Google Antitrust Case: AI Overviews Utilize FastSearch, Not Links

Google’s ‘Antitrust Case’ focuses on AI overviews that leverage FastSearch, moving away from link-based methods in search and advertising.

An excerpt from Google’s antitrust case just gave us all a big revelation, indicating that the AI Overviews feature does not use the classic signal of links when ranking pages.

Instead, it relies on Google’s internal search algorithm called FastSearch, which focuses on speed and semantic relevance rather than on using link analysis to obtain relevance.

Hint from the Antitrust Memorandum Opinion

In his LinkedIn profile, the founder of SERPrecon, Ryan Jones, identified one revealing passage featuring a new Memorandum Opinion discussing how Google is training its Gemini AI models.

Despite many people’s belief that links are still key when it comes to ranking the web pages retrieved by an AI model from a search query via an internal search engine. So, when someone asks Google’s AI Overviews a question, the system queries Google Search and generates a summary based on those results.

However, it seems that is not how things function at Google. The company uses a distinct algorithm that retrieves fewer web documents while doing so more quickly.

The passage reads:

“To ground its Gemini models, Google uses a proprietary technology called FastSearch. Rem. Tr. at 3509:23–3511:4 (Reid). FastSearch is based on RankEmbed signals—a set of search ranking signals—and generates abbreviated, ranked web results that a model can use to produce a grounded response. Id. FastSearch delivers results more quickly than Search because it retrieves fewer documents, but the resulting quality is lower than Search’s fully ranked web results.”

Ryan Jones shared these insights:

“This is interesting and confirms both what many of us thought and what we were seeing in early tests. What does it mean? It means for grounding Google doesn’t use the same search algorithm. They need it to be faster but they also don’t care about as many signals. They just need text that backs up what they’re saying.

…There’s probably a bunch of spam and quality signals that don’t get computed for fastsearch either. That would explain how/why in early versions we saw some spammy sites and even penalized sites showing up in AI overviews.”

He goes on to share his opinion that links aren’t playing a role here because the grounding uses semantic relevance.

FastSearch Overview

The Memorandum also clarifies that FastSearch returns only a finite number of results:

“FastSearch is a technology that rapidly generates limited organic search results for certain use cases, such as grounding of LLMs, and is derived primarily from the RankEmbed model.”

The Memorandum (Elsewhere) explains that Google’s AI Overviews do not make the full web index a query point, but rather run on a separate, faster algorithm, called FastSearch, that returns fewer documents per hour due to its higher speed.

The Memorandum has a passage that explains:

“At the other end of the spectrum are innovative deep-learning models, which are machine-learning models that discern complex patterns in large datasets. …(Allan)

…Google has developed various “top-level” signals that are inputs to producing the final score for a web page. Id. at 2793:5–2794:9 (Allan) (discussing RDXD-20.018). Among Google’s top-level signals are those measuring a web page’s quality and popularity. Id.; RDX0041 at -001.

Signals developed through deep-learning models, like RankEmbed, also are among Google’s top-level signals.”

RankEmbed Model Background

At the heart of the system is RankEmbed, a deep-learning technique that sifts through massive datasets to detect what the researchers call “latent semantic information” and relationships among data points, without understanding them like a person.

Of significance, RankEmbed’s ranking is based on “user-side” data, such as click behaviors, as will be evident from the description above, and disclosed examples.

The Memorandum frames RankEmbed as:

“User-side Data used to train, build, or operate the RankEmbed model(s); “

Elsewhere, it shares:

“RankEmbed and its later iteration RankEmbedBERT are ranking models that rely on two main sources of data: _____% of 70 days of search logs plus scores generated by human raters and used by Google to measure the quality of organic search results.”

Then:

…RankEmbedBERT needs to be retrained to reflect fresh data…”

“The RankEmbed model itself is an AI-based, deep-learning system that has strong natural-language understanding. This allows the model to more efficiently identify the best documents to retrieve, even if a query lacks certain terms. PXR0171 at -086 (“Embedding based retrieval is effective at semantic matching of docs and queries”);

…RankEmbed is trained on 1/100th of the data used to train earlier ranking models yet provides higher quality search results.

…RankEmbed particularly helped Google improve its answers to long-tail queries.

…Among the underlying training data is information about the query, including the salient terms that Google has derived from the query, and the resultant web pages.

…The data underlying RankEmbed models is a combination of click-and-query data and scoring of web pages by human raters.

AI Search and Ranking Implications

This change means that AI search does not depend on the old link structure of the web graph, instead concentrating more on speed and relevance while maintaining the philosophical connection to semantic models.

Ryan Jones suggests that Google has a different index in FastSearch whose main criteria are frequently visited, “quality” sites.

The scale of the web also makes it prohibitive for human raters to label examples that distinguish between good and bad content to help AI systems learn, which is why we exploit sparingly utilized human rater data that creators used to train these models.

Final Thoughts

This transition from link-based ranking in AI Overviews to more recent behaviourally informed and semantically driven models such as FastSearch and RankEmbed, at Google, marks a structural shift in how search results are produced for AI-powered searches.