Some SEO concepts arrive quietly. No announcement, no documentation update, just a pattern that starts showing up in the data, and once you see it, you cannot unsee it. Grounding chunks may be one of those.
Recent research by Dan Petrovic offers a rare look into how Google’s AI systems, particularly those powered by Gemini, actually process content. Not full pages. Not complete articles. Fragmented, prioritised segments of text, selected based on criteria most publishers have never considered.
Once you understand that, many things start making more sense.
Google Doesn’t Read Your Page. It Samples It
The idea behind grounding chunks is deceptively simple.
For each query, Google works with a fixed processing budget of roughly 2,000 words. That budget is not per page. It spans all sources contributing to a given answer. Your content is not competing for a ranking position in isolation. It is competing for a share of a constrained word pool, alongside every other source the system considers relevant.
Most of that content never gets read. On average, only around 300 to 500 words are actually extracted from any single page. The rest, regardless of quality or optimisation, does not make it into the model’s consideration.
That is a different kind of visibility problem than SEO has traditionally dealt with.
Ranking Still Matters, but Differently
Higher-ranked pages receive a larger share of the available word budget. The top result captures roughly 28% of the total allocation. Each lower position gets less.
This means rankings still function as a gatekeeper for AI visibility, but the mechanism is different from what most people assume. Position does not determine whether you appear or not. It determines how much of your content the system actually reads. A page in position one does not just show up more often. More of what it says gets considered.
That is a subtle distinction, but it changes how you think about the value of ranking improvements.
Why More Content Doesn’t Always Mean More Visibility
Long-form content has been a dependable SEO approach for years. More depth, more coverage, more keyword surface area. Grounding chunk data puts pressure on that assumption.
Content length does improve the number of relevant words extracted, but only to a point. Past roughly 2,000 to 3,000 words, the gains flatten. Meanwhile, the proportion of content actually selected drops sharply as pages get longer.
A shorter page might have more than half its content represented in extracted chunks. A longer page, with more total information, might have only a small fraction included. More words on the page does not mean more words in the answer. For specific sections, it can actually mean less.
The Emergence of “Chunk-Level SEO”
If Google evaluates content in segments rather than as whole pages, your optimisation needs to work the same way. That is the core idea behind chunk-level SEO.
Stop thinking in terms of pages. Start thinking in terms of:
- Sections that can stand independently
- Paragraphs that directly answer specific questions
- Blocks of content that are contextually complete
Each of these is a potential entry point into an AI-generated response. Key insights buried three scrolls into a long article are unlikely to be selected. Information that is accessible, clearly defined, and self-contained has a much better shot. A section that requires surrounding context to make sense is a section that gets skipped.
Sliding Windows and Contextual Extraction
The research points to Google using something similar to sliding windows when processing content.
The system does not read a page from top to bottom in one pass. It scans through segments, evaluating relevance independently as it goes. This is consistent with how large language models handle long inputs generally, breaking them into pieces small enough to process efficiently.
For your content, this means local context within each section is not optional. A paragraph that relies on the three preceding paragraphs for meaning will not perform well when evaluated in isolation. Each section needs to carry its own meaning. That changes how you write introductions, transitions, and explanations. Dependency is a liability. Clarity is not just a readability goal. It is a selection factor.
What This Means for Content Strategy
None of these points argues against long-form content. It argues for building it differently.
Length still contributes, but only when the structure supports meaningful segmentation. A 2,000-word article with clearly defined, self-contained sections will extract better than a 4,000-word article where the same insights are spread thin across the text.
Precision matters more than coverage. Specific, high-value answers inside clearly identifiable sections are what get selected. Broad topic coverage without clear segmentation produces content that is technically present but functionally invisible to extraction systems.
Other structural elements reinforce these principles:
- Internal linking reinforces topical clusters
- Headings provide structural cues for segmentation
- Formatting improves readability for both users and systems
These are not decorative choices. They directly affect how your content gets chunked and selected.
A Subtle Alignment With Broader Google Systems
Grounding chunks fit cleanly with patterns confirmed elsewhere, in leaked documentation, in observed ranking behaviour, and in how AI Overviews select source material.
Google consistently moves toward:
- Evaluating content at a granular level
- Prioritising usefulness over volume
- Using behavioral and contextual signals to refine outputs
Grounding chunks are not a standalone feature. They are one expression of a broader shift toward modular content evaluation that is visible across multiple parts of Google’s systems.
Conclusion: Visibility Is Now Fragmented
The assumption that a page ranks as a whole is losing accuracy.
In AI-driven search, your content gets broken into parts, evaluated in segments, and selectively included in responses. The best page on a topic does not automatically get the most representation. The page with the most extractable, clearly structured insights does.
The question worth asking about any piece of content is no longer only what it says. The question is which parts are worth selecting and if they are easy to find and understand.
Take Action
If your content is long but underperforming in AI-driven search features, start with the structure.
Identify:
- Whether key insights are clearly defined
- If sections can stand independently
- Where important information might be buried
Optimisation is no longer page-level. It is section-level, sometimes sentence-level. That is where the next gains come from.