Google says its new Search model runs simultaneous queries across more subtopics and connects users to "a broader range of helpful links than before". Gemini 4 Hydrogen became the default model in AI Mode on 12 November, globally and for every user, and that line about fan-out is the part of the announcement with consequences for a menu page.
Hydrogen is the first Gemini 4 model given to everyone
Google has put three Gemini 4 models into Search in a month and gated them differently. Argon arrived on 13 October for paid AI Pro and Ultra plans. Carbon began rolling out in AI Mode to Ultra subscribers on 10 November. Hydrogen, two days after that, went to everybody: "Hydrogen is now available globally in Google Search for all users."
The post is by Robby Stein, VP of Product for Google Search, and calls Hydrogen "the biggest model upgrade to Search yet". It describes improvements across all supported AI Mode languages. It does not mention AI Overviews anywhere. That detail sits in Stein's post on X instead, where he writes that Hydrogen is "now the new default model in AI Mode & being used for your hardest questions in AI Overviews". The reach into AI Overviews, which turns up for people who never open the AI Mode tab, was announced socially rather than in the documentation.
A wider fan-out means narrower questions
Query fan-out is the mechanism underneath the claim. Google describes it as "conducting simultaneous queries across all the relevant subtopics", and says the effect is that "Search helps you find sites you previously might not have discovered".
Those subtopics are where menus live. A diner's question rarely arrives as two words and a place name. It arrives with conditions attached: a dietary requirement, a time, a number of people. The fan-out splits that into its constituent questions before any page is picked, so a page is competing for the sub-query rather than for the sentence the diner typed.
That cuts both ways once the markup is stale. A kitchen takes a dish off the lunch menu and the JSON-LD still carries it; the sub-query that matches now matches something untrue.
Google published the claim without numbers
Neither post carries a figure for it. Luis Rijo noted at PPC Land on 12 November that the announcement gives no count of sub-queries, no average number of links per answer, and no indication of which sites benefit. The companion model post quotes Artificial Analysis scores alongside speed and hallucination claims, all of them from Google or from evaluations Google selected, none independently reproduced. There is no click or referral data in either post.
"A broader range of helpful links" is therefore a statement of design intent, not a measured outcome. It is still worth taking seriously, because it tells you what Google is optimising towards, but it is not evidence that any particular site is being cited more this week than last.
Nothing measurable this week will settle it
The November 2026 core update began on 10 November and may take up to a fortnight to finish rolling out. Hydrogen landed two days into it. The ranking system and the model that selects and summarises sources are both moving inside the same window, which leaves any traffic change this month attributable to either and provable for neither.
The reporting has a hole in it as well. Search Console's Generative AI performance report under-logged impressions from 9 to 11 November. Google filed the issue on its data anomalies page with the note that "this issue affects data logging only", and Barry Schwartz wrote it up at Search Engine Roundtable on the same day Hydrogen shipped. A dip in AI impressions across those three days is a reporting fault rather than a result.
Hydrogen changes how many questions Google asks the web on a diner's behalf. It does not change what a restaurant's own pages are able to answer. A fan-out that reaches further still has to land somewhere that states the allergen and the price in a form a machine can match.
