMS can hold a locale folder without every folder becoming a URL segment.
- The start page stays in the root, and the segment’s own root redirects to it.
- No location prompt, no IP-based switch of what Googlebot sees. Googlebot crawls from the US and must get the full page.
How I work
My method is in a whitepaper, written in German, with the examples for each setup. These are the steps I run:
- Inventory every version. Domains, subdomains, folders, languages, markets, sitemaps, and hreflang in the head, in headers and in sitemaps. In the 2025 audit that was dozens of versions.
- Look at Search Console twice: site view (which version gets which markets) and market view (where each market lands). I build the views per audit, with a PDF export for the team.
- Do the crawl arithmetic. Crawls per day, known pages, days per full crawl. If it is over 30 days, tags are not the first problem.
- Pick the setup per case. My rules, from the whitepaper:
- One language, one market: one domain, no special markup.
- Several languages, one market: one domain with a language folder, start page in the root. Over 1 million URLs sent to Google: a domain per language, because each gets its own crawl budget.
- Several markets, one language each: a domain per market. Markets drift apart fast, and separate domains give the most room to act.
- One language, several markets: separate domains, or subdomains if domains are not available. Folders are the riskiest: too many pages, and if Google cannot crawl all market versions, hreflang is void.
- Several markets and several languages: an English international domain, one domain per single-language market, folders for multi-language markets, plus correct hreflang. No perfect setup, only different degrees of pain.
- Write the spec per page type. URL, canonical, the full hreflang set, x-default, redirects, sitemap, and the language of navigation, footer and main content, which must be the same on every page.
- Test every release. Reciprocity, status 200, canonical, and the language for each request with and without a header. In one 2026 spec I asked for a generator that builds the alternate set from one list of published languages, so tags cannot drift.
- Re-measure per market after two weeks and after the crawl cycle has run once.
Two points where the whitepaper and Google’s current text differ, and what I do:
- Hreflang for translated sites. My whitepaper calls it fully optional when the languages differ. Google’s doc now lists a fully translated site as a case where indicating alternates is recommended. I set it on every translated site, and always on start and brand pages.
- What x-default does. In the whitepaper I describe it as the preferred market. Google: “The reserved x-default value is used when no other language/region matches the user’s browser setting.” (Google Search Central) I treat it as the fallback, ideally a language or country selector page.
If you are choosing a setup, read the whitepaper first: My whitepaper “Multimarkt & Multi Language SEO” (PDF, German, 9 pages). It carries the URL examples and the hreflang sets for each case.
This site: one language on purpose
This site is English only, on one domain. I checked the live home page: <html lang="en">, one canonical, no alternate links. It needs no hreflang, and I did not add any.
In my book I wrote the rule behind it: “Always start with a single domain!” (Understanding SEO, Multiple domains and languages) A site with several languages should still put each language in its own folder, except the start page, and submit one sitemap per language. The book says a single domain worked for sites with up to about 100,000 pages in total.
You can test the claim yourself: open the home page, view the source and search for hreflang. It is not there.
Questions I get before every expansion
Should we send visitors to their language automatically?
No. Google: “Avoid automatically redirecting users from one language version of a site to a different language version of a site.” (source) Show a banner and let the visitor choose. The language switcher must be real links.
Do we need hreflang if the languages differ?
Google says it is “usually best” to indicate your language and region pages. (source) My view: for same-language market versions it is not optional at all.
Does hreflang need the same domain?
No. Google: “Alternate URLs do not need to be in the same domain.” (source)
Folder, subdomain or a domain per market?
It depends on the case, and the rules are in step 4 above. For separate markets I recommend separate domains. They are harder to run, and each has its own crawl budget.
What goes into x-default?
A fallback for users no version matches. Google: “Consider adding a fallback page for unmatched languages, especially on language/country selectors or auto-redirecting home pages.” (source)
How long until the right page ranks?
Google does not document a time. In the 2025 audit a full crawl took about 5 weeks, so no fix could show faster than that. My inference: count in crawl cycles, not days.
When should we call you?
Before the new country is live, or when the first wrong-market page ranks. A team that wants to learn the method can take the International SEO Workshop.
Sources, read on 2026-11-12: Tell Google about localized versions of your page (last updated 2026-09-21) and Managing multi-regional and multilingual sites (last updated 2025-12-10), Google Search Central. My setup rules are in my whitepaper Multimarkt & Multi Language SEO, V1.1 (PDF, German, 9 pages). The client numbers come from my own audits, workshops and handovers in 2025 and 2026.
From real jobs
- 2026: two sites in one language, no hreflang on the pages I probed; almost 40 URL variants of one article.
- 2026: language chosen by the request on one URL; about 4 in 10 Google clicks already on language versions.
- 2026: canonical and hreflang that contradict each other on variant pages.
- 2026: a market segment in the URL before the second country; about 30k URLs re-mapped once.
- 2025: dozens of market and language versions, hreflang correct, and still under half of US impressions on the US version.
- 2023: the same page types tested from four markets.
- Since 2018: hreflang and multi-language specs for product and media sites.
Some of our customers
Examples, not the full list
Teams like yours
Some of our customersOnline shops
Webshops and marketplaces: category pages, product pages, speed.
ExamplesTravel and tourism
Marketplaces, operators, tourism boards: many markets, many languages.
ExamplesSoftware, SaaS and apps
SaaS and apps: JavaScript, docs, growth.
ExamplesIndustry and B2B
Manufacturers and B2B brands: products sold worldwide.
ExamplesSomething else in the way?
How consulting works- External SEO lead
- Relaunch and migration SEO
- Technical SEO
- Web performance
- SEO for newsrooms
- Content strategy and search demand
- Search data and traffic forensics
- SEO agency cleanup
- GEO and AI search
- Vibe coding training
- AI automation consulting
- AI-first transformation
- Custom tools for one job
- Design thinking
- Growth
- Growth strategy analysis
- SEO workshops and training
- Keynotes and talks
- Sparring and second opinion
- Dev, product and marketing mediator