Google does not hold many events where its search engineers speak candidly about how the system actually works. When it does, practitioners should pay close attention. At Search Central Live in Milan this week, Google covered chunking, site-wide signals, paywalled content, commodity versus non-commodity content, and what clicks from AI Overviews actually look like. Each of these topics connects directly to how brands earn - or fail to earn - visibility in AI-generated results.
Chunking Is Not Just a Technical Detail
Chunking refers to how Google breaks content into discrete, processable segments before feeding it into AI systems. This is not a minor implementation detail. It is the mechanism that determines whether a specific claim, definition, or answer within your content gets surfaced in an AI Overview - or gets ignored entirely.
If your content is structured as dense, undifferentiated prose, Google's chunking process will struggle to extract clean, citable segments from it. Content that uses clear headings, logical paragraph breaks, and self-contained explanations is far easier to chunk usefully. This is why the structural advice that has long applied to featured snippets carries over so directly to AI Overviews. The underlying mechanism is similar, even if the output looks different.
Practically, this means reviewing your most important pages for chunk-readiness. Can any individual paragraph in your content stand alone as a meaningful answer to a question? If not, restructure. Short, purposeful paragraphs with a clear subject per section are not just good writing practice - they are now a technical requirement for AI citation.
Site-Wide Signals Still Matter - Perhaps More Than You Think
Google's Milan briefing touched on site-wide signals, which is a reminder that AI Overviews do not evaluate individual pages in isolation. The broader trust and quality picture of your entire site feeds into whether Google is willing to cite you at all. A handful of strong pages cannot compensate for a site that carries thin content, poor technical hygiene, or inconsistent quality across its sections.
This is particularly relevant for brands that have grown their content output rapidly over the past few years, often through scaled AI content programmes. If those programmes produced large volumes of low-quality or duplicative material, that content is now acting as a site-wide drag. The solution is not simply to produce more high-quality content - it requires auditing and either improving or removing what is already there.
Site-wide signals also interact with topical authority. Google is assessing not just whether a page is well-written, but whether the site as a whole demonstrates genuine expertise in the subject area. Brands that cover a topic superficially across many pages will score lower than those with fewer, deeper pieces that demonstrate real command of the subject.
Commodity Content Will Not Get You Cited
One of the more direct points from the Milan event was the distinction between commodity and non-commodity content. Commodity content is the kind of material that appears in near-identical form across thousands of sites - generic definitions, basic how-to guides, broad category overviews. Google has little reason to cite any specific source for this type of content because the information is undifferentiated.
Non-commodity content, by contrast, offers something that cannot easily be found elsewhere in the same form. Original analysis, proprietary data, expert opinions, specific case studies, detailed process documentation - these are the formats that give Google a reason to cite your source rather than synthesise around it. The AI Overview needs to be able to point somewhere meaningful. That means your content needs to be meaningfully distinct.
For UK brands, this is a useful filter when planning content. Before commissioning any new piece, the question worth asking is: does this exist in substantially the same form on fifty other sites? If the honest answer is yes, the content is unlikely to contribute to AI visibility regardless of how well it is written or optimised at a technical level.
Paywalled Content and What Google Actually Does With It
Google addressed paywalled content at the Milan event, which is an important topic for publishers and subscription-based businesses. The core position is consistent with what Google has communicated before - properly implemented paywall schema allows Google to understand that content exists and to index it appropriately, without requiring full exposure of the content to anonymous users.
For news publishers in particular, this matters for AI Overview visibility. If subscription content is not correctly marked up, Google may not treat it as a trusted source in the way it otherwise would. The structured data implementation - specifically the CreativeWork schema with the isAccessibleForFree and hasPart properties, which are established industry practice for communicating paywall status - is the mechanism that communicates paywall status to Google in a way it can process. Note that these specific properties were not detailed at the Milan event itself.
The broader implication for subscription businesses is that being behind a paywall is not itself a barrier to AI citation, provided the technical implementation is correct. What matters is that Google can understand your content's existence, authority, and structure - even if it cannot reproduce the full text. Brands with gated content should audit their structured data to confirm they are not inadvertently signalling the wrong thing.
What AI Overview Clicks Actually Look Like
Google spoke at the Milan event about what click behaviour from AI Overviews looks like in practice. This connects directly to a question that every brand investing in GEO wants answered - does appearing in an AI Overview actually drive traffic? The signal from Google is that clicks do come from AI Overviews, but the behaviour is different from traditional organic clicks.
Google Search Console includes AI-related settings and data that can help teams understand how their content is performing in this context - a topic covered in more depth in our recent piece on GSC's AI Overview data. The presence of this data within GSC confirms that Google considers AI Overview performance to be a distinct and measurable dimension of search visibility - not simply an extension of organic rankings. Teams that are not yet using this data are missing visibility into a meaningful portion of their search presence.
The practical takeaway is not to assume that AI Overview citations are purely a brand awareness play with no conversion value. The click behaviour is different - users who click through from an AI Overview have typically already received some information and are seeking more detail or are ready to act. It is plausible that the intent level of AI Overview visitors is higher than average organic traffic, though Google did not confirm this directly at Milan and it remains an inference worth testing rather than an established finding. Tracking this traffic distinctly and analysing its on-site behaviour is worth prioritising if you have not already done so.
Putting It Together for Your AI Visibility Strategy
The Milan briefing did not introduce entirely new concepts, but it did provide direct confirmation of several things that practitioners have been working with as educated assumptions. Chunking matters. Site-wide quality matters. Content differentiation matters. Technical markup for paywalled content matters. These are not independent optimisation tasks - they are interconnected dimensions of the same underlying question: does Google trust your site enough, and find your content distinctive enough, to cite it in AI-generated results?
For brands currently working on GEO strategy, this briefing provides a useful framework for prioritising effort. Start with site-wide quality - that is the foundation. Then focus content investment on non-commodity material that genuinely adds something. Ensure your technical structure supports clean chunking. And if you have paywalled content, confirm your structured data implementation is correct. None of this is quick, but each element builds directly on what Google has now stated openly.