AI search is already changing brand visibility and click-through rates, whether that’s through people asking ChatGPT a question or seeing an AI Overview at the top of Google’s results. Now, Google AI Mode is speeding that change up.

AI Mode launched in the UK on 28 July 2025 and has since expanded to more than 200 countries and territories and over 35 languages.

Instead of a ranked list of links that users click through to piece information together, AI Mode answers queries with conversational paragraphs, backed by linked citations. Bing already offers something similar with Copilot Search, but Google’s market dominance means AI Mode is likely to do far more to bring AI search into the mainstream.

All of this gives marketers possibly their strongest incentive yet to adapt quickly, so they can maintain or even improve their brand’s visibility rather than watch it erode.

That urgency makes an easily implemented, templated approach very tempting. Predictably, there are already plenty of supposedly “AI-optimised” content templates, frameworks and page layouts out there, all promising to improve your chances of being cited in AI-generated responses.

But before you invest in one of these template-based approaches, do they actually work?

The short answer, based on our research, is no. Template-led “AI-optimised” layouts don’t reliably improve your chances of being cited.

The more nuanced answer is that some aspects of page structure do matter – just not the ones the templates focus on.

Here are five things you need to know to understand what really matters when it comes to AI search.

1. Why content templates won’t impress LLMs

When discussing AI search, we always need to distinguish between Google’s AI products (AI Mode and AI Overviews) and LLM systems such as ChatGPT, Claude and Perplexity. The way they operate and validate information is fundamentally different.

When systems such as ChatGPT or Claude fetch a page to answer a question, the page is typically converted to plain text first. HTML tags, CSS styling and JavaScript drop away, so the model sees blocks of text – not your carefully crafted page formatting, navigation or visual hierarchy.

So for these systems, there’s no such thing as an AI-optimised template, because your template is invisible to them.

2. Why Google AI Mode is not like ChatGPT

Unlike standalone LLMs, Google’s AI products combine Google’s traditional search infrastructure with LLM processing.

Google AI Mode and AI Overviews use a “grounding” mechanism, checking their responses against Google’s search infrastructure to reduce the risk of hallucinations and improve accuracy. This is why your traditional SEO foundations still matter for visibility in Google’s AI products.

ChatGPT and similar systems don’t share Google’s search index, Knowledge Graph and ranking systems, even when they add browsing or search tools. Instead, they rely on vast amounts of training data from sources such as Common Crawl, supplemented with real-time search to find fresh sources and validate information.

That means they’re working with the same fallible web as the rest of us, without the authority signals, knowledge graphs, and other systems that power Google’s search infrastructure.

This difference has strategic implications. Content that performs well in traditional Google search has a better chance of being cited by Google’s AI products, because both draw on the same underlying infrastructure. Strong E-E-A-T signals, topical authority and technical SEO fundamentals are still relevant.

3. Schema markup: the indirect truth

If a page is converted to plain text before an LLM processes it, any schema markup is usually stripped out too. So the model typically can’t use your schema to understand how your content is structured.

Yet Google, Microsoft and OpenAI have all suggested that structured data matters for AI search. Google’s May 2025 guidance on succeeding in AI search recommends making sure structured data matches your visible content. Bing’s Fabrice Canel said in March 2025 that schema markup helps Microsoft’s LLMs understand content. And OpenAI says ChatGPT’s shopping results draw on structured product data from third-party providers.

No wonder marketers get confused.

In our view, schema’s clearest AI-related value is indirect. LLM-based systems use search data from Google, Bing and others, plus information from content partners, to identify, prioritise and validate information before they respond. Structured data feeds into that infrastructure – but it’s no guarantee that an LLM “reads” the JSON-LD on your page.

For example, while ChatGPT doesn’t directly “read” schema markup when processing a page, its shopping results draw on third-party websites and data feeds that use structured metadata for pricing, product descriptions and reviews.

So schema markup matters in the same way it always has: it gives algorithms the structured data they need to define entities and relationships, add context and build knowledge graphs. You now also get an extra, indirect benefit from AI systems drawing on the same information through third-party sources.

4. Why “above the fold” doesn’t matter in AI Mode

Content marketing advice often recommends putting the most important information within the first 800 pixels of a webpage, above the fold.

The logic seems sound. A short summary or scene-setting opening quickly confirms what the content is about, so surely AI systems would also benefit from key information near the top of the page that signals its relevance to a query.

We tested this hypothesis in our AI Mode research. AI Mode often includes text-fragment links in its responses, which take the user straight to the relevant highlighted text in the cited page. By measuring the exact pixel depth of each highlighted fragment, we could see whether AI Mode was more likely to surface content from the first 800 pixels.

It wasn’t. AI Mode pulls information from anywhere on the page, regardless of pixel depth. What counts is the most relevant passage, not where it sits in the layout.

5. Descriptive subheadings do matter in AI Mode

While pixel depth wasn’t a factor, our research did reveal one clear pattern: AI Mode tends to select text fragments made up of a subheading followed by the first line of the text beneath it.

This suggests AI Mode uses subheadings to navigate the information hierarchy of a page.

But we already knew that descriptive headings and strong opening lines matter to Google. While it’s not new advice, it reinforces long-standing content best practice.

Conclusion: The real strategy for AI visibility

You might be disappointed to learn there’s no magic template to optimise your content for AI. There is no silver bullet.

On the other hand, you probably don’t need a costly rebuild around an “AI template”. Your budget is better spent on hierarchy, clarity and topical depth – things that already serve SEO.

What worked for SEO still matters for AI search, even if some of the impact is indirect. Focus on information hierarchy, descriptive subheadings and comprehensive topic coverage.

The best approach is the same one that’s worked for years: create well-structured, authoritative content that genuinely serves user intent.

Need help improving your AI visibility? Find out more about our AI search optimisation services, or get in touch.