For twenty years, Google's job was to point. You asked, it pointed you to ten pages, and the page did the work. That job description just changed again.
What Google actually announced
On the Google blog this week, the company said Gemini 4 Hydrogen is now available globally in Google Search for all users. The claims are specific. Search handles multi-part questions, typos and specific requirements better, thanks to what Google calls deeper instruction following and tool calling.
Responses now include more images, structured layouts, comparison tables and "custom tools built for you." The upgrade applies across all languages supported by AI Mode. Google also says an improved query fan-out runs simultaneous searches across subtopics, which connects users to a broader range of links than before.
What the post does not include is just as telling. No numbers on usage, accuracy or traffic. No named publishers. No commitment about how many of those broader links get clicked.
From pointing to assembling
Read the feature list again with a builder's eye. A comparison table means Google read several vendor pages and normalised them into columns. A custom tool means Google built a small calculator or planner on the spot instead of sending you to the site that already had one.
This is the shift from a search engine to an answer engine to something closer to a page factory. I covered the commercial version of this in Google wants to be your Booking.com. Hydrogen generalises the idea. Every query can now get a generated interface.
There is also a quieter risk. When Google writes the table, Google also decides which attributes matter. If the columns are price, rating and delivery time, a premium vendor that wins on expertise looks expensive and slow.
The comparison frame is now set by the machine. Frames decide outcomes before any human reads a word.
For a business, the consequence is architectural. Your website used to be the finished product. Now it is increasingly a parts supplier. Google takes your specs, your prices and your claims, then assembles them next to your competitors' in a layout you did not design.
What gets taken apart, and what survives
When a machine builds a comparison table, it needs clean, extractable facts. Price, dimensions, compatibility, delivery times, service areas. If those facts are buried in a PDF, locked in an image or phrased as marketing fluff, the table either skips you or fills the cell with someone else's description of you.
That is the uncomfortable test. Open your top three product or service pages and ask: could a machine fill a comparison table from this page without guessing? In most audits I see, the answer is no. The page was written to persuade a human who already arrived, not to feed a system deciding who gets included.
I made a related argument in your product page was written for the wrong reader. Hydrogen makes it more urgent, because the machine is no longer only summarising. It is rebuilding the comparison from scratch.
What survives disassembly is different. Original data survives, because Google has to cite where a number came from. Clear opinions survive, because a generated table cannot hold a point of view. And brand survives, because when the user sees five options in a table, they still pick the name they already trust.
Query fan-out is a visibility lottery with rules
The fan-out detail matters more than it sounds. When Google splits one question into several subtopic searches, it pulls sources for each piece. A page that answers one narrow subtopic extremely well can be included even if it would never rank for the head term.
That rewards depth over breadth. One definitive page on "how long does installation take for X" can earn a citation inside a much bigger answer. The broad "ultimate guide" competing with fifty other ultimate guides often earns nothing.
In practice, I would map the subtopics Google is likely to split a commercial question into. For a B2B software query, that might be pricing, integrations, implementation time, security and alternatives. Each one deserves a page or a clearly marked section that answers it directly, with numbers.
It also makes rankings a weaker proxy than ever. You can be cited in the answer without ranking in the top ten, and rank first without being cited. Measurement built only on position tracking now misses part of the picture.
What I would do this month
First, pick the ten queries that matter most commercially and run them in AI Mode. Screenshot what Google builds. Note whether you appear, how you are described, and whose numbers fill the table.
Second, restructure one key page so its core facts sit in plain HTML: a short spec block, clear prices or price ranges, explicit service areas, a clean FAQ. Nothing clever. Just extractable.
Third, publish one thing nobody else has. A benchmark, a survey of your own customers, a cost breakdown from real projects. Generated pages need inputs, and original data is the input that carries your name with it.
Google is building the page now. Your job is to make sure it cannot build a good one without you.