Skip to content
Skip to content
News5 min read

Gemini 4 Hydrogen is now the default model in AI Mode

By Editorial standards

Google made Gemini 4 Hydrogen the default model in AI Mode on 12 November, for every searcher, and is using it for the hardest questions in AI Overviews, Google's Robby Stein posted on X, as Search Engine Roundtable reports. Gemini 4 Argon, which reached AI Mode in October, was offered only to paying subscribers, whereas Hydrogen answers everyone. In our UK readings in September, before the change, AI Mode named the business ranked first on Google for the same buying question in 625 of 1,630 answers.

What did Google change?

Google's Search announcement, by Robby Stein, Google's VP of Product for Search, calls Hydrogen "the biggest model upgrade to Search yet" and says it is now available globally in Google Search for all users. It names no countries, so it does not mention the UK by name.

Google lists four improvements. Search should handle multi-part questions, typos and highly specific requests better, and follow instructions more closely. Its query fan-out, which runs several searches at once across a question's subtopics, now connects people to a broader range of links than before, Google says. Answers get more images, structured layouts, comparison tables and custom tools. And responses are more nuanced in every language AI Mode supports.

The model announcement rolls Hydrogen out the same day in the Gemini app's model menu, where free users get it through automatic model selection when it suits the prompt, and to developers through the Gemini API. Search Engine Roundtable notes that Google released it during the November core update, which began on 10 November, and days after Gemini 4 Carbon reached AI Mode for paid subscribers.

What could it change for your brand?

Google does not say how the new model picks the businesses and pages an answer names, so nobody outside Google can say yet whether your brand gains or loses. Two of its claims are worth watching. A broader range of links could spread citations across more sites, including ones that were not cited before. Better handling of specific questions matters because buyers often add their size, their tools or their location to a question rather than asking only for the "best" option.

That second point is where we have a baseline. The UK AI Shortlist Study asked each buying question in two forms: a broad one, such as "best payroll software for small business UK", and narrower versions, such as "payroll software for weekly paid hourly staff UK". Before Hydrogen, Google's AI answers named the business ranked first on Google a little less often on the specific questions.

How often did Google's AI name the top business before?

In September, AI Mode named the business whose own website ranked first on Google in 175 of 428 answers to broad questions and 450 of 1,202 answers to specific ones. AI Overviews named it in 257 of 388 and 674 of 1,082. The Gemini app named it in 141 of 426 and 315 of 1,201.

The gaps for AI Mode and AI Overviews are small, around four percentage points, and could come from which businesses happen to rank first for each kind of question. The Gemini app's gap is wider, at about seven points. These readings come from one collection on 6 and 7 September 2026, when neither Argon nor Hydrogen existed, so they describe the older models only and are a baseline to compare new readings against.

625 of 1,630

AI Mode answers that named the business whose own website ranked first on Google, before Gemini 4 Hydrogen

UK AI Shortlist Study, 6 and 7 September 2026

Share of answers naming the business ranked first on Google, by type of question
Google AI serviceBroad questionsSpecific questions
Google AI Overviews66.2%62.3%
Google AI Mode40.9%37.4%
Gemini33.1%26.2%

UK AI Shortlist Study, 6 and 7 September 2026, before Gemini 4 Hydrogen. AI Overviews: 257 of 388 broad and 674 of 1,082 specific answers. AI Mode: 175 of 428 and 450 of 1,202. Gemini: 141 of 426 and 315 of 1,201. AI Overviews are scored only where one appeared.

What should you do this week?

Treat 12 November as a new starting point for AI Mode. Any reading of how AI Mode describes your business from before that date came from an older model, so compare new readings with each other rather than with last month's.

Expect it to be hard to say why something moved. The core update began on 10 November and Hydrogen arrived two days later, and both can change who an AI answer names. Search Console adds a third problem: Google's data anomalies page says a logging error caused a decrease in impressions on the Generative AI performance report for 9 to 11 November, and that it affects data logging only. So a fall in AI impressions on those three days is a logging gap, and your visibility may not have dropped at all.

Check the specific questions your buyers ask, with their conditions in, as well as the broad "best" questions. To see who AI names in your category now, the free AI visibility check asks buyer questions about your business on Claude, ChatGPT and Google AI Overviews, with no login, and shows who each answer names and the pages it cites. Our Google AI Mode tracker page explains how Discoverable reads AI Mode answers.

How we measured it

The figures come from the UK AI Shortlist Study. Its 1,000 buying questions in 25 categories were fixed before collection, each family made of one broad question and narrower versions of it, and were asked in three rounds on 6 and 7 September 2026 with the country set to the UK. Google was searched for each question in each round. Wherever a business's own website held Google's first organic position for that question and round, we checked whether each AI answer named it. A business counted as named when a verified name or alias appeared in the answer text, with the study's published correction for words the text extraction had joined together. Answers that only linked its website count as not named. Missing answers and unclear matches were left out, and AI Overviews were scored only where one appeared.

The counts are the first-position rows in rank-1-to-3-rows.csv with the corrections in glue-correction.json applied, split by each question's type in questions.csv. All three files are in the evidence bundle free to download from the study page, and the totals add up to the study's published first-place rows. The details of the model change come from Google's announcements and Search Engine Roundtable.

Sources

Keep exploring: Free AI visibility checkerWhat is GEO?The best GEO tools (2026)AI Visibility Index: 40 brands measuredGEO statistics 2026Google AI Overviews trackerChatGPT visibility trackerPerplexity visibility trackerGemini visibility tracker

See where your business stands.

Run a free AI visibility check and get your score in about a minute, no login.