AI Mode Adds Hindi, Japanese, Korean and More Languages

If your customers search Google in Hindi, Indonesian, Japanese, Korean or Brazilian Portuguese, AI Mode now speaks their language. On 8 September 2025 Google added these five AI mode languages, so people can ask long, complex questions in their own language and get an AI answer with links to supporting pages.

For site owners in those markets, this is the moment AI Mode stops being an English-only experiment and starts affecting your audience. Below I cover what launched, how your pages get included, how it shows up in Search Console, and what I would do this week.

What Google Launched on 8 September 2025

Google announced the change in a post on its Search blog by Hema Budaraju, Vice President of Product Management for Search. The headline is simple: AI Mode now works in five more languages, and the rollout started the same day for users around the world.

• Hindi
• Indonesian
• Japanese
• Korean
• Brazilian Portuguese

Google says AI Mode uses a custom version of Gemini 2.5 (its AI model) for these languages. The point is that the model can understand local phrasing and nuance, not just translate an English answer. You can read the short announcement on the Google blog.

People can reach AI Mode through its own tab in Search results or at google.com/ai. They can ask by typing, by voice, or by taking or uploading a photo, and then ask follow-up questions in the same conversation.

How This Fits the Wider AI Mode Rollout

Until now, AI Mode was English-only. It started as a Labs experiment in the US in March 2025, opened to everyone in the US in May, then reached India in June and the UK in July.

In August Google took it to more than 180 new countries and territories, but still only in English. I covered that step in AI Mode Expands to 180 Countries. The September update is the first time AI Mode understands and answers in languages other than English. AI Overviews went through a similar step in 2024, which I covered in AI Overviews expanding to more than 100 countries.

That matters a lot for markets like Japan, Korea, Indonesia and Brazil, where most people simply do not search in English. For them, AI Mode was technically present but practically irrelevant. Now it is usable.

Why the AI Mode Languages Update Matters for Your Site

Each new language opens AI Mode to a very large group of searchers who could not really use it before. If your business sells to people in India, Indonesia, Japan, Korea or Brazil, some of your future visitors will now meet your brand inside an AI answer first.

AI Mode does not show ten blue links. It writes an answer and places links to supporting pages inside and beside it. To build that answer, Google uses a technique it calls query fan-out: it splits the question into smaller sub-questions and runs many related searches at once.

In practice this means three things for you:

• Longer questions: people tend to type full, detailed questions into AI Mode, not two-word keywords. Your content has to answer specific needs, not just match a head term.
• More pages in play: because of fan-out, a page can be pulled in for a sub-question even if it would not rank on page one for the original query.
• Fewer, more qualified clicks: some questions get fully answered on Google. Google says clicks from its AI features tend to be higher quality, with people spending more time on the site.

My honest view: in these markets, the share of searches going through AI Mode will start small. But this is the best time to get your content right, before habits settle and before competitors notice.

Eligibility: What Your Pages Need to Appear in AI Mode

The good news is that there is nothing new to install. Google’s documentation on AI features and your website is clear on the basics.

1. The page must be indexed. If Google has not indexed it, it cannot be a supporting link.
2. The page must be eligible to show with a snippet. A snippet is the short text preview under a search result. If you block snippets, you block AI Mode too.
3. The page must meet Google’s normal Search technical requirements. Google says there are no extra technical requirements for AI features.

Google also says you do not need any special schema markup, AI text files or new machine-readable files. Standard SEO fundamentals apply. Meeting the requirements does not guarantee your page will be used, just as it does not guarantee a ranking.

The Controls That Affect AI Mode

If you want to limit what AI Mode shows from your pages, you use the same controls as regular Search:

• nosnippet: stops Google showing any snippet from the page, which also keeps it out of AI answers.
• data-nosnippet: an attribute you add to a specific section of HTML so that section is not used in snippets.
• max-snippet: sets a maximum length for snippets.
• noindex: removes the page from Search entirely.

One mistake I see often is people thinking the Google-Extended token in robots.txt controls AI Mode. It does not. It covers use in other Google AI systems, not Search. Blocking it will not remove you from AI Mode, and allowing it will not get you in.

In my experience, most businesses should not restrict snippets at all. Every one of these controls also reduces how you appear in normal results, so the cost is usually bigger than the benefit.

Local-Language Content: What Actually Helps

When someone asks AI Mode a question in Japanese, they expect an answer in Japanese. Google has not published rules on how AI Mode picks supporting links across languages. But in my experience, Google tends to favour pages that match the searcher’s language and region, and I would plan on that being true here too.

So if you serve these markets with English pages only, I would treat this update as a push to fix that.

Write for the Market, Not Just in the Language

A straight translation of an English page is rarely enough. AI Mode is built to handle detailed, local questions, so your page needs local detail to be useful for a sub-question.

• Use local currency, units, payment methods and delivery details.
• Answer the questions people in that market actually ask, which are often different from your English FAQ.
• Use local examples, place names and terms people search with.
• Get a native speaker to review anything produced by machine translation.

Google’s spam policies target content mass-produced mainly to manipulate rankings. Automated translation published at scale with no human review can fall into that trap. I’d rather publish 20 well-reviewed pages in Korean than 2,000 rough ones.

Market Notes for Each Language

Hindi. Many people in India write Hindi in the Latin alphabet, not in Devanagari script. When I research Hindi keywords, I check both forms, because the questions can look very different. India also had English AI Mode since June, so you may already see a mix of English and Hindi queries for the same topic.

Indonesian. Indonesian searchers often mix in English terms, especially for tech and finance. Check which words people really use before you pick your headings.

Japanese. Japanese mixes several writing systems, so the same idea can be written in more than one way. Make sure your key pages use the terms your audience uses, and keep important content in plain text, not inside images.

Korean. Google is not the only big search engine in Korea, so check how much of your Korean traffic comes from Google before you shift your plans. For the part that does, the same rules apply: clear Korean text, local details and crawlable pages.

Brazilian Portuguese. Google named Brazilian Portuguese specifically. Brazilian and European Portuguese differ in spelling and vocabulary, so if you serve both, keep separate versions instead of one page for both audiences.

Get the Technical Side of Multilingual Sites Right

If you run the same content in several languages, Google needs to understand which version is for whom. These are the checks I make first:

• One language per URL. Do not switch languages with cookies or browser settings on the same URL. Googlebot will usually see only one version.
• hreflang tags. These are tags that tell Google which language and region each page is for, such as hi-IN, ja-JP, ko-KR, id-ID and pt-BR. Each version should list all the others and itself.
• Clear structure. Use subfolders or subdomains such as /ja/ or /pt-br/, and link the language versions to each other. My guide to site architecture for SEO covers how to plan this.
• Translated titles and descriptions. Many sites translate the body but forget the title tag, meta description, image alt text and structured data.

Here is a simple hreflang example for a page that exists in English, Japanese and Brazilian Portuguese:

<link rel="alternate" hreflang="en" href="https://example.com/guide/" />
<link rel="alternate" hreflang="ja-JP" href="https://example.com/ja/guide/" />
<link rel="alternate" hreflang="pt-BR" href="https://example.com/pt-br/guide/" />
<link rel="alternate" hreflang="x-default" href="https://example.com/guide/" />

The same block goes on all three pages. The x-default line tells Google which page to show when no version matches the searcher.

How AI Mode Shows Up in Search Console

This is where many site owners get confused. AI Mode data is in Search Console, but you cannot see it on its own.

• Where it lives: AI Mode clicks and impressions are included in the Performance report under the “Web” search type, mixed in with regular results.
• Clicks: Google’s help docs say a click on a link to an external page in AI Mode counts as a click.
• Impressions: standard impression rules apply, so a link generally needs to be shown to the user.
• Position: AI Mode follows the same position rules as a normal results page.
• Follow-up questions: Google treats a follow-up as a new query, so its impressions, clicks and position belong to that new query.
• No filter: there is no AI Mode filter or search appearance option, and Search Labs experiments are not included.

So you will not get a neat “AI Mode in Hindi” line. You have to look for signs in the data instead.

Ways to Spot the Change in Your Data

Here is what I check for each market:

1. Filter by country. In the Performance report, add a Country filter for India, Indonesia, Japan, South Korea or Brazil and compare the weeks before and after 8 September 2025.
2. Filter by script. Search Console lets you filter queries with a custom regex (a pattern that matches text). This lets you split queries by writing system, which is a quick way to separate local-language searches from English ones.
3. Look for long queries. AI Mode questions tend to be longer. A rise in impressions for long, question-style queries is a useful signal.
4. Watch click-through rate. If impressions rise for local-language queries while clicks stay flat, AI answers may be satisfying some searches on Google.

These regex filters work in the Query filter (choose “Custom (regex)”):

p{Devanagari}            Hindi in Devanagari script
p{Hangul}                Korean
p{Hiragana}|p{Katakana}  Japanese kana
^(S+s+){7,}S+$         queries of eight words or more

Japanese and Korean do not always use spaces between words, so the word-count pattern works best for Hindi, Indonesian and Portuguese. For deeper analysis across many months, I prefer the Search Console bulk export to BigQuery, because the interface keeps only 16 months of data and shows at most 1,000 rows per table.

Give it time. A rollout of this size takes a while to show in user behaviour, and one week of data proves very little.

Practical Steps for Sites in These Markets

Here is the order I would work in if I had a site with audiences in any of the five languages.

1. Record a baseline. Export clicks, impressions, CTR and position by country and by script before the new data builds up. You cannot measure change without a starting point.
2. Check indexing and snippets. Use URL Inspection on your key local-language pages. Look for stray noindex, nosnippet or very short max-snippet values that a plugin or template may have added.
3. Check crawl access. Make sure robots.txt, your CDN and any bot protection let Googlebot reach the local-language folders. Firewall and bot protection rules sometimes block Googlebot by mistake.
4. Fix your hreflang and language URLs. Every language version should have its own URL, correct hreflang, and translated titles and meta descriptions.
5. Close your biggest content gaps. List the top questions customers ask in each language, then add clear answers to your most important pages. Start with pages that already earn impressions.
6. Update your local business data. Google advises keeping Merchant Center and Google Business Profile information up to date, because AI features can draw on it for shopping and local questions.

Steps 2 to 4 are the ones I see skipped most. They are also the cheapest to fix, and a broken hreflang setup can undo months of good translation work.

What I Would Not Do

There is already a lot of noise about “optimising for AI Mode”. A few things I would skip:

• Do not buy special “AI schema” or AI text files. Google says none are needed for AI Mode.
• Do not machine-translate your whole site overnight. Thin, unreviewed pages are more likely to hurt than help.
• Do not block snippets out of fear. You lose regular visibility, not just AI visibility.
• Do not rewrite everything as short Q&A blocks. Helpful, well-structured pages with clear headings and direct answers already work well.

Conclusion

The AI mode languages update on 8 September 2025 brought Google’s AI search to Hindi, Indonesian, Japanese, Korean and Brazilian Portuguese speakers worldwide. Your pages need nothing special to take part: they must be indexed, snippet-eligible and technically sound.

What makes the difference is good local-language content, correct hreflang, and a clear baseline in Search Console so you can measure what changes. My suggested next step is to export your by-country Search Console data today, then work through the six steps above, starting with indexing and hreflang.