The 12 International SEO Mistakes We See Most Often

Technical targeting mistakes

1. Broken or missing hreflang

Hreflang is the most technically demanding part of international SEO and the most frequently misconfigured. One error can invalidate an entire annotation set. The failures we find most often:

• No self-referencing annotation. Every page in a set must point at itself, or the whole set can be ignored.
• Annotations that are not reciprocal. If your English page declares a French equivalent, the French page must declare the English one back. One-sided declarations are ignored.
• Invalid ISO codes. en-UK is not valid; en-GB is. This one is insidious rather than obvious, because Google ignores the part of the annotation using a reserved code, so en-UK degrades silently to a generic en. The language code must come first, and a country code alone is not valid.
• Relative URLs. Alternates must be fully qualified, including the protocol. Pointing at HTTP URLs from an HTTPS site is the same error.
• Annotations pointing at redirects or noindexed pages. If the target redirects or carries a noindex directive, the relationship is weakened or dropped.
• Canonicalising across language versions. The error most likely to undo otherwise careful work. Every page needs a self-referential canonical in its own language, or the closest substitute.

The myth: declaring hreflang in two places is a conflict

It is not. Google treats the three implementation methods, page head, HTTP header and XML sitemap, as equivalent, and states there is no benefit in using more than one. Pick a single method because maintaining three is harder, not because combining them causes a technical conflict.

The myth: Search Console will flag your hreflang errors

It will not. Google’s Page Indexing documentation states that alternate language pages are not detected by Search Console, and the URL Inspection tool does not address hreflang at all. The International Targeting report was the monitor and it was deprecated with no replacement. Audit with a crawler that maps the full annotation graph, such as Screaming Frog or Sitebulb, and use the Merkle hreflang tester or Aleyda Solis’s generator for individual pages.

2. IP-based redirects, and the header-sniffing version of the same mistake

Sending visitors to a regional version based on their IP address feels helpful and creates a serious crawl problem. Google advises against automatically redirecting users between language versions, because it can prevent both users and search engines from seeing all of them, and cautions that IP location analysis is unreliable and may stop it crawling your variations properly.

The mechanism is simple. Google crawls predominantly from United States IP addresses and does not vary the crawler source for a single site to find page variations. A site redirecting US IPs to its US version will show Googlebot that version and little else, so your other market pages are never discovered. Teams who fix this by switching to Accept-Language sniffing have the same problem in a less obvious form: the crawler sends requests without setting that header at all. One nuance worth keeping honest: Google notes it may crawl from other countries where it detects a site blocking US requests, so treat this as a high-probability failure rather than a certainty, and test rather than assume.

• The fix: give every version its own URL, use hreflang to signal which serves which audience, and let people choose. A soft banner suggesting the local version is fine, and a cookie may remember that choice between distinct URLs. It should never swap the content served at a single URL.

3. Inconsistent URL structures across markets

This is almost always organisational rather than technical. It appears when regional teams have each made their own decision over the years, so France ends up on a ccTLD, the Netherlands on a subdomain, and Poland in a subfolder. Each choice was defensible alone. Together they fragment authority.

For most expansion, subfolders are the sensible default because authority compounds rather than fragmenting. ccTLDs earn their cost where local trust is decisive or you genuinely run separate country operations. Whichever you choose, choose once: changing architecture after launch means redirects, re-indexing, and no guarantee of full recovery.

4. No x-default fallback

When no annotation matches a visitor’s language or country, Google picks a URL using other signals, and it may not be the one you would have chosen. Adding hreflang=”x-default” gives it an explicit fallback. Google notes that although x-default works on any page, it was designed for language selector pages, so a selector page is the stronger choice and a global homepage a reasonable substitute. This matters most on large sites, where visitors from unannotated markets still arrive.

Localisation mistakes

5. Publishing machine translation without native review

The myth: machine-translated content is spam

It is not, and the wrong version of this advice is everywhere. Google’s scaled content abuse policy concerns value rather than production method, and applies regardless of how content was created. Google has stated its policies do not strictly define AI-translated content as spam, and it removed its older advice about blocking automatically translated pages in June 2025. Hand translation is not a safe harbour either if the output is thin.

So unreviewed machine translation is a quality problem, not a policy violation. The practical issues:

• The keyword targeting reflects the queries the source page was written for, not what local users search. This is the strongest objection, and better translation does not fix it.
• Unnatural phrasing costs readers and conversions. That cost is commercial rather than algorithmic: Google has consistently denied using engagement metrics such as bounce rate as ranking signals.
• It produces recognisable sentence patterns that native readers spot immediately, undermining trust before anyone reads your offer.

Machine translation with native-speaker post-editing is a reasonable default. Publishing raw output at scale with no local keyword work behind it is the mistake. That gap is the difference between translation and localisation.

6. Regional variants that are near-identical

Publishing the same content for British and American English without meaningful differentiation is the case hreflang cannot rescue. This is not a penalty: Bing states plainly that duplicate content does not trigger search penalties on its own. What happens instead is consolidation. Google commonly picks one near-identical same-language variant as canonical and serves that, so one market quietly loses its page.

Bing has documented an AI dimension Google has published nothing equivalent on: large language models cluster near-duplicate URLs and pick one page to represent the set, so the model may select a version that is outdated or not the one you intended. Treat that as Bing’s account of its own systems rather than a universal rule.

If you target several English-speaking markets, the variants need real differences: pricing, regulatory context, availability, local examples. Where none of those actually differ, one page beats two thin ones.

7. Localisation that stops at the body copy

Translating the main content and leaving everything else in the source language produces a page that reads as imported, which costs trust at exactly the moment you are trying to earn it. What gets missed:

Alt text belongs on that list too, and it is almost always the last thing anyone remembers.

Authority and local signal mistakes

8. Building links only from home-market domains

Links from your home market do not establish authority abroad. A site with several thousand referring domains, all British, looks strong domestically and weak in Germany, and both search engines and AI systems know the difference. Earning authority in a new market means earning it there: local trade publications, industry associations, regional partners, and original research published in the local language. None of it is fast, and little can be outsourced cheaply.

• What does not work: link schemes targeting country-coded directories, mass outreach in a language nobody on your team speaks, and translated press releases, which announce themselves as imported the moment a local editor reads them.

9. Leaving the cheap local signals switched off

Google documents four signal groups for working out which locale a page is for: country-code top-level domains; hreflang in tags, headers or sitemaps; server location, which it notes is not definitive given how CDNs work; and a fourth group covering local addresses and phone numbers, local language and currency, links from local sites, and Business Profile signals. That fourth group is cheap and the one most often ignored, because it does not feel like SEO. A real local address, phone number, and currency are targeting signals, not just trust signals.

The myth: check your country targeting setting in Search Console

You cannot. Google removed that setting in September 2022, so it cannot be misconfigured. If a guide tells you to check it, that guide predates the removal, which tells you something about the rest of its advice. There is no markup substitute either: Google ignores locational meta tags such as geo.position and geotargeting HTML attributes.

Strategy mistakes

10. Targeting languages when your commercial reality is country-specific

Country targeting treats the UK, Germany, and Japan as discrete units, and is right when local context genuinely differs: pricing, regulation, availability, currency, payment methods. Language targeting treats English, German and Japanese as the unit wherever the reader sits, and is right when the offer is identical and only language varies. Most SaaS sits in the second camp.

The mistake is not picking one over the other. It is arriving at the answer by accident. A single Spanish page serving both Spain and Latin America is sound for a software product and poor for an ecommerce brand with different pricing, payment methods and consumer law in each. Decide deliberately, and write the decision down.

11. Translating the keyword list instead of researching the market

Translate intent, not keywords. A list converted word for word fails twice over: volumes diverge, so a term worth chasing in the UK may be worth nothing in Germany, and native speakers do not search the way translation tools assume.

This holds within a single language: German buyers in Germany, Austria and Switzerland use different terms for the same products, and Spanish diverges enough between Spain and Mexico that one strategy cannot serve both. The reliable workflow starts with intent. Establish what the user is trying to accomplish, work with a native-speaker specialist to find the phrases that intent surfaces in, validate volumes against country-segmented data, then read the pages already ranking there. If their headings consistently use different phrasing from yours, trust the results page over the translation.

12. Assuming Google is the only engine that matters

Google’s share is dominant globally but not universal. Baidu leads in China and requires Chinese-language content, mainland hosting or a credible CDN arrangement, and an ICP licence that itself requires a Chinese business entity. Yandex holds a significant share in Russia and the CIS. Naver dominates South Korea, blending search, blog, forum, and shopping results in a way that rewards being present across its properties.

There is now a second version of this mistake: assuming Google is the only surface that matters. AI assistants select and cite sources differently from classical ranking, and their training data skews heavily towards English, so a brand with a strong English presence and a thin local-language footprint is routinely described accurately in English answers and poorly, or not at all, elsewhere. Our piece on SEO, GEO, and AEO sets out how the surfaces differ, and our look at how AI Overviews are changing global search covers what that means commercially.

Which of these apply to you? A step-by-step checklist

Which mistakes are present depends on three things: the markets and languages you target, the platform you run on, and the URL structure you chose.

Your markets and languages

Your platform and URL structure

The remediation order

Fix in this sequence. The first three cause the most damage for the least effort, and everything after depends on the pages being crawlable.

1. Test for IP redirects and Accept-Language sniffing, including at the CDN or edge layer. If Googlebot is being redirected, nothing else matters.
2. Crawl the full hreflang annotation graph and check self-references, reciprocity and ISO codes.
3. Fix canonicals pointing across language versions. Every page needs a self-referential canonical in its own language.
4. Add x-default, ideally on a language selector page.
5. Confirm indexation coverage per market. Missing pages are a discovery problem, not a ranking one.
6. Audit whether regional variants are genuinely differentiated, then merge them or differentiate them properly.
7. Redo keyword research natively for your highest-value pages, starting with those that convert at home.
8. Localise metadata, interface, and trust signals on those same pages, not just the body copy.
9. Switch on the cheap local signals: local address, phone number, currency, and links from local sites.
10. Start local authority building in one market, and measure for three to six months before scaling.

If you would rather have specialists run the diagnosis and remediation for you, our international SEO agency team does exactly this across every market and language a brand operates in. So, if your international SEO is showing several of the mistakes we’ve discussed, then get in touch today to have our team diagnose your setup and prioritise the fixes.