Google made Gemini 4 Hydrogen available to every Google Search user worldwide on 12 November 2026, two days after its November core update began. Google calls it the biggest model upgrade Search has had and says its wider query fan-out links to more sites, so any AI visibility baseline taken before 12 November is out of date.

Key takeaways

  • Google announced Gemini 4 Hydrogen in Search on 12 November 2026 and says it is available globally to all users, not only subscribers.
  • Google says Hydrogen widens query fan-out, running simultaneous searches across subtopics to link to a broader range of sites.
  • Google also promises more images, structured responses, comparison tables and custom tools in AI answers.
  • The November 2026 core update started on 10 November, so two Google changes hit the same week.
  • Search Console’s generative AI report undercounted impressions from 9 to 11 November 2026 because of a logging error Google says affected data only.

What Google shipped on 12 November

Robby Stein, VP of Product for Google Search, wrote Google’s Search post, which says Hydrogen “is now available globally in Google Search for all users”. The post names four improvements, mentions AI Mode languages but not AI Overviews, and gives no figures for links, clicks or answer quality.

The model itself came from Google DeepMind the same day. Koray Kavukcuoglu’s launch post calls Hydrogen Google’s “daily driver for agentic workflows”, built for fast execution, while Google’s frontier Carbon and Argon models handle deep, long-horizon reasoning. Google says Hydrogen cuts fabricated claims by more than five times against Gemini 3.8 Flash, and its Search comparison images set it against Gemini 3.5 Flash. Developers pay $1.50 per million input tokens and $7.50 per million output tokens, and Google says free Gemini app users reach it through automatic model selection.

What Google says changedWhat it means for a brand
Better reading of complex, multi-part, typo-filled questionsLong buyer questions get fuller answers, which favours pages that answer the specifics
Wider query fan-out across subtopics, with more linksMore pages can be cited per answer, including sites that didn’t rank before
More images, structured responses and comparison tablesProduct data, specs and clear tables give Google material to lift
More nuanced answers in every supported AI Mode languageUK and non-English answers may change as much as US ones

Which searches does Hydrogen change?

Google’s post says Hydrogen handles multi-part, typo-filled and highly specific questions better. In a fan-out, Google splits one question into subtopic searches, runs them at once and builds the answer from what comes back, which we explained in The citation moved off the page as the reason rankings stopped predicting citations. On that description, a shopper asking which waterproof running shoes suit wide feet on muddy trails under £150 would set off searches on fit, terrain, waterproofing and price, and each one is a chance for a different page to be cited.

Our read is that the searches a buyer types before choosing a brand will change most, while head terms that already return a short answer will change least. Google hasn’t published which query types Hydrogen handles in AI Overviews, so treat AI Mode as the confirmed surface and AI Overviews as likely but unconfirmed.

What it means for SEO and content teams

A wider fan-out rewards coverage of the subtopics buyers care about, written so each section answers one of them on its own. A brand with one long guide that buries fit advice in paragraph nine gives Google less to cite than a brand with a clear section, table or page for each concern. The SISTRIX citation study found structural signals such as tables of contents travelled with more citations, and Google is now promising more tables in its own answers.

Google’s October link changes matter more now too. We covered how Google moved inline links higher in AI Mode on 15 October, and Google says Hydrogen links to a broader range of sites than before. The citation race is wider, so a mid-sized brand has a better chance of a link and a weaker hold on any one of them.

Our view at Manual Focus is that SEO teams should treat 12 November as a new baseline and rebuild their AI visibility tracking around the multi-part questions buyers ask, because Google has changed the model that writes the answers and its fan-out decides which pages get the links.

How do you measure it this month?

Google’s own reporting makes this week hard to read. Its data anomalies page says a logging error lowered impressions in the generative AI report for 9 to 11 November 2026, a window that includes the start of the core update. Google puts the lower figures on those dates down to logging, so don’t read them as a Hydrogen or core update effect.

Analytics won’t separate it either. Google’s new help page on identifying Search traffic says Search sends only the scheme and host, such as https://www.google.co.uk/, with no query, and the page describes no separate value for AI Mode. It also says Search Console counts a click when someone selects a result, while analytics counts a visit only once the page loads and its tag runs. Search Engine Roundtable flagged the timing and suggested, without Google confirming it, that Hydrogen touches AI features rather than core rankings.

Five jobs before December

  1. Mark the dates. Add 10 November (core update), 12 November (Hydrogen) and 9 to 11 November (logging error) as annotations in GA4, Looker Studio and your SEO reports, so nobody reads a logging dip as a ranking loss.
  2. Rebuild the prompt set. Write 20 to 30 multi-part questions your buyers ask, with budget, fit and use case in them, and record which brands and pages AI Mode cites for each from a UK location. Re-run them weekly to the end of December. Light Meter runs this kind of panel across six engines, including AI Mode and AI Overviews.
  3. Map the fan-out gaps. For each question, list the subtopics Google would search and check you have a page or clear section answering each one. Fill the gaps with the SEO cluster generator playbook.
  4. Give Google tables to lift. Put specs, sizes, prices and comparisons on your key product and guide pages as plain HTML tables, because Google says Hydrogen returns more comparison tables.
  5. Compare clicks with visits. Set up a referrer segment for Google domains in analytics, as Google’s help page describes, and compare it with Search Console clicks weekly. The attribution tear-down playbook shows how to keep the two honest.

What Google hasn’t said

Google gives no number for how many more links Hydrogen shows, how often it cites sites outside the top results, or whether clicks rose or fell. It doesn’t say whether AI Overviews run on Hydrogen for every query or only some, or which model Search used before for all users, beyond the Gemini 3.5 Flash comparison images. It also doesn’t say how Hydrogen interacts with ads in AI Mode, which Google is bringing to the UK this quarter, as we set out in our AI Mode ads post. When we covered Lens data in Search Console in September, the generative AI report showed impressions without clicks, so it can’t show whether Hydrogen’s extra links send anyone to your site. Until Google publishes more, brands will have to measure the effect on their own accounts.

Further reading

If you want help rebuilding your AI visibility baseline after the Hydrogen change, our services page explains how we work.

Frequently asked questions

What is Gemini 4 Hydrogen in Google Search?

Gemini 4 Hydrogen is a Google model that the company made available globally in Google Search for all users on 12 November 2026. Robby Stein, VP of Product for Google Search, called it the biggest model upgrade to Search yet. Google says it reads complex and typo-filled questions better, runs a wider query fan-out and returns more images, structured answers and comparison tables.

Does Gemini 4 Hydrogen send more traffic to websites?

Google hasn't said. Its 12 November 2026 Search post says Hydrogen's query fan-out connects people to a broader range of helpful links, including sites they previously might not have discovered. It gives no figures for links shown, clicks or traffic, so brands should compare their own Search Console and analytics data from before and after the change.

Is Gemini 4 Hydrogen the same as the November 2026 core update?

No. Google's Search Status Dashboard lists the November 2026 core update from 10 November, and Google announced Gemini 4 Hydrogen in Search two days later, on 12 November. Search Engine Roundtable suggested, without confirmation from Google, that Hydrogen affects AI Overviews and AI Mode rather than core rankings. Both changes landed in the same week, so traffic shifts are hard to attribute.

Can I see Gemini 4 Hydrogen traffic separately in Google Analytics?

Not from the referrer alone. Google's help page on identifying Google Search traffic says Search sends only the scheme and host, such as https://www.google.co.uk/, with no query. The page doesn't describe a separate value for AI Mode or AI Overviews, so analytics groups those visits with other Google Search clicks. Search Console's generative AI report shows impressions only.

Sources

From the podcast

James and Hannah talk through Gemini 4 Hydrogen, which Google made available to every Google Search user worldwide on 12 November 2026 and called the biggest model upgrade Search has had. They explain query fan-out, why a wider fan-out rewards brands with a page or section for each question buyers ask, and why this week’s Search data is hard to read, with a core update from 10 November and a Search Console logging error from 9 to 11 November. Then three jobs to hand to AI so you can set a new baseline.

Get AI on this

Paste these into whichever AI assistant you use. Swap the bracketed parts for your own details. The first needs an assistant that can browse the web and run on a schedule, and the other two need file uploads or browsing.

1. Rebuild your prompt set and check it weekly

I run marketing for [brand], [one line on what you sell] at [website], selling
mainly in [market]. Our main competitors are [competitor 1], [competitor 2]
and [competitor 3].

Google now builds AI Mode answers by splitting a question into several
subtopic searches. I want to track how we appear for the long, specific
questions buyers ask before they choose a brand.

Step 1. Write 25 questions a buyer in [market] might type into Google before
choosing a [product category]. Each should combine at least two of budget,
fit or size, use case, conditions and comparison with an alternative. Avoid
our brand name and competitor names. Show them as a numbered list and wait for
me to approve or edit them.

Step 2. Once I approve the list, set this up as a routine that runs every
Monday morning until [end date]. For each question, search Google in AI Mode
from a [market] location and record:
- every brand the answer names, in order
- every website the answer links to, in order
- whether [website] is linked, and which page

Give me a table with the columns: Date | Question | Brands named | Sites
linked | Our page linked (URL or "no") | Competitors linked.

After each run, list what changed since the previous week. If AI Mode didn't
show an answer or wouldn't load for a question, say so in that row rather than
leaving it blank or guessing.

2. Map the fan-out gaps

I run marketing for [brand] at [website], selling [product category] in
[market].

Attached are:
- our list of buyer questions (from our weekly AI Mode tracking)
- our sitemap or a list of our page URLs with their titles

For each question, list the three to six subtopics a search engine would
probably search separately to answer it (for example fit, terrain, price,
durability, care). Then check whether our site has a page, or a clearly
headed section on a page, that answers each subtopic directly.

Give me a table with the columns: Question | Subtopic | Our page or section
that answers it (URL and heading, or "none found") | How directly it answers
(direct, partial, buried).

Then list the subtopics with no page or only a buried answer, ranked by how
many questions they appear in. Base the coverage check only on the pages you
could open or the titles in the file, and say which pages you couldn't check.

3. Find the missing tables

I run marketing for [brand] at [website].

Attached is a list of our 20 most important product and guide pages.

Google says its AI answers now show more comparison tables, so I want the
facts on our pages laid out in a form it can lift.

For each page, find any specifications, sizes, prices, compatibility details
or comparisons written in paragraphs or bullet points rather than in a table.

Give me:
1. A table with the columns: Page URL | Facts currently in prose | Suggested
   table (yes or no) | Why.
2. For each "yes", a draft HTML-ready table with a header row, using only the
   facts on the page. Don't add or round any figure.

Flag any page you couldn't open, and any fact that looked inconsistent between
pages, rather than choosing one yourself.

And one for you

Add three annotations in your analytics and SEO reports: 10 November 2026 for the core update, 12 November for Gemini 4 Hydrogen, and 9 to 11 November for the Search Console logging error. Then compare Search Console clicks with Google referral visits in analytics weekly, so you can see which way each moves.

Read the episode transcript

James: If your search report looks odd this week, Google gave you three reasons for it in one week. On the twelfth of November Google put a new model, Gemini four Hydrogen, into Search for everyone, two days after a core update started. I'm James Vickers, this is Manual Focus, and Hannah's here to untangle which change did what.

Hannah: Hi James. Some of it can't be untangled yet, so I'll say where the line is.

James: Start with what Google said.

Hannah: We read two Google posts from the twelfth. Robby Stein, who runs product for Google Search, called Hydrogen the biggest model upgrade to Search yet, and said it's available globally in Search for all users, not only subscribers. The second post, from Google DeepMind, launched the model itself and calls it their daily driver for agentic work, built for speed, while its frontier models, Carbon and Argon, handle the long reasoning.

James: What does Google say changes for someone typing a search?

Hannah: Four things. It reads complex, multi-part and typo-filled questions better. It widens what Google calls query fan-out. More images, structured answers and comparison tables. And more nuanced answers in every language AI Mode supports. What it doesn't give you is a single number on links, clicks or traffic.

James: Explain fan-out for anyone who hasn't met it.

Hannah: You ask one question, and Google quietly splits it into several smaller searches, runs them at the same time and writes the answer from what comes back. So a question about waterproof running shoes for wide feet on muddy trails under a hundred and fifty pounds could become searches on fit, terrain, waterproofing and price. Google says Hydrogen does more of that, and links to a broader range of sites, including ones people might not have found before.

James: Which is the mechanism we talked about in The citation moved off the page, back in June. Rankings stopped predicting who got quoted, because the answer was built from searches you never saw.

Hannah: Same mechanism, turned up. And it builds on the link changes we covered in mid October, when Google moved inline links higher in AI Mode.

James: So, the so what for a content team. If the answer is assembled from subtopics, a brand needs a page, or at least a clear section, for each one.

Hannah: That's the read. One long guide that buries fit advice in paragraph nine gives Google less to cite than a page with a proper section on fit. And Google says it's showing more comparison tables, which ties in with the SISTRIX study we covered in October, where tables with a proper header row went with more AI citations.

James: Does it help a mid-sized brand or hurt it?

Hannah: Both. A wider fan-out means more links per answer, so more chances for a brand that didn't rank on the head term. But more links also means any one link is less secure. That's our reasoning. Google hasn't published anything that measures it.

James: And which searches move?

Hannah: Google only confirms AI Mode in its Search post. It doesn't mention AI Overviews at all, so I'd treat AI Overviews as likely but unconfirmed. Our guess is the long questions a buyer asks before choosing a brand change most, and short head terms least.

James: Now the measurement problem, because that's what lands on a marketing director's desk on Monday.

Hannah: Three things overlap. The November core update started on the tenth. Hydrogen came on the twelfth. And Google's own data anomalies page says a logging error lowered impressions in Search Console's generative AI report from the ninth to the eleventh. Google says that one affected data only, so a dip on those days is a logging fault, not lost visibility.

James: Can analytics tell the AI traffic apart?

Hannah: Not on its own. Google published a help page this week on spotting Search traffic in your analytics or server logs. It says Google sends only the site address, google dot co dot uk for example, with no search query, and it describes no separate value for AI Mode. It also explains why Search Console clicks and analytics visits never quite match. Search Console counts the click, and analytics counts the visit only once your page loads and the tag fires.

James: Our view at Manual Focus is that the twelfth of November is a new baseline. Rebuild the tracking around the multi-part questions buyers actually ask, because the model writing the answers has changed.

Hannah: And mark the three dates in your reports first, so nobody reads a logging dip as a ranking loss.

James: So what do you hand over? Get AI on this, first job. Rebuild your prompt set and check it every week.

Hannah: Give it your product range, your market and your main competitors, and it writes twenty to thirty multi-part buyer questions with budget, fit and use case in them. Then set it as a weekly routine, running each one in AI Mode from a UK location and logging which brands and pages are cited.

James: Second, map the fan-out gaps.

Hannah: Give it those questions and your sitemap. It lists the subtopics Google would likely search for each one, and marks which have a page or clear section on your site and which have nothing.

James: Third, find the missing tables.

Hannah: Give it your top twenty product and guide pages. It tells you which ones have specs, sizes, prices or comparisons sitting in paragraphs, and drafts each as a plain table for the content team to check.

James: The prompts are in the show notes. Hannah, what don't we know?

Hannah: How many more links Hydrogen shows, how often it cites sites outside the top results, or whether clicks went up or down. Whether AI Overviews run on it for every query or only some. How it sits with ads in AI Mode, which Google is bringing to the UK this quarter. And when we last looked, Search Console's generative AI report showed impressions but no clicks, so it can't tell you whether those extra links send anyone your way.

James: So the takeaway. Google changed the model that writes its AI answers, in the same week as a core update and a reporting glitch, so annotate the dates, reset your baseline and make sure every question your buyers ask has a page that answers it. The full article is on manual-focus dot co dot uk. If this was useful, follow Manual Focus in Apple Podcasts, Spotify or wherever you listen, and if you want a hand rebuilding your own baseline, book a free working session on the site.

Hannah: Thanks, James. And maybe don't panic about the ninth to the eleventh.