Dynamic AI layouts: how generative UI decides the way your content is shown

Generative UI is an interface that designs the format of each answer on the spot.

The verdict: with dynamic AI layouts, being cited isn’t enough; you also have to win visual prominence inside a layout you can’t predict.

You used to be able to screenshot a search results page and feel confident it would look roughly the same tomorrow.

Ten blue links, a featured snippet in its usual spot, recognisable SERP features. That stability is gone.

AI interfaces render the same question differently every time. That changes what you need to steer on.

What a dynamic AI layout is

A dynamic AI layout is generative UI in action. The interface produces the visual format of an answer based on three inputs:

• The content itself: a comparison, a process, a definition.
• The user’s context: device, history, supplied details.
• The assumed intent: buy, learn or compare.

One user sees a comparison table, another a flowing narrative, a third an interactive decision tree.

So the container your content appears in is no longer a fixed template. It’s a fluid decision made by the model.

This isn’t theory

On 18 November 2025, Google Research described generative UI in the Gemini app and in AI Mode.

In its “dynamic view”, the model designs and codes a custom interactive response for each prompt.

Developers build the same thing today. The Vercel AI SDK describes generative UI as connecting the results of a tool call to a React component, where the model decides when and how to use those tools.

This piece is a deep dive within the ultimate GEO guide, which covers generative engine optimization (GEO) from foundations to measurement.

Test your AI visibility: score your page in half a minute with our GEO check.

Generative UI: from fixed template to generated layout

Classic search results ran on fixed templates. The format varied by query type, but the variations were predictable.

You could optimise for a specific position within a known pattern. Our comparison of AI Overviews vs featured snippets shows how much that has shifted.

The three signals behind a layout

AI systems generate the layout instead. It emerges from three kinds of signals:

• Content-driven structure: a comparison becomes a table, a process a step-by-step plan.
• Context-driven adaptation: mobile gets a vertical, scannable format; desktop a richer multi-column view.
• Intent-driven optimisation: transactional intent gets action elements; comparative intent a structured comparison.

Pros and cons become a side-by-side view. A quick question yields a compact summary, a research question an extended version.

Informational intent, meanwhile, gets a readable narrative.

The same prompt, three layouts

You can already see the effect across engines. We gave three assistants the exact same prompt on 17 October 2026: explain GEO in three short bullet points for a B2B marketing manager.

ChatGPT returned three bold-led bullets and no sources:

ChatGPT free (temporary chat), 17 October 2026: three bullets, about 90 words, no sources.

Gemini added an intro sentence before its bullets:

Gemini (Google Workspace account, Flash model), 17 October 2026: an intro sentence plus three bullets, about 120 words, with no sources shown.

Perplexity put a source chip under every bullet:

Perplexity free, logged out, 17 October 2026: three bullets, a source chip on every bullet and 10 sources in total. The interface labels are in Dutch because we ran it from Belgium.

Same question, three containers. Only one of them gives a source any visible space.

What happens behind the scenes

The model roughly goes through six steps:

1. It synthesises the content.
2. It analyses the structure: list, comparison or process.
3. It assesses the user’s context.
4. It picks or generates a format.
5. It decides what gets visual emphasis.
6. It renders the whole thing.

Every one of those steps contains a judgement that touches your visibility.

Here’s that sequence as a picture, from raw content to the answer on screen:

Google’s AI features may also run query fan-out first: several related searches merged into one answer.

Why is visual prominence a new competitive dimension?

Because when layouts are dynamic, being selected is only half the contest.

The other half is how noticeable and emphasised your content is inside the rendered answer.

The four forms of prominence

We distinguish four forms:

• Position prominence: where you sit in the visual hierarchy.
• Space prominence: how much visual space you take up.
• Formatting prominence: how your content is styled.
• Interaction prominence: whether you’re tied to the primary action.

At the top of the answer is strongest. Inline in the flow is medium, and lower sections or collapsed areas are weakest.

Space runs from an extended treatment with its own section down to a brief mention. Formatting runs from highlighted or boxed out to greyed out and pushed aside.

Interaction means the first button, or a hidden action that only appears after expanding.

What decides prominence

Three factors weigh most:

• Relevance: a direct answer gets the most emphasis.
• Authority: recognised brands and expert sources can get a boost.
• Content structure: well-structured content can appear in rich formats.

Content structure is where you have the most influence. Unstructured content stays limited to a basic view.

How to build your text for extraction is worked out in content architecture for AI extraction. How engines pick their sources in the first place is in AI source selection.

Which formats can your content surface in?

Four formats come back most often, and each asks for its own optimisation approach.

Narrative summaries

The model writes prose that synthesises multiple sources.

What counts is whether you’re selected as the source for a core claim, how clear the attribution is, and how prominently your brand is named.

Your move: write clear, quotable statements with specific, referenceable data. Build brand-topic associations so your name comes up naturally.

Comparison tables

Structured comparisons of options. What counts is which column you’re in, which rows get included and how your values are displayed.

The first column usually catches the eye first.

Your move: make your differentiators table-worthy attributes. Deliver clean, comparable data points that are easy to extract.

Step-by-step plans

Procedural content in sequence. It comes down to which steps cite your content, and whether your solution surfaces at the right moment.

Your move: create complete procedures that break into clear steps. Place your solution at the logical point.

Decision trees

Interactive decision flows that guide the user through criteria. The point is which branches you appear in, and at what position: first recommendation or alternative.

Your move: document explicitly which criteria and use cases your solution fits. Then the decision logic can place you.

The four formats at a glance

How does personalisation make visibility plural?

Dynamic layouts make personalisation possible. The same question can render your content completely differently per user.

For example, picture three buyers asking the same question:

• An enterprise buyer sees your platform prominently, with enterprise features highlighted.
• A small business owner sees it as one option among SMB-friendly alternatives.
• A technical evaluator gets specifications and integration details blown up.

The signals behind it

That personalisation is fed by four kinds of signals:

• Explicit: settings and supplied context.
• Behavioural: search history and earlier interactions.
• Contextual: device, location and time.
• Inferred: the assumed role or use case.

You can’t steer those signals. But you can make sure you have something to offer every profile.

What to build

Concretely, four moves:

• Segment coverage: enterprise and SMB content, deep dives and business overviews.
• Use case specificity: industry-specific or role-specific scenarios personalisation can pick up.
• Persona alignment: separate content for the decision maker, technical evaluator and end user.
• Adaptable structure: modular sections at several levels of detail, segmented by audience.

Rule of thumb: a buyer type without its own page probably won’t see you prominently.

How do you optimise for flexible rendering?

You don’t know in advance which format your content ends up in. So build content that works everywhere.

We apply three principles.

Modular architecture

Make self-contained sections, each with its own value, clear boundaries and flexible combinability.

A module’s got to be able to surface without the rest of the page.

Multiple granularities

Offer the same information at three levels:

• a summary of 50 to 100 words
• a standard version of 200 to 400 words
• a detailed version of 500 words or more

Each level’s complete on its own. Then the model can pick what fits a quick or a deep question.

Format-agnostic core

Make sure your core information doesn’t depend on the format.

The meaning has to hold up as a table, a list or a narrative.

Your essential points shouldn’t lose prominence in any view.

Structure helps, magic schema doesn’t

Clean structure helps the model understand your content: extractable tables, consistent lists with clear labels, and markup that matches the visible text.

Our guide to structured data covers that markup.

But don’t expect a magic schema. Google says there are no additional requirements or special optimizations needed to appear in AI Overviews or AI Mode.

It’s the same build that helps when an AI agent reads your page in pieces: see grounding snippets.

How do you measure prominence over time?

You can’t steer what you don’t measure. Visual prominence is largely assessed qualitatively.

A simple routine

Five steps:

1. Query your target terms across different platforms.
2. Capture the answers as screenshots.
3. Score each on position, space, formatting and emphasis.
4. Compare against competitors.
5. Repeat on a fixed rhythm and track the change.

For example, a scoring sheet per prompt could look like this:

Our AI visibility audit uses the same idea at a larger scale.

The quantitative indicators

There are also indirect, quantitative indicators:

• Click-throughs from AI answers: Google counts AI Overviews and AI Mode in the Search Console Performance report under “Web”.
• Share of citations: how often you’re the cited source, see AI share of voice.
• First-mention rate: how often you’re named first.
• Featured treatment: how often you get your own section or box.

None of them measures prominence perfectly. Together they give a reliable picture of your direction, as our guide to AI visibility metrics explains.

Keep it per platform

Google AI Mode, ChatGPT and Perplexity each render according to their own patterns. What’s prominent in one format can fall away in another.

Our comparison of AI search engines shows how they differ, with engine-specific advice for ChatGPT and Perplexity.

Watch out: generative UI is still young.

Google’s researchers say it can take a minute or more, and that occasional inaccuracies remain.

Frequently asked questions

What is the difference between a dynamic layout and a classic search result?

A classic search result uses a fixed template: positions and formats are more or less set and vary predictably by query type. A dynamic AI layout generates the visual format on the spot, based on the content, the user and the context. As a result, the same question can produce a table, a narrative or a decision tree for different people.

Should I optimise my content for one specific format?

No, that’s exactly the risk. Because you don’t know in advance which format the model will pick, you don’t optimise for a single view but for flexible rendering. Build modular, self-contained sections at multiple levels of detail, with a core that keeps its meaning as a table, a list or prose.

How does personalisation affect my visibility?

Personalisation means the same question renders your content differently per user, depending on their assumed role, device and behaviour. You don’t steer those signals, but you improve your odds by covering multiple segments and use cases: enterprise and SMB, technical and business, decision maker and end user.

Does this also apply to agentic AI and autonomous assistants?

Yes. As AI assistants become more autonomous and assemble interfaces themselves, the way they show content becomes even more variable. The principles stay the same: modular content, visual prominence and coverage across segments. In that sense, dynamic layouts are a preview of how ever more fluid interfaces will present your content.

Need help?

Want to turn this into execution? See how we approach it with AI visibility.

Start with our AI citation checklist, or score a page with the GEO check.

Further reading

• How the new AI search architecture works (retrieval, generation and RAG)