This is a submission for the Google I/O Writing Challenge
There is a question that Google's I/O 2026 keynote answered without anyone on stage saying it out loud.
The question is: at what point does a search engine stop needing the web it searches?
AI Mode — the thing Google demonstrated with visible pride — doesn't just summarize your content anymore. It generates answers. It generates interactive visuals. It builds entire experiences on the fly. And it does this personalized to the query, in a format no static webpage can match.
Every content creator watching that demo should have felt the same thing — not panic, but a shift. The ground rules just changed again, and we don't know yet how far.
A disclosure before we go further: I'm not writing this from the sidelines. For the past six months, I've been building an AI content pipeline — a system that generates SEO-optimized articles through four AI personas, each with a distinct voice, with human editorial oversight, claim verification, and structured content planning. Before that, I spent years as a copywriter building WordPress sites and teaching myself SEO by doing it wrong enough times to start getting it right.
We're transparent about what the site is: the content is AI-generated and we say so — in the article footers, on the editorial standards page, in the schema markup. We think that honesty is more valuable than pretending a human wrote every word. The site has been live for about a month, Google has had no issues indexing it, and roughly a third of the planned articles are published. We're adding more slowly, deliberately, because each batch teaches us something about what works and what doesn't.
In parallel, we're designing new content pipelines for specific developer and analyst roles — focused not on broad concepts but on concrete problems these roles actually face in their daily work. What to build next is the hardest question right now, because the landscape shifts faster than any content plan can anticipate. Building a content system in 2026 means rebuilding parts of it every few weeks.
Which is exactly why the I/O demo hit the way it did. I was learning AEO and query fan-out patterns, optimizing infographics for Google Image search, fine-tuning article structure for AI citation — to get good at a game whose rules, it turns out, were about to change again. Not someday. This summer.
Three eras, each shorter than the last
The first era was links. You wrote something good, Google ranked it, users clicked. The business model was clear: quality content → visibility → traffic → revenue. This lasted roughly twenty years.
The second era was snippets and AI overviews. Google started answering questions directly, pulling fragments from your page. You still got a citation, sometimes a click. The game became: structure your content so the machine can extract a clean answer. AEO — answer engine optimization — is this era's discipline. It assumed the machine still needed your words.
The third era is what Google just showed us. The machine doesn't just extract your words. It generates its own content, its own visuals, its own interactive experiences. The competition is no longer about who gets the click. It's about who creates the experience. And for the first time, the search engine itself is a competitor in that race.
Each era is shorter. Each era changes what "being visible" means. The direction is worth naming honestly, even if the destination isn't clear yet.
The format war nobody expected
Here is what Google actually said on the I/O stage:
"Search can build you the ideal format exactly for your question, completely custom, on the fly. We're talking dynamic layouts, interactive widgets, entire experiences, all created just for you. This is agentic coding at the scale of search."
The demo showed a student asking "how do black holes affect space time?" and receiving an interactive visual — not a link to a page with a diagram, but a generated, manipulable visualization built on the fly. The student followed up with a more specific question about binary black holes and gravitational waves, and Search dynamically built a brand new interactive visual in real time.
This is not a better snippet. This is generative UI — the search engine becoming an application.
Google announced rollout for summer 2026. We haven't seen the real product yet — only a stage demo. And stage demos have a history of overpromising. NotebookLM already generates nine types of output from uploaded sources — podcasts, mindmaps, reports, presentations — and the quality varies. Generating a good infographic from a well-scoped source is a solved problem. Generating one from the entire web, on the fly, at the quality level of that demo? That's a harder claim to evaluate without seeing it work at scale.
But here's what matters regardless of execution quality: Google is no longer competing for your traffic. It's competing for your format. Even if the first version is mediocre, the direction is clear. The search engine wants to be the experience, not the directory to the experience.
What this means for informational content
Text-based AEO has a logic you can work with. Structure your content clearly, lead with the answer, use schema markup, earn citations. It's not easy, but it's legible. You can see the rules and play by them.
Visual AEO doesn't exist yet — and it might never exist in a form content creators can influence.
When Google generates an interactive explainer in response to a query, it draws from the structured information across thousands of pages and renders its own experience — interactive, personalized, potentially better than any static image. For sites that explain technical concepts — how attention works, how state space models replace quadratic scaling, how RAG pipelines process queries — this is the most direct challenge. These are exactly the kind of conceptual visualizations that generative UI could build on the fly.
Could. Not will. We don't know yet whether it handles complex technical concepts or only "101" level demos. We don't know whether it cites sources for generated visuals. We don't know whether it works for every query or only for a narrow set of educational topics.
These are questions that will have answers by late 2026. Right now, they're open.
What can't be generated (and the catch)
This is where the essay could turn reassuring. Focus on first-party data, original research, and personal experience — the things AI can't fabricate. And that's true.
But it deserves a more careful statement.
AI can't fabricate your specific experience. It can't invent the fact that your content pipeline costs $8.12 per entity, or that you iterated a Gemini image prompt three times before it stopped putting your face at 40% of the canvas. It can't know that you tried putting everything in CLAUDE.md and the model got noticeably worse. These are things only you can source.
The catch: these things only matter if someone is looking for them. If the query is "how do Claude Skills work," Google can answer that without you. If the query is "what was it like building a Claude Skill for a Hugo content pipeline," you're the only source — but the audience is smaller.
First-party data is a moat, but it's a moat around a castle whose size you don't control. The question is whether that castle becomes more valuable as everything around it gets commoditized — or whether it just becomes more lonely.
The honest middle: what's changing, what's not, what we don't know
What's probably changing:
Generic explainer content — "What is RAG?", "How does attention work?" — is losing its discovery value. Google will answer these with or without your page. If your strategy depends entirely on ranking for these queries, diversify.
What's probably not changing yet:
Mid-funnel evaluation content — "RAG vs. fine-tuning for my use case" — still requires specific constraints, infrastructure context, cost structures that Google can approximate but not personalize without your data. AEO still works here. For now.
What's growing in relative value:
Experience-based content. Not because it's getting better — but because everything around it is getting commoditized. When the machine can generate any explanation, the only scarce thing is the explanation it can't generate: the one that comes from having done the thing.
What we genuinely don't know:
- Will Google cite sources for generated visuals, or will they be unattributed?
- Does generative UI work for complex technical topics or only accessible ones?
- Will there be a way to influence what visual gets generated — a kind of visual AEO — or is it a closed system?
- How good is the real product vs. the stage demo?
These aren't rhetorical questions. They're the things worth watching when the product launches in summer 2026.
What builders should probably do
I don't have a strategy. I have an observation and a set of bets.
The observation: the content that survived every previous Google shift — Panda, featured snippets, AI overviews — was content that had something Google couldn't replicate. In 2012, that was quality. In 2020, that was structure. In 2026, it's experience and data that only exists because you did the work.
The bets:
Document process, not just conclusions. The builders who share the messy middle — failed attempts, surprising costs, workflow decisions — remain citable even when the machine can generate the conclusion itself. The conclusion without the process is just another generated answer. The process is the proof.
Watch the launch closely. When generative UI rolls out, test your key queries. What does Google generate? Does it cite sources? Is the quality real or demo-grade? The answers will tell you more about the next two years than any prediction written today.
Don't optimize for a format war you can't win. If Google generates interactive infographics, competing with better static infographics is a losing game. Compete on the axis where you have an advantage: specificity, experience, data nobody else has. How do you compete on visuals when the search engine generates its own? We don't have that answer yet. We're looking for it.
Keep building anyway. The web has survived every prediction of its death. It survived apps, social media, and featured snippets. It will probably survive generative UI too — just in a different shape, serving a different function. The shape isn't clear yet. But the builders who are still building when it becomes clear will be the ones who define it.
This essay was written by a human, with AI assistance, about the uncertain future of content in a world where the search engine generates its own. Whether you found it through Google, through a direct link, or through a recommendation from someone who read it — that path is already part of the story this essay is trying to tell.
Top comments (0)