website SEOWebsite Traffic
Download your cheat sheet and checklist to start building content that works harder.
In this exclusive session, you’ll learn about the new SEO playbook for AI.
In this exclusive session, you’ll learn about the new SEO playbook for AI.
In this exclusive session, you’ll learn about the new SEO playbook for AI.
This template is your no-nonsense roadmap to a flexible, agile social media strategy.
In this exclusive session, you’ll learn about the new SEO playbook for AI.
Google’s June update is over. Two interesting changes at Google may partially explain what changed.
Google’s June 2025 Core Update just finished. What’s notable is that while some say it was a big update, it didn’t feel disruptive, indicating that the changes may have been more subtle than game changing. Here are some clues that may explain what happened with this update.
Although a lot of people are saying that the June 2025 Update was related to MUVERA, that’s not really the whole story. There were two notable backend announcements over the past few weeks, MUVERA and Google’s Graph Foundation Model.
MUVERA is a Multi-Vector via Fixed Dimensional Encodings (FDEs) retrieval algorithm that makes retrieving web pages more accurate and with a higher degree of efficiency. MUVERA was announced by Google on June 25 of this year. But the research paper was published on arXiv.org in May 2024, which is unusual because the paper and announcements are usually done closer together in time.
In some cases the announcement coincides with the implementation of an algorithm. An example is the announcement of “Optimizing LLM-based trip planning” on June 7th, three months after the implementation of the algorithm within Google in March. It’s possible that this algorithm has already been in place before the Google update. But we don’t really know.
The notable part for SEO is that it is able to retrieve fewer candidate pages for ranking, leaving the less relevant pages behind and promoting only the more precisely relevant pages.
This enables Google to have all of the precision of multi-vector retrieval without any of the drawbacks of traditional multi-vector systems and with greater accuracy.
Google’s MUVERA announcement explains the key improvements:
“Improved recall: MUVERA outperforms the single-vector heuristic, a common approach used in multi-vector retrieval (which PLAID also employs), achieving better recall while retrieving significantly fewer candidate documents… For instance, FDE’s retrieve 5–20x fewer candidates to achieve a fixed recall.
Moreover, we found that MUVERA’s FDEs can be effectively compressed using product quantization, reducing memory footprint by 32x with minimal impact on retrieval quality.
These results highlight MUVERA’s potential to significantly accelerate multi-vector retrieval, making it more practical for real-world applications.
…By reducing multi-vector search to single-vector MIPS, MUVERA leverages existing optimized search techniques and achieves state-of-the-art performance with significantly improved efficiency.”
A graph foundation model (GFM) is a type of AI model that is designed to generalize across different graph structures and datasets. It’s designed to be adaptable in a similar way to how large language models can generalize across different domains that it hadn’t been initially trained in.
Google’s GFM classifies nodes and edges, which could plausibly include documents, links, users, spam detection, product recommendations, and any other kind of classification.
This is something very new, published on July 10th, but already tested on ads for spam detection. It is in fact a breakthrough in graph machine learning and the development of AI models that can generalize across different graph structures and tasks.
It supersedes the limitations of Graph Neural Networks (GNNs) which are tethered to the graph on which they were trained on. Graph Foundation Models, like LLMs, aren’t limited to what they were trained on, which makes them versatile for handling new or unseen graph structures and domains.
Google’s announcement of GFM says that it improves zero-shot and few-shot learning, meaning it can make accurate predictions on different types of graphs without additional task-specific training (zero-shot), even when only a small number of labeled examples are available (few-shot).
Google’s GFM announcement reported these results:
“Operating at Google scale means processing graphs of billions of nodes and edges where our JAX environment and scalable TPU infrastructure particularly shines. Such data volumes are amenable for training generalist models, so we probed our GFM on several internal classification tasks like spam detection in ads, which involves dozens of large and connected relational tables. Typical tabular baselines, albeit scalable, do not consider connections between rows of different tables, and therefore miss context that might be useful for accurate predictions. Our experiments vividly demonstrate that gap.
We observe a significant performance boost compared to the best tuned single-table baselines. Depending on the downstream task, GFM brings 3x – 40x gains in average precision, which indicates that the graph structure in relational tables provides a crucial signal to be leveraged by ML models.”
It’s not unreasonable to speculate that integrating both MUVERA and GFM could enable Google’s ranking systems to more precisely rank relevant content by improving retrieval (MUVERA) and mapping relationships between links or content to better identify patterns associated with trustworthiness and authority (GFM).
Integrating Both MUVERA and GFM would enable Google’s ranking systems to more precisely surface relevant content that searchers would find to be satisfying.
Google’s official announcement said this:
“This is a regular update designed to better surface relevant, satisfying content for searchers from all types of sites.”
This particular update did not seem to be accompanied by widespread reports of massive changes. This update may fit into what Google’s Danny Sullivan was talking about at Search Central Live New York, where he said they would be making changes to Google’s algorithm to surface a greater variety of high-quality content.
Search marketer Glenn Gabe tweeted that he saw some sites that had been affected by the “Helpful Content Update,” also known as HCU, had surged back in the rankings, while other sites worsened.
Although he said that this was a very big update, the response to his tweets was muted, not the kind of response that happens when there’s a widespread disruption. I think it’s fair to say that, although Glenn Gabe’s data shows it was a big update, it may not have been a disruptive one.
So what changed? I think, I speculate, that it was a widespread change that improved Google’s ability to better surface relevant content, helped by better retrieval and an improved ability to interpret patterns of trustworthiness and authoritativeness, as well as to better identify low-quality sites.
Google MUVERA
Google’s Graph Foundation Model
Google’s June 2025 Update Is Over
Featured Image by Shutterstock/Kues
I have 25 years hands-on experience in SEO, evolving along with the search engines by keeping up with the latest …
Join 75,000+ Digital Leaders.
Learn how to connect search, AI, and PPC into one unstoppable strategy.
Join 75,000+ Digital Leaders.
Learn how to connect search, AI, and PPC into one unstoppable strategy.
Join 75,000+ Digital Leaders.
Learn how to connect search, AI, and PPC into one unstoppable strategy.
In a world ruled by algorithms, SEJ brings timely, relevant information for SEOs, marketers, and entrepreneurs to optimize and grow their businesses — and careers.
Copyright © 2025 Search Engine Journal. All rights reserved. Published by Alpha Brand Media.