Knowledge base v2.13.0 · Community edition · data through 2 October 2026
Day 2 · Thursday 1 October 2026 · 14:15
Language versions can also be language-plus-region variants: Google's example site had a generic English page (en), a UK English page (en-gb) and a Spanish page (es), each on its own URL.
Speaker GoogleEvidence slide photo, transcript
Used byrequirement DEV-INT-01
Google's hreflang example lists every language version in the page's HTML head as a link element with rel=alternate, the version's full URL and its hreflang code (en, en-gb, es), including the page's own URL.
Speaker GoogleEvidence slide photo, transcript
Used byrequirement DEV-INT-03glossary term hreflang
Missing return links are a common hreflang mistake: if page X names page Y as a language version, page Y must link back to page X.
“If page X links to page Y, page Y must link back to page X.”
Wording checked against the slide or recording
Speaker GoogleEvidence slide photo, transcript
Used byrequirement DEV-INT-03glossary term hreflang return link
hreflang language codes must be ISO 639-1 codes: Google's list of common mistakes marked se, dk and cz (the country codes of Sweden, Denmark and Czechia) as wrong and sv, da and cs (Swedish, Danish, Czech) as right.
Speaker GoogleEvidence slide photo, transcript
Used byrequirement DEV-INT-04
hreflang region codes must be ISO 3166-1 Alpha-2 country codes; frequent mistakes are UK instead of GB for the United Kingdom, SW instead of CH for Switzerland and GE instead of DE for Germany.
Speaker GoogleEvidence slide photo, transcript
Used byrequirement DEV-INT-04
EU is not a valid hreflang region code: sites use it hoping to target the European Union, but hreflang targeting works only by country.
Speaker GoogleEvidence slide photo, transcript
Used byrequirement DEV-INT-04
Google also flagged LA used as a region code for Los Angeles (a city, not a country) and SA used for South Africa as wrong hreflang region codes.
Speaker GoogleEvidence slide photo, transcript
Used byrequirement DEV-INT-04
A region code cannot be used on its own in hreflang: en and en-GB are valid values, but GB alone is not, because the value must name a language.
Speaker GoogleEvidence slide photo, transcript
Used byrequirement DEV-INT-04
Google may find a site's different language versions on its own, without the site owner doing anything, but owners can take steps to make this easier for Google.
Speaker GoogleEvidence transcript
Each language version should have its own URL; switching languages with cookies or by updating the content in place makes the versions very difficult for Google to handle.
Speaker GoogleEvidence transcript
Used byrequirement DEV-INT-01
Each page in an hreflang set must also list its own URL, and a missing self-reference is a common mistake Google sees.
Speaker GoogleEvidence transcript
Used byrequirement DEV-INT-03
When the HTML head is not suitable for a page, hreflang can be given by other methods instead, such as listing the language versions in a sitemap.
Speaker GoogleEvidence transcript
Used byrequirement DEV-INT-05
Any one hreflang method is fine, but a site should use only one: Google often sees sites using two or three methods at the same time, with annotations that conflict.
Speaker GoogleEvidence transcript
Used byrequirement DEV-INT-05
Google ignores hreflang annotations that point only one way, where page A names page B as a language version but page B does not name page A back.
Speaker GoogleEvidence transcript
Used byrequirement DEV-INT-03glossary term hreflang return link
Google requires hreflang return links because an alternate URL is a full URL that can be on any domain, so without them any site could declare itself a language version of someone else's website.
Speaker GoogleEvidence transcript
Used byrequirement DEV-INT-03
Swedish content for Sweden is marked sv-SE, not se-SE: the language part must be the Swedish language code sv, even though Sweden's country code is SE.
Speaker GoogleEvidence transcript, slide photo
Used byrequirement DEV-INT-04
Site owners do not need to memorise hreflang language and region codes, but should double-check every code when implementing hreflang.
Speaker GoogleEvidence transcript
Czech is a frequent hreflang error: CZ is the country code and ccTLD of Czechia, but the language code for Czech is cs.
Speaker GoogleEvidence transcript, slide photo
Used byrequirement DEV-INT-04
Google does not use hreflang to determine a page's language when indexing the page.
Speaker GoogleEvidence transcript
Used byrequirement DEV-INT-07glossary term hreflang
Google determines a page's language for indexing from the page content, not from a language code in the URL.
Speaker GoogleEvidence transcript
Used byrequirement DEV-INT-07
Google annotates each page with only one language in the index.
Speaker GoogleEvidence transcript
Pages that mix several languages make it very hard for Google to decide which language a page is in, so each page should make its target language obvious and avoid mixing languages.
Speaker GoogleEvidence transcript
Used byrequirement DEV-INT-07
Scattered boilerplate in a second language is sometimes fine but still not recommended, partly because mixed languages on one page are odd for users.
Speaker GoogleEvidence transcript
Used byrequirement DEV-INT-07
The ccTLD (country-code top-level domain) is one of the most important and strongest signals Google uses to decide which country a site targets; Google also considers other signals, which matter less.
Speaker GoogleEvidence transcript
Used byrequirement DEV-INT-09glossary term ccTLD
Server location is not really a reliable country-targeting signal nowadays, so Google does not use it much; the recording is unclear on the word 'server'.
Speaker GoogleEvidence transcript
For country versions of a site, a ccTLD, subdomains or subdirectories are all acceptable choices; the right one depends on the site's needs, goals and resources.
Speaker GoogleEvidence transcript
Used byrequirement DEV-INT-09
Of the URL structures for country versions in a table on Google's slide (not photographed), the speaker called only one a wrong choice: the last option, which the speaker did not name but said they really do not recommend.
Speaker GoogleEvidence transcript
A separate ccTLD for each country version is not necessarily better, even though the ccTLD is one of the strongest country signals, and sites should not move their domain for that reason.
“Don't try to move your domain now.”
Speaker GoogleEvidence transcript
Used byrequirement DEV-INT-09
Whether to use ccTLDs depends on whether the domain can be obtained, what it costs, local laws or regulations, and how committed the business is to expanding in that country.
Speaker GoogleEvidence transcript
Used byrequirement DEV-INT-09
Google usually understands non-English words typed 'in English' (probably meaning romanised, Latin-letter spellings, which the speaker did not spell out); the speaker said this belongs to query interpretation, a Day 3 (serving) subject.
Speaker GoogleEvidence transcript
Google usually understands a query word whether it is written with or without diacritics (accents).
Speaker GoogleEvidence transcript
Sites need no extra work to appear in AI Overviews and AI Mode: both work on top of the existing web results, so what works for web results also works for the AI features.
Speaker GoogleEvidence transcript
Used byrequirement DEV-AIF-01story angle A-001
In classic results users can usually tell when content in their language is missing, poorly translated or from another country, but AI answers synthesize many sources and hide this; the speaker called it an invisible gap.
“an invisible gap”
Speaker GoogleEvidence transcript
Users often assume AI is all-knowing and borderless, but AI is still language-dependent: if an AI answer is synthesized from the top results, a query in another language draws on a totally different set of data.
“AI is still language-dependent”
Speaker GoogleEvidence transcript
In the speaker's own experience, the differences between languages are larger in AI answers than in classic search results.
Speaker GoogleEvidence transcript
Site owners may want to research whether they show up in AI answers in each language and market they serve, to find blind spots.
Speaker GoogleEvidence transcript
Whether machine-translated content is acceptable depends on the case and is the site owner's decision, after weighing three things machine translation can miss: translation quality, local conventions and cultural adaptation.
Speaker GoogleEvidence transcript
Used byrequirement DEV-INT-10
To show uneven machine-translation quality, the speaker showed Google Translate output of attendee questions from a visit to China with a Google colleague, Gary, some clear and some unintelligible, and a restaurant menu whose English listed an item called 'aggressive element'.
Speaker GoogleEvidence transcript
Localisation should account for local conventions such as date formats, which differ between Europe, the UK, the US and other countries, and calendars (in Thailand the current year is 2569), or a date can point users to the wrong day.
Speaker GoogleEvidence transcript
Used byrequirements DEV-INT-10, DEV-SDA-05
According to the speaker, shoppers in Europe and the US pay a lot of attention to promotions and discounts when deciding to buy (no source for this was captured), one of the cultural factors localisation should take into account.
Speaker GoogleEvidence transcript
According to consumer data shown on a slide in the talk (source not captured), US consumers judge product quality more by user feedback and reviews, while European shoppers seem to look more at brand reputation.
Speaker GoogleEvidence transcript
According to consumer data shown on a slide in the talk (source not captured), German shoppers also rely heavily on expert recommendations and certifications when judging product quality.
Speaker GoogleEvidence transcript
According to consumer data shown on a slide in the talk (source not captured), French shoppers particularly want to know where a product comes from.
Speaker GoogleEvidence transcript
Google said that working out a site's localized pages and which countries it targets is one of the difficult, complex tasks for it in indexing.
Speaker GoogleEvidence transcript
Google said incorrect hreflang language or region codes are a mistake it finds very often, with the same wrong codes coming up again and again, especially in Europe, because site owners assume the codes are easy.
Speaker GoogleEvidence transcript
Used byrequirement DEV-INT-04
Google's documentation says Googlebot mostly crawls from US IP addresses and sends no Accept-Language header, so pages that change content or redirect by the visitor's perceived country or language may not have every version crawled, indexed or ranked; it recommends separate URLs annotated with hreflang.
Publisher Google Search Central
Used byrequirements DEV-INT-01, DEV-INT-02
Google's hreflang guide says language-specific subdomains such as en, en-gb or de in a URL are not used to determine a page's target audience; site owners must map the audience explicitly, for example with hreflang.
Publisher Google Search Central
Google's hreflang guide says its three methods (link elements in the HTML head, an HTTP Link header, which suits non-HTML files such as PDFs, and an XML sitemap) are equivalent; using several at once is allowed but brings no benefit in Search and is harder to manage.
Publisher Google Search Central
Used byrequirement DEV-INT-05
Google's hreflang guide says a site that finds it hard to keep every language pair bidirectional may omit some languages on some pages, because Google still processes the pairs that point to each other; new language versions should at least link both ways with the original or dominant language.
Publisher Google Search Central
Used byrequirement DEV-INT-03
Google's hreflang guide says alternate URLs must be fully qualified, including the protocol (https://example.com/foo, not //example.com/foo or /foo), and need not be on the same domain.
Publisher Google Search Central
Used byrequirement DEV-INT-03
Google's hreflang guide says pages that translate only the template (navigation, footer) around main content in one language still count as duplicates, because localized versions are duplicates only if the main content stays untranslated; it recommends hreflang for such pages.
Publisher Google Search Central
Used byrequirement DEV-INT-07
Google's documentation calls a ccTLD a strong signal that a site is meant for a certain country and still lists server location as a possible signal of a site's audience, though not a definitive one because sites use CDNs or are hosted abroad; it does not rank the signals.
Publisher Google Search Central
Used byglossary term ccTLD
Google's table of URL structures for country targeting gives pros and cons for a country-specific domain (example.de), a subdomain (de.example.com) and a subdirectory (example.com/de/) on a generic domain, and marks URL parameters (site.com?loc=de) as not recommended.
Publisher Google Search Central
Used byrequirements DEV-INT-01, DEV-INT-09
Google's Search Help says the language a query is typed in is an important factor in choosing the language of results, alongside interface and device language, location and site owners' annotations, and that Google may also show results in other languages when they are helpful or when there is not enough information in the query's language.
Publisher Google Search Help
Google's spam policies define scaled content abuse as generating many pages mainly to manipulate rankings, with little or no value to users, no matter how they are created, and list automated translating of scraped content among the examples.
Publisher Google Search Central
Used byrequirement DEV-INT-10glossary term Scaled content abuse
Google's hreflang guide says the ISO 639-1 language code can be followed by a script in ISO 15924 format, such as zh-Hant or zh-Hans, and by an optional ISO 3166-1 Alpha-2 region code, as in zh-Hans-US.
Publisher Google Search Central
Used byrequirement DEV-INT-04
Audit hreflang per cluster rather than per page: check that every page lists itself and all its alternates, that every alternate links back, and that only one method (HTML head, HTTP header or sitemap) supplies the annotations.
Author Ibrahim Anjro
Used byrequirement DEV-INT-03
Validate hreflang values against the ISO 639-1 and ISO 3166-1 Alpha-2 lists in a crawler rule or build check rather than by eye: reserved codes such as UK and EU are simply ignored, but some plausible mistakes are valid codes for something else (se is Northern Sami, GE Georgia, LA Laos, SA Saudi Arabia) and silently point the annotation at the wrong audience.
Author Ibrahim Anjro
Used byrequirement DEV-INT-04
Because Google reads a page's language from its content and stores one language per page, translate templates (navigation, footer, buttons, legal notices) together with the main text; large untranslated blocks risk the page being annotated with the wrong language.
Author Ibrahim Anjro
Used byrequirement DEV-INT-07
At the point where one recording names the ccTLD as the most important country-targeting signal, a second attendee recording heard 'IP address' instead; Google's multi-regional sites documentation names the ccTLD as a strong signal of a site's target country and server location only as a possible, not definitive one, so the ccTLD is the better-supported reading.
Author Ibrahim Anjro
For a business testing new countries, subdirectories on the existing domain are usually the cheapest start; a ccTLD's stronger country signal pays off only where domain availability, cost, local rules and long-term commitment justify it, and an established domain should not be moved just for the signal.
Author Ibrahim Anjro
Used byrequirement DEV-INT-09
Check AI Overviews and AI Mode with native-language queries in each target market, not with translated English keywords, and compare with the country breakdown of Search Console's generative AI performance report; topics where competitors are cited and you are not point to missing or weak local content.
Author Ibrahim Anjro
Used byrequirement DEV-MON-07
Treat machine translation as a first draft: have a native speaker review it and adapt dates, calendars and units before publishing, because unreviewed bulk translation that adds little value can also fall under Google's scaled content abuse policy.
Author Ibrahim Anjro
Used byrequirement DEV-INT-10
Localise the trust signals on product pages, not only the words: if the findings shown hold for your category, lead with reviews in the US, brand reputation elsewhere in Europe, expert certifications in Germany and origin information in France, and test the order per market.
Author Ibrahim Anjro
Sites need no extra work to appear in AI Overviews and AI Mode: both work on top of the existing web results, so what works for web results also works for the AI features.
AI Overviews and AI Mode are built on top of Search results: they are a different experience of the same content Google already has.
Users often assume AI is all-knowing and borderless, but AI is still language-dependent: if an AI answer is synthesized from the top results, a query in another language draws on a totally different set of data.
AI Mode and AI Overviews use the same crawling and the same index as Search. At serving they add grounding on the Search index and query fan-out.
Check AI Overviews and AI Mode with native-language queries in each target market, not with translated English keywords, and compare with the country breakdown of Search Console's generative AI performance report; topics where competitors are cited and you are not point to missing or weak local content.
Google's launch post for the Generative AI performance reports in Search Console (3 June 2026; rolled out to all sites worldwide by 31 August 2026) lists impressions, pages, countries, devices (Search only) and dates, and no click or query metrics.
Treat machine translation as a first draft: have a native speaker review it and adapt dates, calendars and units before publishing, because unreviewed bulk translation that adds little value can also fall under Google's scaled content abuse policy.
The answer is not a claim that Google detects AI text. It says ranking favours text that reads as natural to people. The risk with AI content is scale without value, which falls under Google's scaled content abuse policy, not the tool itself.
A community speaker said language versions need adjusting for regional variants of one language: 90 is 'quatre-vingt-dix' in France but 'nonante' in Belgium, and Swiss German differs from German, for example in whether the letter ß (Eszett) is used.
Language versions can also be language-plus-region variants: Google's example site had a generic English page (en), a UK English page (en-gb) and a Spanish page (es), each on its own URL.
Google detects the language of each document and weights language in index selection so that the index is not dominated by one or two languages.
Google determines a page's language for indexing from the page content, not from a language code in the URL.
Users of non-Latin-script languages do not always search in their own script: the same Persian query may be typed in Persian script or in Latin letters, with the same intent and the same expected results.
Google usually understands non-English words typed 'in English' (probably meaning romanised, Latin-letter spellings, which the speaker did not spell out); the speaker said this belongs to query interpretation, a Day 3 (serving) subject.
In the presenter's observation, AI answers to Persian queries are written in Persian but cite some English sources; the presenter explained that Persian content on a topic is often thinner, of lower quality or less relevant, so the systems retrieve from languages with better content, which the presenter called cross-lingual retrieval.
Users often assume AI is all-knowing and borderless, but AI is still language-dependent: if an AI answer is synthesized from the top results, a query in another language draws on a totally different set of data.
Google generally treats spellings with and without diacritics as synonyms behind the scenes, for example a German 'ü' written as 'ü', as 'ue' or left out.
Google usually understands a query word whether it is written with or without diacritics (accents).
Users expect content written the way they search: in some languages they search in Latin characters, in others in the local script, and Hindi users, for example, search both in Hindi and in Latin letters.
Google usually understands non-English words typed 'in English' (probably meaning romanised, Latin-letter spellings, which the speaker did not spell out); the speaker said this belongs to query interpretation, a Day 3 (serving) subject.
Some countries rely a lot on social proof such as reviews, Google's speaker said, recalling a point from Google's Day 2 internationalisation talk.
According to consumer data shown on a slide in the talk (source not captured), US consumers judge product quality more by user feedback and reviews, while European shoppers seem to look more at brand reputation.
The cheaper tokens become, the more AI slop is created, and Google counts AI slop as scaled content abuse.
Google's spam policies define scaled content abuse as generating many pages mainly to manipulate rankings, with little or no value to users, no matter how they are created, and list automated translating of scraped content among the examples.
Google's Search Quality Rater Guidelines point out that the tool used to create content is not the problem, but how it was used and what for.
Google's spam policies define scaled content abuse as generating many pages mainly to manipulate rankings, with little or no value to users, no matter how they are created, and list automated translating of scraped content among the examples.
Sites need no extra work to appear in AI Overviews and AI Mode: both work on top of the existing web results, so what works for web results also works for the AI features.
Google's answer to 'SEO is dead, long live GEO' is not to worry about the name: good SEO is good GEO and AEO.
Sites need no extra work to appear in AI Overviews and AI Mode: both work on top of the existing web results, so what works for web results also works for the AI features.
Three reasons were given: generative AI features are built directly on the core ranking systems, query fan-out expands the original query to find related information, and generative AI features highlight content indexed by Google Search.