Google switched the default model in AI Mode to Gemini 4 Hydrogen on 12 November 2026, just two days after the November 2026 core update began.
This is a significant change to how Google researches questions, finds sources and presents AI-generated answers. It does not, by itself, change how websites rank in Google’s traditional organic search results.
For businesses, the main concern is visibility. Your website could hold the top organic position for an important keyword but still be missing from the AI answer. That’s because AI Mode can explore several related searches and use different websites to answer different parts of a question.
This article explains what Google has changed, why it matters for SEO and what content, marketing and development teams should do next.
1. What Happened?
On 12 November 2026, Google made Gemini 4 Hydrogen available in Search worldwide.
Robby Stein, Google’s Vice President of Product for Search, described it as the biggest model upgrade to Search yet. Search Engine Journal and Search Engine Roundtable report that Hydrogen is now the default model in AI Mode and handles the most difficult questions in AI Overviews.
According to Search Engine Journal, Gemini 3.5 Flash was the previous default model worldwide.
Hydrogen is the faster, lower-cost model in the Gemini 4 family. Google DeepMind lists its pricing at $1.50 per million input tokens and $7.50 per million output tokens.
Google also reports that Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default setting, compared with 50 for Gemini 4 Argon. It says the new model produces more than five times fewer fabricated claims than Gemini 3.8 Flash.
The combination of speed, cost and accuracy makes Hydrogen suitable for use across everyday Search queries, rather than limiting it to paid Gemini users.
The four changes Google announced
Google highlighted four main improvements to Search.
1. Better understanding of what people are asking
Search can handle more complicated questions, including questions with several parts, spelling mistakes and specific requirements. The aim is to understand what the user needs and provide an answer that covers those requirements.
2. More searches and a wider range of sources
Hydrogen can run several related searches at the same time to investigate different parts of a question. This is known as query fan-out.
Google says the improved process helps connect users with a wider range of useful websites, including sources they might not have found through a normal search.
3. More visual answers
AI Mode can provide more images, structured answers, comparison tables and tools created around the question being asked.
4. Better understanding of language
Google says the model provides more detailed and natural answers across the languages supported by AI Mode.
What Google’s before-and-after examples show
Google published three comparisons when it announced the upgrade. These are examples selected by Google, so they shouldn’t be treated as independent research or proof that every AI Mode answer will change in the same way.
| Search query | Gemini 3.5 Flash | Gemini 4 Hydrogen |
|---|---|---|
| Making a studio flat feel like two rooms for under $200 | A text-based list of suggestions, starting with free ways to change the layout. | Starts with a specific recommendation, followed by a product type, a price range of $80–$140 and an image carousel. |
| Boston neighbourhoods for a 28-year-old starting a job downtown | Paragraphs and bullet points describing different areas. | A map organised around transport routes, followed by a comparison table covering neighbourhood, atmosphere, main transport line and social scene score. |
| Breathable slip-on trainers with arch support | A retailer product card showing a price and star rating. | Refers to the shoes the user already owns before showing image cards with alternative recommendations. |
Two things stand out.
First, the sources cited in the answers changed in all three examples.
Second, Hydrogen presented information in a more structured way. Its answers included details that could be compared, such as prices, transport routes and scores.
For SEO teams, this raises an important question: does your website provide the specific information an AI answer needs to compare products, services or locations?
The Gemini upgrade is separate from the November 2026 core update
Both changes happened within a few days of each other, but they are different updates.
| November 2026 core update | Gemini 4 Hydrogen in Search | |
|---|---|---|
| Started | 10 November 2026 | 12 November 2026 |
| What changed | Google’s core ranking systems | The model used in AI Mode and for more difficult AI Overview questions |
| Main impact | Organic search positions, Discover and other Search features | The sources used in AI answers and how those answers are presented |
| Duration | Google says the rollout can take up to two weeks | Available worldwide from launch |
| Where to measure | Google Search Console Performance report, after the rollout | Search Console’s Generative AI performance report and repeated prompt testing |
Barry Schwartz at Search Engine Roundtable has also discussed the distinction, explaining that Hydrogen affects AI Overviews and AI Mode rather than traditional core rankings. This is his interpretation, not an official Google statement about ranking systems.
The important point is not to treat every change in organic traffic or AI visibility as the result of the same update.
2. Why Does the Gemini Upgrade Matter for SEO?
The model that researches and presents Google’s AI answers has changed, but your traditional organic rankings don’t automatically change as a result.
This creates a visibility gap.
A page can rank first for its main keyword and still be absent from an AI answer. The reason is that AI Mode can break a question into smaller searches and use different sources to answer each part.
Three changes are particularly important.
AI Mode searches across several subtopics
A traditional SEO strategy might focus on getting one page to rank for a main keyword.
AI Mode can approach the same question differently. It may run several searches covering related products, prices, locations, specifications, comparisons and other requirements.
A page that only answers the main question may not provide enough information to support the full response.
For example, someone asking for the best software for a particular job might also need information about pricing, integrations, features, alternatives and ease of use.
Your website may rank well for the main term but miss the smaller questions that help Google build its answer.
Answers rely on information that can be compared
Hydrogen’s examples show more structured responses, including tables, price ranges, scores and image cards.
These formats rely on clear information.
For example, a comparison table needs individual details such as:
- Product price.
- Dimensions.
- Available features.
- Compatibility.
- Performance.
- Limitations.
A page that provides these details clearly gives the model more useful information to work with.
If the same information only appears inside images, PDFs or content that loads after a user interacts with the page, it may be harder to access.
Google changes how it presents information
A traditional list of search results is not the only way people can discover websites through Google.
Depending on the question, AI Mode may show a map, comparison table, image carousel or a tool built around the user’s requirements.
This changes where websites can appear and how their information is used.
Five possible SEO and marketing impacts
1. AI citations can change without organic rankings moving
Google says its improved query fan-out can find a wider range of useful websites. It hasn’t published figures showing how much more fan-out Hydrogen uses.
However, earlier research gives us an idea of why this matters.
In a November 2025 study, Seer Interactive analysed 501 prompts using Gemini 3. It found an average of 10.7 fan-out queries per prompt, compared with 6.01 for Gemini 2.5. The study also reported that 95% of those searches had no recorded search volume.
This research relates to earlier Gemini models, not Hydrogen. We shouldn’t assume the same figures apply to the new model.
However, it highlights a wider problem for SEO teams: many of the searches used to build AI answers may not appear in traditional keyword research tools.
2. Pages with clear product and service information have more opportunities to appear
Google’s examples show answers using product cards, comparison tables and other structured formats.
Websites that publish clear specifications, prices and other useful details give AI systems more information to work with.
For ecommerce businesses, this means keeping product information accurate and consistent across the website, structured data and Merchant Center feed.
For service businesses, it means explaining what is included, who a service is suitable for, how much it costs and any important limitations.
3. The effect on clicks is not yet clear
Richer AI answers could mean that some users get what they need without visiting a website.
However, Google’s wider selection of links could also help users discover websites they might not have found previously.
Google hasn’t published click-through, citation or conversion figures for Hydrogen.
We therefore don’t have enough evidence to say whether the upgrade will increase or decrease website traffic.
Businesses should measure their own results rather than assume the change will have the same effect across every industry.
4. It will be harder to identify what caused a traffic change
The November core update started on 10 November. Gemini 4 Hydrogen arrived in Search on 12 November.
The core update can take up to two weeks to complete. That means changes in organic traffic and AI visibility could happen within the same reporting period.
There is also seasonal activity to consider, particularly with Black Friday on 27 November and Cyber Monday on 30 November.
A change in traffic could be related to the core update, Hydrogen, seasonal demand or a combination of factors.
Don’t rush into major changes before you have enough data to understand what happened.
5. Accurate, supported information could become more important
Google says Hydrogen produces fewer fabricated claims than Gemini 3.8 Flash.
That doesn’t prove that Google AI Mode will always prefer a particular type of content. However, it reinforces the importance of publishing specific information that can be checked.
Avoid broad claims that you cannot support with evidence. Use clear prices, specifications, dates, ratings and other relevant facts where appropriate.
This is a sensible approach for both customers and search visibility.
What hasn’t changed
Google’s existing guidance for AI features still applies.
A page must be indexed and eligible to appear with a snippet to be used as a supporting link in AI Overviews or AI Mode.
Google has also said that websites don’t need special technical requirements, additional schema or AI-specific content to appear in these features.
The May 2026 guidance on generative AI features makes the same general point: good SEO practices and useful content still matter.
The recommendations in this article are therefore not about finding a new technical trick. They’re about making sure your pages answer the smaller questions behind important customer searches.
What we know and what Google hasn’t published
It’s important to separate Google’s announcements from the questions that still need further research.
| Status | Information |
|---|---|
| Confirmed by Google | Hydrogen is available in Search worldwide. Google has announced improvements to query fan-out, language understanding and visual answers. |
| Reported by Google staff and industry publications | Hydrogen is the default model in AI Mode and handles more difficult questions in AI Overviews. |
| Not yet published | Hydrogen’s average number of fan-out queries, citation share by website type, click-through rates and traffic impact. |
| Not yet explained | Exactly what Google considers a difficult AI Overview question and when it uses Hydrogen for those answers. |
3. Which Industries Are Most Affected?
The change is most relevant to websites that attract customers searching for recommendations, comparisons and answers to questions with several requirements.
These searches are particularly important in industries where customers need to compare different products, services or providers before making a decision.
| Sector | What may change | First priority |
|---|---|---|
| Ecommerce and retail | Product answers may include image cards and comparison tables based on price, specifications and suitability. Google’s own example also shows recommendations based on products the user already owns. | Keep product specifications, prices, availability and identifiers consistent across product pages, structured data and Merchant Center. |
| Local and multi-location businesses | Answers about areas and local services may include maps and comparisons between locations. Google’s Boston example compares neighbourhoods using transport, character and other attributes. | Make sure location pages provide clear information about transport, parking, opening hours, services and prices. |
| SaaS and B2B technology | Searches for the best tool for a specific job may cover features, pricing, integrations and alternatives. | Publish clear pricing, feature and integration information, supported by useful comparison pages. |
| Finance and insurance | Questions about eligibility, fees and costs may be answered together in one response, sometimes using tables. | Publish accurate rates, fees, eligibility requirements and dates. Include clear review information for important content. |
| Healthcare | Google says its safeguards are better at distinguishing genuine medical questions from queries that present a risk. More health searches may therefore receive detailed answers. | Publish clinically reviewed content that answers specific patient questions, with clear information about the author or reviewer. |
| Publishers and media | More searches across individual subtopics could lead Google to use specialist websites rather than relying on a general website that ranks for the main keyword. | Develop specialist coverage of the subjects you know well, supported by original research, images and clear authorship. |
| Enterprise and multi-market businesses | The upgrade applies across supported AI Mode languages. Weaknesses in shared page templates could affect thousands of URLs across several markets. | Fix shared templates first, then test important prompts in each market. |
Who is less likely to notice a change?
Websites that mainly receive traffic from simple navigational or transactional searches may see less direct impact from the changes to AI Mode.
For example, someone searching for a business by its exact name may have little need for a long AI-generated comparison.
If your organic rankings have changed since 10 November, investigate the November core update separately rather than assuming Hydrogen is responsible.
4. What Should Businesses Do?
The main priority is to make sure your website provides useful information for the different questions behind important customer searches.
You also need to make that information easy to access, compare and verify.
There are five steps I would recommend.
Step 1: Record your current performance before making changes
Owner: Marketing and SEO team · Timing: This week
Before changing your website, establish a clear baseline.
1. Record the two update dates.
Add annotations to your analytics, Search Console exports and rank tracking reports:
- 10 November 2026: November core update started.
- 12 November 2026: Gemini 4 Hydrogen became available in Search.
This will help you separate the two changes when reviewing performance.
2. Wait for the core update to finish before diagnosing ranking changes.
Google says the November core update can take up to two weeks to roll out.
Once it finishes, wait at least one full week before making a detailed comparison. Compare that week with the week before 10 November.
This gives you a cleaner comparison than relying on daily ranking movements during the rollout.
3. Save your AI search baseline.
Export the Google Search Console Generative AI performance report for the 28 days before 12 November.
Google introduced this report in June 2026 to provide information about impressions from AI Overviews and AI Mode.
Save the available data by page, country and device so you can compare it with future results.
4. Create a list of 30 to 50 prompts for each market.
Write questions in the way your customers would ask them.
Include specific requirements such as:
- Budget.
- Location.
- Product specifications.
- Compatibility.
- Eligibility.
- Delivery times.
- Other buying requirements.
For example, don’t only test “best accounting software”. Also test questions such as “best accounting software for a small business with fewer than ten employees and a budget under £50 per month”.
5. Run each prompt at least three times.
Test the same prompts in AI Mode on separate days.
Record:
- Whether your brand is mentioned.
- Which of your URLs are cited.
- Which competitors appear.
- Which other websites are cited.
- Whether the answer uses a table, map, image cards or another format.
- Whether the information about your business is correct.
One screenshot isn’t enough to establish a pattern. Answers can change between runs, locations and users.
Step 2: Find the smaller searches behind your priority topics
Owner: Content and SEO teams · Timing: Weeks one and two
For each important topic, think about the different searches Google may run to answer a customer’s question.
I recommend grouping them into five areas.
1. Definition and suitability
What is the product or service? Who is it for? Who might it not suit?
2. Comparisons
How does it compare with named alternatives? What are the differences? Is there a suitable alternative to buying or using it?
3. Specific requirements
Does it meet a particular budget, location, size, compatibility requirement, eligibility rule or deadline?
4. Evidence
What reviews, test results, accreditations, qualifications or guarantees support the claims?
5. The next step
How can someone buy, book, install, switch or cancel?
Once you have the questions, compare them with the content on your website.
Mark each one as:
- Covered well: the page provides a clear and complete answer.
- Partly covered: the information is there, but it’s incomplete or difficult to find.
- Not covered: the website doesn’t answer the question.
The questions that are missing or only partly answered should become your content improvement plan.
Prioritise them by commercial value rather than treating every question as equally important.
Traditional keyword tools may not show search volumes for many of these more specific questions. Use customer conversations, sales enquiries, support tickets, on-site searches and your AI prompt testing to find them.
Step 3: Improve your priority pages
Owner: Content team · Timing: Weeks two to four
The next step is to improve the pages that need to answer these questions.
I recommend using the following seven checks as your content standard.
1. Answer the main question immediately.
The opening two sentences beneath the main page heading should clearly explain what the page is about and answer the main question.
Don’t make readers work through several paragraphs before finding the information they need.
2. Give each important question its own section.
Use headings that reflect the questions customers ask.
Start each section with a direct answer, then add the supporting information.
3. Use proper comparison tables.
Present prices, dimensions, limits, delivery times and other comparable information in HTML tables.
Include units and currencies so the information is clear.
4. Explain who your product or service is suitable for.
Don’t only describe the benefits. Explain relevant limitations, requirements, price ranges and situations where the product or service may not be suitable.
5. Support important claims with evidence.
Make sure figures and claims have a clear source.
For important content, include the author’s name or the name of the person who reviewed it, along with the date it was last checked.
6. Use original images where they add value.
Use relevant product and service images, with descriptive alt text and captions where appropriate.
Make sure the images help explain what is being discussed rather than simply filling space.
7. Keep information consistent across your website.
Prices, specifications, opening hours and other important details should match across your main page, structured data, product feed and any other pages where they appear.
These are straightforward content improvements that make information easier for customers to use. They aren’t a special writing formula for Google AI Mode.
Step 4: Make sure search engines can access the information
Owner: Development team · Timing: Weeks two to four
Content improvements won’t help if important information is difficult for search engines to access.
Your technical SEO team should check the following.
- Priority pages return a successful 200 status code.
- Pages use the correct canonical URL.
- Important pages aren’t blocked by
robots.txt. - Priority URLs are included in the XML sitemap.
- Pages aren’t using
noindexunnecessarily. - Important content isn’t blocked by
nosnippet,max-snippet:0or an inappropriatedata-nosnippetattribute. - Main content, specifications and tables are available in the page’s rendered HTML.
- Tabs and accordions don’t hide important information from search engines.
- Key facts aren’t available only inside images, PDFs or videos.
- Tables use proper
<table>,<th>and<td>HTML elements. - Headings follow a clear structure, from H1 to H3 where appropriate.
- Structured data matches the information visible on the page.
- Product information in Merchant Center matches the website.
- Images have crawlable URLs, descriptive file names, alt text, width and height attributes.
- Important sections have their own IDs where direct links to those sections are useful.
- Hreflang tags are correct across international versions of the website.
- Automated checks confirm that important page components and structured data remain in place after each release.
For structured data, use the relevant types for the content, such as Product, Offer, LocalBusiness, Organization, Article and BreadcrumbList.
The markup must match the visible page.
Google has said that special structured data isn’t required for its AI features. Don’t add markup expecting it to guarantee a citation.
Step 5: Roll out the changes across the website using templates
Owner: SEO lead and product owner · Timing: From week three
For large websites, changing every page individually is rarely the best approach.
Start by identifying the templates used across the site.
1. List your page templates.
Include product pages, category pages, location pages, service pages, comparison pages, articles and help pages.
Record the number of URLs, revenue and AI visibility for each template.
2. Decide which templates to fix first.
Prioritise templates that combine high commercial value with weak AI visibility.
3. Improve shared page components.
Instead of manually rewriting hundreds of pages, identify components that can be improved once and reused.
For example:
- Product specification tables.
- Comparison tables.
- Question-and-answer sections.
- Author and reviewer information.
- Last-reviewed dates.
- Images and captions.
4. Use one reliable source for your data.
Where possible, connect these components to your product database, content management system or location database.
This means a price or specification can be updated in one place rather than manually changed across several pages.
5. Test the changes on a smaller group of pages.
Start with 20 to 50 URLs using the same template.
Keep a similar group of pages unchanged so you have something to compare the results against.
6. Measure the results over four weeks.
Compare AI visibility, prompt-test citations, organic clicks and conversions between the updated pages and the control group.
7. Roll out the changes across the template.
Once the pilot has been reviewed, extend the improvements across the remaining URLs.
Update the content briefs, CMS requirements and development checks so that new pages follow the same standard.
Who should be responsible?
A successful rollout needs clear ownership.
| Team | Main responsibility | What they deliver | When the work is complete |
|---|---|---|---|
| Marketing and SEO | Baseline, prompt testing and reporting | Dashboards, AI visibility reports and lists of related searches | AI visibility is measured separately from traditional rankings. |
| Content | Page improvements and content planning | Updated priority pages, content briefs and fact records | Priority pages meet the agreed content standard. |
| Development | Templates, shared components and technical checks | Tables, comparison sections, author details, structured data and feed checks | Technical checks pass across the agreed templates. |
| Product or ecommerce | Accurate product and service information | Complete and consistent product, service and location data | Required information is present on live pages. |
| Leadership | Business priorities and approvals | Agreed rollout plan and approval for the pilot | Teams have agreed priorities and the pilot has been reviewed. |
A practical 30-day action plan
| Week | Dates | Main activities |
|---|---|---|
| 1 | 13–20 November | Record the update dates, save AI visibility data, create the prompt list and fix any serious indexing or snippet issues. |
| 2 | 23–27 November | Map the related searches for the ten most important topics, review page templates and check product information. Retailers should also respect any Black Friday development freeze. |
| 3 | 30 November–4 December | Review the core update once a clean comparison period is available. Start the first page template pilot. |
| 4 | 7–11 December | Repeat the prompt tests, compare the pilot pages with the control group and agree the order for the wider rollout. |
What businesses should avoid
There are several mistakes I would avoid during this process.
- Don’t blame Hydrogen for an organic ranking drop. Investigate the core update and other possible causes separately.
- Don’t add an AI text file or invented AI schema. Google says these aren’t required for its AI features.
- Don’t rewrite the whole website during a core update. Making widespread changes at the same time makes it difficult to understand what caused a performance change.
- Don’t buy brand mentions to influence AI answers. Google’s Gary Illyes has reportedly warned website owners against this approach.
- Don’t make decisions based on one AI answer. Repeat the tests and look for consistent changes before adjusting your strategy.
5. What We’re Watching Next
There are six developments worth monitoring as Google continues to roll out Gemini 4 Hydrogen in Search.
1. Completion of the November core update
Google’s stated rollout period is up to two weeks from 10 November.
Once the update has finished, we can start making cleaner comparisons between organic search performance before and after the update.
This is important because Hydrogen arrived during the rollout.
2. Independent research into Hydrogen’s query fan-out
Seer Interactive’s earlier research found a substantial increase in the number of searches generated by Gemini 3 compared with Gemini 2.5.
We expect further research into how Hydrogen handles query fan-out.
The number of searches it generates, the types of websites it finds and how those sources are selected will be particularly interesting for SEO teams.
3. How widely Hydrogen is used in AI Overviews
Google has said Hydrogen handles the hardest questions in AI Overviews, but hasn’t explained exactly what qualifies as a difficult question.
AI Overviews appear across a wider range of Search queries, so understanding how far Hydrogen is used in this feature will be important.
4. More reporting in Google Search Console
The Generative AI performance report provides a way to monitor AI search impressions.
However, more reporting on clicks and the queries behind those impressions would make it easier to measure AI search as a customer acquisition channel.
5. Custom tools in AI answers
Google says Hydrogen can create tools for individual questions.
We’ll be watching how these tools work, whether they include links to their sources and how they affect the way people discover websites.
6. Different Gemini models in AI Mode
Gemini 4 Carbon became available to Ultra subscribers in AI Mode on 10 November, while Argon is available to paid users.
Different models may return different answers and cite different websites for the same query.
This creates a measurement challenge. One result from one model may not represent what every Google user sees.
Why repeated testing will become more important
One of the most interesting parts of Google’s examples is that Hydrogen considers information about the user’s existing products and preferences when making recommendations.
For example, its trainer example takes into account the shoes the user already owns before suggesting alternatives.
As AI answers become more tailored to individual users, checking one ranking or taking one screenshot will provide a less complete picture of search visibility.
SEO teams will need to test important questions repeatedly, record the sources and brands that appear, and look for patterns over time.
6. About Szymaniak Digital
Szymaniak Digital is an Enterprise AI SEO Consultancy based in Romsey, Hampshire, led by Konrad Szymaniak.
We help marketing, content and development teams understand how Google and AI platforms find, use and recommend information about their brands.
Our work combines technical SEO, AI search research, content strategy and performance measurement. The aim is to turn changes in search into practical improvements that support customer acquisition, revenue growth and market share.
