Google Rolls Gemini 4 Hydrogen Into Search for All Users

By Amara Osei

Google has switched on Gemini 4 Hydrogen for every Google Search user worldwide, and the company calls it the biggest model upgrade Search has ever had. This is not a beta or a limited Labs experiment. It is live for everyone at once — a signal of how central Google now believes AI-generated answers are to its core product.

Google is pitching four improvements. The first targets messy, real-world queries. Search should now handle long, multi-part questions, poor spelling and highly specific conditions, and address each part rather than answering only the easiest component. Google credits stronger instruction following and tool calling (agent-style skills that let a model execute tasks and use external tools in sequence) — the same capabilities AI labs have prioritised across their models this year.

What is query fan-out?

The second change is what Google calls query fan-out (breaking a question into subtopics, searching them in parallel, then combining the results). Google says Hydrogen's reasoning makes this process more thorough and surfaces a wider range of websites, including ones people might not otherwise find.

That claim deserves scrutiny. Since AI Overviews (Google's AI-generated answer summaries) launched in 2024, followed by AI Mode, publishers have argued that AI answers keep readers on Google instead of sending them to the sites the information came from. The promise that Hydrogen surfaces more links reads as a direct answer to that criticism. Whether those links actually get clicked is a separate question — and Google has shared no traffic data to support the claim.

How do the answers themselves change?

Third, answers become more visual. Google promises more images, structured layouts, comparison tables and custom interactive tools built for specific questions — useful for comparing laptops or planning a trip. It also means more of the answer happens on Google's own page, not on the publisher's site.

Finally, Google says responses will be more nuanced across every language AI Mode supports. That matters in regions like the Middle East, where Arabic-language AI search has generally lagged behind English in quality.

What is Google not saying?

The gaps matter as much as the announcements. Google has published:

• no accuracy figures for Search;
• no data on how often Hydrogen gets facts wrong compared with earlier models;
• no explanation of how it chooses which sources to cite.

AI answers in Search have made embarrassing mistakes before, and a more capable model does not remove that risk. It mainly makes confident mistakes sound more convincing.

So what?

Google is no longer treating AI answers as an optional layer on top of its links. With Hydrogen live for everyone, the real test will be whether users trust its answers and whether publishers see any measurable benefit from the extra links Google says it is surfacing — and on that second point, the burden of proof sits squarely with Google.