Why is note's Generative AI traffic so strong? [Part 1] The reality of traffic via Generative AI that exceeds expectations based on search engine traffic volume by 4.2 times.

In this article, we use behavioral logs from VALUES, Inc. to attempt to verify the common theory that the more search traffic a site has, the more traffic via Generative AI is expected. In conclusion, the results confirmed a certain correlation between the two (as discussed later, it is a state similar to the Double Jeopardy effect).

On the other hand, at the same time, exceptional cases that deviated from that correlation were also observed.

A representative example of such an exception is note.com. In this article, based on our proprietary data, we will explore in two parts (Part 1 and Part 2) why note is strong in SEO in the first place, and how it gains traffic via Generative AI (hereinafter referred to as 'Excess AI Traffic') that exceeds the expectations derived from search traffic.

Note: 'Excess' is a word indicating [surplus, overabundance, exceeding, excessive]. In the Double Jeopardy law of marketing, brands that acquire loyalty beyond the behavioral loyalty expected from the brand's scale are expressed as having 'excess loyalty' or similar terms.

Now, let's begin.

Introduction

Verification Theme: The relationship between search traffic and traffic via Generative AI

The more search traffic a site has, the more traffic via Generative AI is expected—I started this analysis because I thought this theme might have a structure similar to the Double Jeopardy law.

In other words, sites with large scale appear frequently in LLM pre-training data, and given the current situation of SERPs favoring domains, they also tend to be hit easily in real-time search (so-called RAG).

'The rich get richer'—I thought it was almost certain that such a situation exists, but just like with Double Jeopardy, the insight lies not in just looking at that simple law, but in asking what are the outliers (such as the excess loyalty mentioned at the beginning), and why are they happening? That is the goal of this verification theme: to clarify that point.

What is traffic via Generative AI?

Traffic via Generative AI (also called GEN AI) refers to traffic generated from URLs that Generative AI tools like ChatGPT, Perplexity, and Gemini reference and present when answering questions or summarizing. Note that this data does not include 'AI Overviews' displayed by Google on search results.

Unlike traditional search traffic (clicks from Google, Yahoo!, and Bing), this traffic via Generative AI involves a process where the AI selects information sources, so the SEO community is buzzing about how much new countermeasures are needed. Recently, terms like 'AI Visibility' have also become common.

Now, let's take a look at the data.

Correlation between search engine traffic vs. traffic via Generative AI (May 2025 – June 2025)

Aggregation period: May 2025 – June 2025
Survey subjects: 692 sites included in 25 categories
Devices: PC & Smartphone
Source data: VALUES, Inc. behavioral log data

Conclusion: Overall, there is a certain correlation. Sites such as note, Wikipedia, and GitHub were observed as exceptions.

In this graph, the horizontal axis represents traffic via search engines during the period, and the vertical axis represents traffic via Generative AI. The URLs aggregated were surveyed across approximately 700 sites included in 25 categories (product comparison, shopping, media, corporate/government/organizations, etc.). The R-squared (coefficient of determination) of the trend line shown in the graph is 0.5551, and the P-value is less than 0.0001.

Takeaways from this article

The first takeaway from this article is that, based on correlation, the expected value of traffic via generative AI can be derived from traffic via search engines, is something that can be done. As a hypothesis for the direction of causality, we currently assume that 'high ranking in search engines → more likely to be included in LLM pre-training data and more likely to be a target for RAG,' which means we can calculate the expected traffic volume for generative AI based on the scale of search engine traffic.

The second takeaway is that, while there is a certain correlation, there are sites that users are actually more likely or less likely to navigate to after being cited by generative AI, which is a fact. In the Japanese internet market, note, the subject of this article, was a prime example of a site gaining excess traffic.

Furthermore, we also found that there is a gap between (1) the rate of citation and (2) whether users actually navigate or take action (this is the third takeaway).

For example, according to an ahrefs study, research data has been published showing that Wikipedia is cited more frequently than search engine traffic in both ChatGPT and Perplexity (this article also serves as a reproduction study of that). However, according to data from Values, Inc., where the author works, the actual volume of traffic to Wikipedia via generative AI is significantly lower than the expected value.

Unlike an approach that mechanically inputs prompts and calculates citation rates, this is a point that became clear precisely because it is behavioral data that can grasp whether users are actually navigating.

Now, as a point of interest, we will move on to our analysis of note.
(I plan to summarize other notable sites besides note separately.)

An astonishing existence that deviates from the correlation—what are the characteristics of note.com?

How note.com deviates from the correlation line

Based on this double jeopardy correlation, the traffic to note.com via generative AI, which would be expected from the scale of its search engine traffic, is actually 4.2 times the expected value. This is a significant outlier. Since its search engine traffic is similar to Instagram's, its generative AI traffic should ideally be at a similar level, so why was this excess AI traffic created?

The strength of note.com's SEO as a premise

1. Domain evaluation from Google
First, what should be noted is the evaluation of the note domain by users and Google. Due to many years of operation and content publication by many users, it has received a high evaluation from Google, even leading to a capital alliance with Google.

Furthermore, the following is the result of a survey of note.com using a Domain Power (DP) survey tool by SEO expert Tsuyoshi Kashiwazaki (Thank you for the useful tool, Kashiwazaki-san!)

Domain power is a numerical value aggregated by various third-party tools, but it is a proxy indicator that shows the reliability and authority of a website, and it is said that the higher the value, the easier it is to rank high in search results.

Note that these domain metrics from third-party tools do not perfectly match the features held by Google. As a feature that indicates a site's authority, Google is known to use a site_quality score. Its existence has been revealed through public documents from the May 2025 hearing records, past leaks, and patent information. I have listed my post that serves as a reference below.

2. Ease of sharing on SNS → Speed of URL discovery
I'm sure you have seen note articles shared on SNS. In addition to the good usability that allows authors/readers to easily share on SNS, content related to people's opinions and experiences is easily shared, creating a cycle where URLs are easily discovered through SNS diffusion and articles reach more users.

3. Speed until indexing
note has the characteristic that new articles are recognized by Google relatively quickly after publication and are easily displayed in search results. The background to this is that the processing on Google's side for indexing is optimized according to the update interval and frequency of the entire site, and for note, where a very large amount of content is updated, the speed until indexing is processed in a shorter time compared to other sites.

In other words, the structure is such that individual users participating in note can enjoy the benefits of indexing speed that large-scale sites receive.

4. User behavior improves due to the site's high usability
Checking user behavior on note based on data from Values, Inc., the trends for sessions per user, page views per user, and average dwell time are all trending positively.

According to research on search marketing by Values, Inc., which I oversee (analysis of Google's algorithm using a machine learning model called Random Forest), it has been quantitatively shown that user behavior is a clear ranking factor. In particular, it is known that [Dwell time on page] and [Sessions per user] always have high scores for importance (feature importance) calculated by Random Forest is a known fact.

Guiding users to relevant content from a vast amount of article content and the difficulty of leaving due to the lack of ad displays are likely what create good user behavior. Based on past leaked information from Google, it is said that the dwell time of each site/page serves as a signal, and I believe this high-quality user behavior leads to the strength of SEO.

I understand it's strong in SEO, but why does it generate power that deviates from the correlation?

Should note.com not simply be viewed as a [blog media] that is strong in SEO?

So far, we have looked at the background of why note is strong in traditional SEO, which I believe many people in the SEO community have already felt firsthand. And the purpose of this article is to approach not only the strength of traditional SEO, but also the point of Why is it able to acquire more traffic via generative AI than expected from SEO traffic?.

Beyond SEO strengths such as domain authority, what is the reason why users actually go to see the source after a conversational AI cites it on the platform?

The keyword is "note.com should not be viewed as a mere [blog media]". The scope of this survey includes other blog platforms that could be similar services (ameblo, fc2 blog, etc.), but excess AI traffic has not been observed in those media. *fc2 is almost on the trend line, and ameblo has only acquired a slight amount of excess traffic.

In the next article, as another set of data, I will organize whether the law of double jeopardy applies similarly to each site category. I would like to think about what implications can be obtained when classifying sites by category and confirming the correlation, and combine those implications with the content of this first part.

In situations where users are exploring information, I believe that this kind of excess traffic (excess AI traffic) occurs precisely in themes where it is questioned who to obtain information from is what happens. Next time, I plan to share some of my own hypotheses. This excess traffic (excess AI traffic) occurs

Note: The next article will not cover topics such as LLMs.txt or structured markup. On note, structured markup is very simple (only BreadcrumbList and BlogPosting in each article), there are no notable points specialized for GEO (LLMO), and LLMs.txt is not installed. I hope this will be an opportunity to think about how to engage with media, rather than just focusing on immediate, detailed measures.

Summary of Part 1

The common belief that 'sites with more search traffic are expected to have more traffic via generative AI' has been quantitatively demonstrated, but it is the exceptional sites that offer insights and hints.

Using behavioral logs from Values, Inc., we were able to verify the common belief that 'sites with more search traffic are expected to have more traffic via generative AI'.

On the other hand, it also became clear from actual user behavior data that there are sites that users are actually more or less likely to navigate to after being cited by generative AI.

While preliminary data that aggregates citation rates and mention rates by AI by inputting large amounts of prompts has begun to exist, we were able to clarify whether users are actually navigating to those sites using behavioral logs from Values, Inc. (To my knowledge, there are no other examples that express the relationship between search engine traffic and generative AI traffic using large-scale data from this perspective!)

As an aside, Values, Inc. and Google have jointly announced the following research results based on such unique behavioral log data.

Traffic to note.com via generative AI has reached 4.2 times its expected value.

'Sites with more search traffic have more traffic via generative AI' Within this simple correlation, it has been revealed that note is acquiring an astonishing amount of excess traffic.

As a premise for this, we also organized that note's SEO strength lies in the structure where individual users can enjoy the benefits of a large-scale site. Specifically, the fact that [the evaluation and benefits of note as a whole] can be [enjoyed by individual users], which includes (1) site authority (site_quality score), which is one of the features of Google's algorithm, (2) speed of URL discovery, (3) speed of indexing, and (4) high-quality user behavior signals across the entire site, provides a strong incentive (ease of ranking higher) for article writers.

Finally

Values, Inc. supports search marketing based on unique data analysis.

We provide support for overall search marketing (SEO and advertising based on search behavior data) based on data, including the unique behavioral logs that serve as the data for this article. In an era where search is no longer just about Google, if your company would like to conduct search marketing adapted to the AI era, please feel free to contact us.

I share information related to marketing on X (https://x.com/Roberto_VALUES). I will also announce the second part of this article there, so please follow me!