The U.S. District Court for the District of D.C. has released a Memorandum Opinion, a written decision on the antitrust lawsuit against Google.
Document No. 1436, the main contents of which are to prohibit exclusive agreements for the distribution of search, Chrome, etc.; to determine that a split of Chrome is not necessary; and to impose an obligation to share search index and user interaction data to a certain extent.
It is a primary source.
Source: United States v. Google LLC Memorandum Opinion (Document 1436)
This is not an “ah-ha” story by itself, but since there are some terms that are new to me, I will summarize them as my own notes as well.
Although this is something we have known for some time in the field of SEO and web management, I think it is easy to use this information in various places.
目次
Why should we pay attention to the text of the decision?
The focus of the published decision is to clarify exactly what data Google is using specifically, for what purpose, and how it is being incorporated into its search engine and generative AI.
It may be that now that we know the name, what’s the point, but I think it’s good to have a common language to make things easier to understand.
In particular, the clear indication that Google makes extensive use of user behavior (clicks, dwell time, immediate return, etc.) in evaluating search results.
In attracting visitors from search engines, I would say that more emphasis should be placed on designing a “sense of task completion” (shortening the diameter of information arrival, internal leads, FAQ/step UI) as well as click rate and stay.
After that, it would be nice to have a list of the many internal codenames of the algorithms and projects used in the company. It makes it easier to explain.
Two points to keep in mind from the text of this decision
- The importance of user behavior data is clear
The ruling text makes it clear that Google continuously uses user behavior data in many of its search engine processes, from crawl prioritization, index freshness management, ranking adjustments, and product evaluation. This is an important component of improving search quality. - A number of Google internal code names have been released for better understanding.
The ruling text has clearly organized the previously unclear Google internal designations and roles of each algorithm, such as Navboost, GLUE, RankEmbed, FastSearch, and MAGIT. This allows for a clearer understanding of the relationship between search and generative AI.
Details of internal code names and their role
I have summarized each code name based on the information in the decision letter, as well as a memorandum.
Internal code names mentioned in the decision
Project Magi
- Summary: An internal project to integrate generative AI into Google Search. Various generative AI ideas were incorporated and ultimately realized as the “AI Overviews” feature.
- Main input: search results and related signals
- Output and behavior: presentation of key points in natural sentences at the top of the SERP and further exploration functions by AI Mode
- Positioning: Fundamental technology that supports new search engine experiences.
As for Magi, it was already publicized in 2023 and was not hidden (or even before that).
MAGIT
- Summary: Post-processing model to optimize Gemini foundation model for AI summarization for search. Tuned using search data.
- Main input: search data and instructional information for summary
- Output and Behavior: Generate natural sentences for AI Overviews
- Special note: Stated that no click or query data was used for pre-training of the underlying model.
Gemini is the successor to PaLM2 and LaMDA, and one of the models based on that Gemini and optimized for search is MAGIT (or maybe Magi-T).
AI Overviews and AI Mode (so-called Google AI search)
- Summary: AI Overviews is a function that presents a summary at the top of search results, and AI Mode is a function that supports further information search. We confirmed that the introduction of AI Mode improved search satisfaction.
- Input and Behavior: We will develop links and leads to additional information that will present the main points using the normal search results and the generative model.
- Additional information: While the introduction of AI Overviews reduces direct link clicks, it also increases the number of clicks on referred links.
GLUE (super query log)
- Summary: A logging system that aggregates query content, device information, and overall search behavior such as clicks, hover, scrolling, and spelling correction, including Navboost data.
- Purpose: Basic data for ranking adjustment and behavior understanding
Navboost (NavBoost)
- Summary: A memory-based ranking system that records and statistic user click and query action history and utilizes it for ranking. Uses data from the most recent 13 months.
- Primary use: bottom-up search result quality, including long-tail, local, and new queries.
*NavBoost has been around for quite some time, including the name. Specifically, it is a fundamental system that has been built into Google search since 2005, and is said to record user behavior such as clicks, browsing, and leaving in response to search queries, and reflect this information in improving rankings for similar searches in the next round.
RankEmbed/RankEmbedBERT
- Abstract: Top ranking signals using deep learning. It learns about 70 days of log data and manually evaluated scores.
- Use: Improves semantic matching and is highly effective for long-tail queries.
FastSearch
- Summary: Dedicated to search grounding, using RankEmbed-derived signals to quickly return short, summarized, ranked web results for Gemini answer generation.
- Use: Providing grounding in answer generation for generative AI.
Vertex AI (Search grounding)
- Overview: Cloud functionality that allows third parties to ground with Google search and other data, returning “information” only, not “results” from FastSearch itself.
- Asymmetry: Gemini receives access to some search functions (e.g., Knowledge Graph) via Vertex, but does not provide them to third parties
GCC/Docjoins
- Abstract: A large web corpus (Google Common Corpus) and its management system (Docjoins) used for pre-training Gemini models, larger than Common Crawl.
Gemini Nano/AICore
- Summary: A small LLM (Gemini Nano) running on a terminal and its execution environment (AICore). It features high-speed inference.
- Applications: On-device processing, privacy support.
Key Points for Utilizing User Behavior Data
User behavior is positioned as an important component of search quality.
- Google retains click/stay/return and other behavioral logs for a long period of time (in principle 13 months for NavBoost), which is reflected in rankings, index operations, and freshness maintenance.
- Satisfaction is estimated by “good clicks,” “bad clicks,” “last longest viewed clicks,” etc., and used for the next ranking adjustment.
- Positioning of Glue for real-time movements (sudden changes in news, etc.) and Slices for contextual segments.
The quality of clicks, last long viewed results, dwell time and return rates, hover and scroll on SERPs, and other micro-behavioral data are used by Google to adjust rankings and determine search quality.
It is also used to improve the accuracy of local searches, especially in mobile environments, for long-tail queries and new topics.
In practice, effective measures include content design to reduce direct returns and short-term abandonment, to increase user turnover and stay, to develop structured data, and to present information in a concise manner that summarizes the main points.
It is also important to maintain a content structure and FAQs that are easily adopted by the AI summary function.
summary
The text of the ruling makes it clear that Google is designing its search engine with user behavior at the center.
In the field of SEO and web operations, this is something we have already known for some time, but now that it has been reiterated, I think it is information that is easy to use in various areas. However, I think it is good that it is easy to understand when a common language is created, even if it is just a matter of knowing the name.
中小企業・小規模事業者の方々に向けて、ウェブの活用やホームページの戦略などについてWebコンサルティング、施策代行実施などを行っている、株式会社ラウンドナップ代表取締役の中山陽平です。中小企業のWeb活用をサポートし、そこからの反響獲得を実現させています。→プロフィール詳細はこちらから
