On 12 November 2026, Google rolled out Gemini 4 Hydrogen to Search. It's now the default model in AI Mode and handles more complex questions in AI Overviews, for all users worldwide. Google's Robby Stein described it as the biggest model upgrade to Search yet.
For website owners, the short version is this: Google's AI answers just got better at long, messy, multi-part questions. That means more searches can be answered on the results page, and the pages that do get cited need to be the clearest, most specific source on their topic.
What actually changed
Based on Google's announcement and early coverage, the upgrade does four things:
| Change | What it means for searchers | What it means for you |
|---|---|---|
| Better at multi-part questions | "Best CRM for a 5-person agency that integrates with Xero and costs under £50" gets a direct answer | Pages that cover specific situations get pulled in |
| Better at typos and vague wording | Fewer failed searches | Exact-match keywords matter even less |
| More visual answers | Images, tables, comparisons and small tools inside AI Mode | Structured data, clear tables and good images get reused |
| Default in AI Mode | Every AI Mode answer uses the new model | AI Mode visibility now depends on this model's choices |
The same model is also available in the Gemini app, the Gemini API and Google AI Studio, so the way it reads and summarises pages will show up well beyond Search.
Why the timing confused everyone
Gemini 4 Hydrogen arrived two days after the November 2026 core update began. If your traffic changed that week, you need to work out which one did it.
A simple way to tell them apart in Search Console:
- Average position fell: most likely the core update.
- Average position held but clicks fell: most likely more answers being completed by AI Overviews or AI Mode.
- Impressions rose sharply for long, specific queries: AI Mode's query fan-out showing your page for sub-questions. That's a good sign even when clicks lag.
How AI Mode picks its sources
AI Mode doesn't search once. It breaks a question into several related searches, gathers results for each, then writes one answer. This is called query fan-out. A smarter model fans out more precisely, which rewards sites that cover a topic in depth across several connected pages.
Google has said its AI experiences use core ranking signals to choose what to cite. In practice, pages cited in AI answers almost always rank well organically for at least one of the fanned-out searches. There's no separate back door.
Seven things to do this month
- Answer the question in the first two sentences. A model summarising ten sources prefers the one that states the answer plainly. Put context and caveats after.
- Cover specific situations. "Best accounting software" is crowded. "Accounting software for UK landlords under Making Tax Digital" is the kind of multi-part question Hydrogen now handles well. Write for those.
- Use real tables. Comparisons, prices, specs and steps in proper HTML tables are easy to lift into visual answers.
- Add original information. Your own numbers, test results, photos and examples give the model a reason to cite you rather than the fifth page saying the same thing.
- Mark up what you can. Product, Organization, FAQ and Article structured data help Google understand the page. Our schema markup guide covers the basics and our free schema generator writes it for you.
- Build clusters, not one-offs. One pillar page and several supporting pages that link to each other match how fan-out works.
- Track AI referrals separately. See how to track traffic from ChatGPT, Gemini and Perplexity in GA4.
What not to do
- Don't stuff pages with FAQ blocks to "feed the AI". Write the questions people actually ask, with honest answers.
- Don't publish hundreds of thin variations for every long-tail question. That's exactly what Google's scaled content abuse policy targets.
- Don't block Google's crawler to avoid AI answers. Blocking Googlebot removes you from Search entirely. The Google-Extended token controls use of content for Gemini model training, not AI Overviews in Search.
Does this mean organic traffic is finished?
No, but its shape keeps changing. Simple factual questions are answered on the page. Searches where people need to compare, buy, book, hire or trust someone still send clicks, often better-qualified ones. Businesses that publish genuine expertise and make it easy to act (call, book, buy) are still winning traffic that converts.
If you want to see how visible your site is in AI answers today, try our free AI visibility checker. For a deeper look at the shift, read what AI visibility (GEO) means and our AI Mode SEO guide.
The short version
Gemini 4 Hydrogen is now the brain behind AI Mode and harder AI Overviews. It handles complex questions better, shows more tables and visuals, and still relies on core ranking signals to choose sources. Answer clearly and early, cover specific situations in depth, use tables and structured data, and add information nobody else has.




