Menu

What Gemini 4 Hydrogen Changes in Google Search, and What to Do About It

Nov 18, 2026

Author: Mike Rowan,

Google swapped the model behind AI Mode for every user in the world, two days into a core update. I think that it matters more for our clients than the core update that is getting most of the attention.

Key takeaways

  • On November 12, 2026, Google made Gemini 4 Hydrogen the default model in AI Mode for all users worldwide, and says that it is also used for the hardest questions in AI Overviews.
  • Google says that Hydrogen handles multi-part questions and typos better, runs a wider query fan-out that surfaces more links, and builds more visual answers with comparison tables and custom tools. It published no numbers to back that up.
  • The launch landed two days into the November 2026 core update and right after a Search Console reporting bug, so nobody can cleanly read their AI visibility data for this stretch yet.
  • Our view is that the fan-out change rewards sites that answer the sub-questions behind a topic, and that is where we would put the work.

Google doesn’t usually swap the engine behind a product with more than a billion monthly users in a two-minute blog post, but that is what happened on November 12, 2026.[1][6] Gemini 4 Hydrogen became the default model in AI Mode for every user worldwide that day, and Google called it the “biggest model upgrade to Search yet.”[1]

We spent the afternoon reading both of Google’s announcements and the trade coverage that followed.[3][5] A ranking update reshuffles which pages show up, and a model change rewrites how the answer gets built in the first place, which is why I think that this one deserves more of your attention than it is getting.

What Google announced

Hydrogen is the third model in the Gemini 4 family and the first one that everybody gets. Argon went to paying subscribers in October, Carbon went to Ultra subscribers on November 10, and Hydrogen arrived for all users two days later.[4][6] Google positions it as the fast, efficient model, and it takes the place of the Flash name that Google used for its fast models until now.[2][14]

Search isn’t the only place it landed. Hydrogen also went into the Gemini app, the Gemini API and Google’s developer tools on the same day.[2] Search is where it reaches the most people, though, as AI Mode alone passed a billion monthly users back in May.[6]

What Google says Hydrogen changes in Search
CapabilityWhat Google saysWhat it means for your site
Intent understandingHandles multi-part questions, hard-to-read typos and very specific requests, and addresses each partPeople can ask longer, messier questions and get one answer, so fewer follow-up searches happen
Query fan-out and linkingRuns simultaneous searches across the relevant subtopics and connects people to a broader range of linksMore of your visibility is decided on sub-questions that you may not track as keywords
Visual responsesMore images, structured layouts, comparison tables and custom toolsGoogle builds the comparison itself from facts and images that it can pull from pages and feeds
Language understandingMore nuanced responses in every language that AI Mode supportsNon-English markets get the upgrade at the same time, with no staged schedule announced

Source: Google’s Search announcement.[1] The right-hand column is our reading.

Robby Stein, the VP of Product for Search, added on X that Hydrogen is also “being used for your hardest questions in AI Overviews,” which the blog post itself doesn’t mention.[4] For example, I would expect a long comparison question to be routed to Hydrogen and a simple definition to stay on a lighter model, but Google hasn’t said where that line sits.

The Gemini 4 models in Search
ModelWho gets itRole
Gemini 4 HydrogenAll users, globally, since November 12Default in AI Mode, plus the hardest AI Overview questions
Gemini 4 CarbonGoogle AI Ultra subscribers, since November 10, with daily limitsDeeper reasoning for the toughest AI Mode questions
Gemini 4 ArgonGoogle AI Pro and Ultra subscribers, since October 13The first Gemini 4 model to reach AI Mode

Sources: Search Engine Journal and PPC Land.[4][6]

How the rollout fits together

The timing is what makes this launch hard to read. For example, three separate things happened to your AI visibility data inside four days, and each one would be enough to move a chart without the other two.

Showing all 9 events

DateWhat happened
January 2026ModelGemini 3 becomes the default model for AI Overviews globally[4]
May 19, 2026ModelGemini 3.5 Flash becomes the AI Mode default, and Google reports AI Mode passing one billion monthly users[4][6]
September 2, 2026ModelAI Mode moves to Gemini 3.8 Flash[6]
October 13, 2026ModelGemini 4 Argon reaches AI Mode for AI Pro and Ultra subscribers[4]
October 15, 2026LinksGoogle moves inline links higher in AI Mode and AI Overviews and tests a section for official sites[10]
October 28, 2026AdsAds in AI Mode expand to ten more countries[6]
November 9 to 11, 2026ReportingA logging bug under-counts impressions in Search Console’s Generative AI performance report[8]
November 10, 2026RankingModelThe November 2026 core update begins, and Gemini 4 Carbon starts rolling out to Ultra subscribers[4][7]
November 12, 2026ModelGemini 4 Hydrogen becomes the AI Mode default for all users worldwide[1][4]

Google’s guidance on core updates is to wait at least a full week after the rollout finishes before you analyze the impact.[7] I think that the same patience applies here, as any chart you pull this week has a reporting bug, a ranking update and a model change stacked on top of each other.

On the rollout itself, Google’s wording is that Hydrogen is available globally in Search for all users, in every language that AI Mode supports, starting on launch day.[1] Google didn’t announce a waitlist or a country-by-country schedule. That is a different pattern from Argon and Carbon, which were both held behind paid plans, and it follows a month in which Google also moved inline links higher in AI answers and widened ads in AI Mode.[10][6]

Why we think this matters

The first reason is that Hydrogen is the default. Carbon and Argon are interesting, but they sit behind a subscription. Hydrogen is what your customers get without doing anything, in every language that AI Mode supports.[1]

The second reason is the fan-out, and it is the one that I do think that SEO teams should focus on. Query fan-out is Google’s term for breaking one question into several related searches and running them at the same time. Google says that Hydrogen does this across all of the relevant subtopics of a question and that it surfaces sites people might not have found otherwise.[1]

One question

which crm is best for a 10 person agency, needs gmail sync and under $50 a user

Sub-queries that run at the same time

  • best CRM for small agencies
  • CRM Gmail integration
  • CRM pricing per user
  • CRM reviews from agencies

Passages pulled from many sites

A pricing page, an integration doc, a review roundup and a forum thread can each supply one piece.

One answer

A written recommendation with inline links, images and a comparison table.

An illustration by KPItarget of how query fan-out works. Google doesn’t publish the sub-queries that it runs, so the four shown here are our example.

The research from before this launch already showed citations drifting away from the classic top 10. For example, Ahrefs found in March that about 38% of the pages cited in AI Overviews also ranked in the top 10 for the same query, down from roughly 76% the previous July.[11] A Moz study of nearly 40,000 queries put the exact-match figure for AI Mode at 12%.[12]

Share of AI citations that also rank in the organic top 10

AI Overviews, July 2025 (Ahrefs)76%
AI Overviews, March 2026 (Ahrefs)38%
AI Mode, early 2026 (Moz)12%
Two different studies with different methods, both from before Hydrogen. Ahrefs notes that its improved citation detection may explain part of the drop.[11][12]

I wouldn’t lean on any single number in that chart. The direction is consistent, though, and a wider fan-out should push it further.

The third reason is the visual answers. If Google builds a comparison table for the searcher, your product needs to be a row in that table with the right facts in it. I think that this is good news for specialists, as a mid-size company that answers the ten real questions a buyer has about one service has a better shot at being cited for one of them than it ever had at outranking a national publisher on the head term. But with opportunities come problems, and the same change makes it easier to lose a citation that you didn’t know you had.

There is a paid media angle as well. Ads in AI Mode expanded to ten more countries on October 28, and existing Search, Shopping and Performance Max campaigns are eligible for them automatically.[6] If Hydrogen pulls more long, multi-part questions into AI Mode, I think that more of your paid impressions end up next to an AI answer too, which is one more reason for the SEO and paid teams to look at the same pages together.

What Google hasn’t told us

To be perfectly honest, almost everything we know about the quality of this change comes from Google. For example, the 2.5 times speed gain and the more than fivefold drop in fabricated claims are Google’s own results against Gemini 3.8 Flash, and nobody outside the company has reproduced them yet.[2][6]

Open questions as of November 12
Open questionWhy it matters
How many more links appear“More links” is Google’s claim. Neither the blog post nor Stein’s thread gives a figure.[4]
Which AI Overview questions count as hardestIt decides how much of your AI Overview tracking is affected.
Which model Hydrogen replaced in SearchGoogle’s posts don’t say. PPC Land reports that AI Mode had been on Gemini 3.8 Flash since September.[6]
Whether the rollout is complete everywhereGoogle said global and gave no country-by-country schedule.
What happens to clicksNo click, referral or traffic data was published.

How to optimize for it

We don’t think that this calls for a new playbook. Google’s own guide says that optimizing for its AI features is still SEO, and it specifically says that you don’t need llms.txt files, special schema or content chopped into chunks.[9] What we would change is where the effort goes, and these are the five things we are doing for clients first.

What to doWhy Hydrogen makes it matterHow to verify
1. Annotate November 9 to 12 in every dashboard and hold your read until a week after the core update finishesThree overlapping events make this week’s data unreadableThe Google Search Status Dashboard shows the update as complete
2. Map the sub-questions behind each money page and answer the missing ones on that pageA wider fan-out decides more citations at the sub-question levelEach sub-question has a direct answer under a clear heading on an indexed page
3. Put comparison facts in real HTML tables, add original images, and keep Merchant Center and Business Profile currentHydrogen answers use more comparison tables and imagesView the page as text and confirm that the specs, prices and alt text are there
4. Confirm that key pages are indexed, eligible for a snippet and included in Search generative AI featuresGoogle lists all three as requirements for appearing in AI features[9]URL Inspection in Search Console, plus the generative AI setting for the property
5. Re-run your AI visibility prompts with longer, multi-part questionsAI Mode queries run about three times the length of a classic search[6]Compare citations and mentions against your baseline from before November 9

One caution on the second item. Google says that creating a separate page for every fan-out variation violates its scaled content abuse policy.[9] For example, a service page that is missing pricing context should get a pricing section, and it shouldn’t get nine thin sibling pages. RESONEO’s research on what AI Mode reads points the same way, as it found that a page enters an answer through the passage that answers the question.[13]

On the third item, the test that we use is whether a buyer would compare you on it. If they would, it should be on the page as text, in a table where a table fits, with a real photo next to it. Google’s guide recommends supporting text with high-quality images and video, and it points to Merchant Center feeds and Business Profiles for commerce and local queries.[9]

Map the sub-questions for one topic

Type a product or service, then tick each question that your site already answers directly. These are audit prompts written by KPItarget. Google doesn’t publish its real fan-out queries.

0 of 10

Tick the questions that already have a direct, visible answer on your site.

Find your first clean comparison week

Enter the date that Google marks the core update as complete. The default is Google’s two-week estimate, so update it when the real date is known.

Clean “before” week
November 2 to November 8, 2026
Compare it with
November 25 to December 1, 2026
Earliest day to pull the report
December 3, 2026

The “before” week ends the day before the reporting bug started. The read date allows two days for Search Console data to fill in.

Annotations to paste into GA4 and Looker Studio

Nov 9 to 11, 2026: Search Console Generative AI report logging bug (impressions under-counted)
Nov 10, 2026: Google November 2026 core update began
Nov 12, 2026: Gemini 4 Hydrogen became the default model in AI Mode

If you only do one thing this week, add those three annotations and save your last prompt-tracking run from before November 9. Search Console will keep its history, but a record of what AI Mode said about you before Hydrogen can’t be recreated later.

Frequently asked questions

What is Gemini 4 Hydrogen?

Gemini 4 Hydrogen is Google’s fast, cost-efficient model in the Gemini 4 family, released on November 12, 2026. It is the default model in AI Mode for all users, and Google says that it is also used for the hardest questions in AI Overviews.

Does Gemini 4 Hydrogen change organic rankings?

Google hasn’t said that it does. Hydrogen is the model that writes AI Mode answers and some AI Overviews. Ranking changes this month are more likely tied to the November 2026 core update, which began on November 10 and is a separate event.

Do I need llms.txt or special schema to show up in Hydrogen answers?

No. Google’s guide to optimizing for generative AI features says that Search doesn’t use llms.txt files or special markup, and that structured data isn’t required for AI features, although it still helps with rich results.

When can I trust my AI visibility data again?

Google recommends waiting at least a full week after a core update finishes before analyzing its impact. If the update takes the full two weeks, the first clean comparison week ends around December 1. Compare against a week before November 9, as the Generative AI performance report under-counted impressions from November 9 to 11.

Will Hydrogen send more traffic to websites?

Nobody knows yet. Google says that the improved fan-out connects people to a broader range of links, but it published no click or referral data. Measure your own citations and referrals before and after November 12.

Sources

  1. Robby Stein, Google, “Google Search gets a major upgrade with Gemini 4 Hydrogen”, November 12, 2026. blog.google/products-and-platforms/products/search/gemini-4-hydrogen-search-upgrade
  2. Koray Kavukcuoglu, Google, “Introducing Gemini 4 Hydrogen”, November 12, 2026. blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-hydrogen
  3. Search Engine Land, “Google Search gets major upgrade with Gemini 4 Hydrogen”, November 12, 2026. searchengineland.com/google-search-gets-major-upgrade-with-gemini-4-hydrogen-501867
  4. Matt G. Southern, Search Engine Journal, “Google Makes Gemini 4 Hydrogen The Default Model In AI Mode”, November 12, 2026. searchenginejournal.com/google-gemini-4-hydrogen-default-ai-mode/597096
  5. Barry Schwartz, Search Engine Roundtable, “Google Search Gets Massive Upgrade With Gemini 4 Hydrogen”, November 12, 2026. seroundtable.com/google-search-gemini-4-hydrogen-42526.html
  6. PPC Land, “Google Search gains Gemini 4 Hydrogen for all users, not just subscribers”, November 12, 2026. ppc.land/google-search-gains-gemini-4-hydrogen-for-all-users-not-just-subscribers
  7. Search Engine Watch, “Google’s November core update could reshuffle your search rankings”, November 2026. searchenginewatch.com/google-november-2026-core-update
  8. Search Engine Land, “Google Search Console performance report data bug Nov 9 -11”, November 12, 2026. searchengineland.com/google-search-console-performance-report-data-bug-nov-9-11-501899
  9. Google Search Central, “Optimizing your website for generative AI features on Google Search”, last updated July 10, 2026. developers.google.com/search/docs/fundamentals/ai-optimization-guide
  10. 9to5Google, “Google AI Mode and AI Overview updates work to elevate links”, October 15, 2026. 9to5google.com/2026/10/15/google-ai-mode-overviews-links
  11. Louise Linehan, Ahrefs, “Update: 38% of AI Overview Citations Pull From The Top 10”, March 2, 2026. ahrefs.com/blog/ai-overview-citations-top-10
  12. SEO.com, “What a Recent Study of AI Mode Citations Means for The Future of Search”, summarizing Moz’s AI Mode citation study. seo.com/blog/ai-mode-citations
  13. RESONEO, “AI Overviews and AI Mode: what Google’s AI reads before it answers”, October 2026. think.resoneo.com/ai-overviews-ai-mode
  14. Abner Li, 9to5Google, “Google announces Gemini 4 Hydrogen with ‘significant leaps’ in speed, intelligence”, November 12, 2026. 9to5google.com/2026/11/12/gemini-4-hydrogen-launch

Appendix

A. Glossary

Gemini 4 Hydrogen
Google’s fast, cost-efficient Gemini 4 model, released November 12, 2026, and the default in AI Mode.
Gemini 4 Carbon and Argon
Google’s frontier Gemini 4 models, built for longer and deeper reasoning and limited to paid subscribers in Search.
AI Mode
The conversational experience inside Google Search that writes a full answer with links.
AI Overviews
The AI-written summaries that appear above regular results for some queries.
Query fan-out
Google’s technique of breaking one question into several related searches and running them at the same time.
Grounding
Retrieving pages from the Search index first, then generating a response with links to the pages that support it.
Core update
A broad change to Google’s ranking systems. The November 2026 core update began on November 10 and is separate from the Hydrogen launch.
Generative AI performance report
The Search Console report that shows how a site performs in Google’s AI features.
Inline links
Links embedded in the words of an AI response, as opposed to links listed beside or below it.
Scaled content abuse
Google’s spam policy against producing many pages mainly to manipulate rankings, which Google says covers pages made for every fan-out variation.

B. Model figures as reported by Google

These come from Google’s own announcement and the evaluations that it chose.[2] None has been independently reproduced, and Google hasn’t said which thinking level Search uses.

MeasureGemini 4 Hydrogen
Speed against Gemini 3.8 FlashAbout 2.5 times faster at the default setting
Fabricated claimsReduced by more than 5 times
Artificial Analysis Intelligence Index51 at the default level, against 50 for Argon
Terminal-Bench 4.157%, against 18.2% for Gemini 3.8 Flash
Thinking levelsFour: low, medium, high (default) and xhigh
Context window1 million tokens
API price$1.50 per million input tokens and $7.50 per million output tokens

C. How we researched this post

We read both of Google’s November 12 announcements, then the same-day coverage from Search Engine Land, Search Engine Journal, Search Engine Roundtable, PPC Land and 9to5Google. We checked the optimization advice against Google’s own guide for generative AI features, and used Ahrefs, Moz and RESONEO for the citation research.

We haven’t run our own before-and-after test of Hydrogen in Search yet, and for the reasons in this post we don’t think that a clean one is possible this week. The fan-out diagram and the sub-question prompts are our own illustrations, and they aren’t Google’s actual queries. The Moz figure is cited through a summary of the study.

D. Review notes

Last reviewed on November 12, 2026, the day of the launch. The core update was still rolling out at the time of writing, so check the Google Search Status Dashboard for its completion date.