AI Overview now appears on 92.1% of Japanese searches, and on 100% in seven sectors. Itera publishes its September 2026 LLMO/AIO trend report, free of charge

Itera, Inc. (head office: Shinjuku-ku, Tokyo; Representative Director: Takayuki Muto), which operates LLMOcheki, the SaaS product for measuring citation in AI search, published its monthly research report — the September 2026 LLMO/AIO trend report, eighteen pages in all — free of charge on 5 October 2026.

The report draws together the main developments in AI search and generative search optimisation (LLMO/AIO/GEO) through September 2026, from our own continuous measurement and from public sources in Japan and abroad.

In Japanese Google search, which we measure continuously across thirty sectors and nine hundred queries, the average appearance rate of AI Overview reached 92.1% as of 28 September 2026. That is up from 86.5% at the time of the August survey, and in seven sectors — e-commerce, real estate and home renovation among them — the rate was 100.0%, meaning an AI summary appeared on every query measured. Factoring that in, the estimated click-through rate for the top organic result is 12.7%, less than half the 28.4% commonly used as a benchmark.

September was also the month in which a Google trial programme paying for content that contributes to AI answers came to light, turning citation in AI search from a measure of performance into a source of revenue.

The report is published free as a PDF and can be downloaded by anyone from our site.

Why we publish it

AI Overview has become a standard feature of Japanese search, and the shift towards zero-click — users deciding on the strength of the AI’s answer alone — is advancing. In September 2026, joint research from the University of Pennsylvania and Northeastern University reported that 93% of searches through AI Mode and 83% of searches showing AI Overviews end without a click. Figures that had been industry estimates now have academic backing.

Meanwhile, the means by which a company can see how it appears in AI search are still being built. In September 2026 Google’s John Mueller commented that, for measuring impressions through generative AI, the conventional notion of ranks one to ten is hard to map onto — the party providing search acknowledging the limits of the established framework.

Given that the surfaces are being built out rapidly while the means of measuring them and the yardsticks for judging them lag behind, we set out the state of the market each month in a form practitioners can act on.

Highlights

1. AI Overview appears on 92.1% of Japanese searches, and on 100.0% in seven sectors

In our continuous measurement across thirty sectors and nine hundred queries (survey period 10 August to 28 September 2026), the average AI Overview appearance rate as of 28 September 2026 was 92.1%, up from 86.5% at the time of the August survey.

The seven sectors reaching 100.0% were e-commerce, M&A, web production, programming schools, home renovation, real estate and removals. In these, an AI summary appeared on every query measured.

Across the eight measurements from 10 August to 28 September, the thirty-sector average moved within a range of 84.5% to 93.4%, with no downward trend.

2. A 27-point spread between sectors

While the highest sectors stand at 100.0%, the lowest — hair salons at 73.3% and dental clinics at 76.7% — sit well below. Twenty-seven points separate top from bottom, so how urgent this work is differs considerably by sector.

3. YouTube is the most cited source in AI Overview, by a factor of 7.7

The domain cited most often within AI Overview answers during the survey period was youtube.com, with 4,745 citations — about 7.7 times the 616 of second-placed google.com. Below them come note.com, moneyforward.com, suumo.jp and other operator-run media cited repeatedly within particular sectors.

A separate consumer survey of ours found that only 4.00% of people turn to social media or YouTube to verify what an AI tells them. AI consults YouTube to compose its answer; consumers check that answer against official sites and e-commerce reviews. The structure is asymmetric.

4. Google has begun paying for citation in AI answers

In September 2026 it emerged that Google is running an invitation-only trial through Search Console called the AI Contribution Pilot, which pays for content that contributes to answers generated in the Gemini app, AI Overviews and AI Mode.

Around a hundred companies are said to be taking part, and the formula is not published. Most smaller sites are reported to have received less than 0.1% of their own advertising revenue, so it is not yet a level to rely on as income.

What matters is that a price has been put on contribution without a click. The yardstick that says no traffic means no value has now been rejected by the platform itself.

5. But half of those citations vanish within a year

September also brought follow-up research on how long citations last. Green Flag Digital found that of 215 finance pages cited by ChatGPT in October 2025, 50% had no citations at all by September 2026, leaving a retention rate of 14%.

Winning a citation is not the end of it. Whether you are holding on to it has to be checked continuously.

6. Search rank is no longer a proxy for AI citation

Research from Brandlight, which specialises in GEO, found the overlap between high Google rankings and AI citation falling from over 70% to under 20%. Kingsworth Research, analysing 5.35 million AI citations, likewise concluded that rank does not serve as a proxy for share of AI citation.

A Zyppy study with Search Engine Land, aggregating 131 SEO specialists, 103 factors and 13,665 data points, found AI-specific techniques such as llms.txt and chunking for AI to be of almost no effect. What mattered was relevance to search intent, content quality, and trust in and awareness of the brand.

What the report contains (eighteen pages)

• A timeline of September 2026, week by week, covering eighteen topics
• The AI Contribution Pilot, and how the nature of citation is changing
• The volatility of citation (14% retention) and the movement observed on the citing side in September
• Three studies on the divergence between search rank and AI citation
• The quickening pace of model changes (Gemini 3.8 Flash and 3.8 Live) and the fault that removed citation links
• Academic backing for zero-click
• The expanding surface of AI Overview (US appearance rate 25.8% to 39.4%; brand searches 26% to over 80%)
• Our measurements: appearance rate, zero-click rate and CTR decline for all thirty sectors, with eight weeks of trend
• Our measurements: the top ten cited domains in AI Overview
• Our own survey: how generative AI is used when choosing B2B products
• Progress and limits in the measurement environment, and changes to crawler control
• Market developments: capital concentration abroad, new entrants in Japan
• A five-principle checklist for practitioners

Read the report https://llmocheki.com/blog/news/2609llmotrendreport

Comment from our Representative Director

“September was the month a price was put on citation. Google began paying for it, and in Japan newspaper publishers began selling it as an advertising product. That citation carries business value is now a given.

In the same month, though, follow-up research put the one-year retention rate at 14%. Citation is an asset, but one that loses half its value if left alone.

I have worked in search and measurement for twelve years, through SEO, MEO and LLMO, and I have not seen the yardsticks change this fast before. Google itself acknowledges the limits of the ranking framework.

Did you win it, are you holding it, and how do you compare with your competitors — whether a company can answer those three questions with its own numbers is, I believe, what will separate them from next year onwards. I hope this report helps in putting that in place.”

About the report

• Title: September 2026 LLMO/AIO trend report
• Published by: Itera, Inc. (LLMOcheki)
• Publication date: 5 October 2026
• Format: PDF, eighteen pages
• Price: free
• How to obtain: download from our site

About LLMOcheki

LLMOcheki is a SaaS product that quantifies how much a brand is mentioned and cited within AI answers across eight engines: ChatGPT, Gemini, Claude, Perplexity, Grok, Copilot, Google AI Overview and Google AI Mode.

Eight engines together: ChatGPT’s share of AI prompts has fallen from 70% to 50% as Gemini and Claude grow. Measuring one engine cannot capture how the market as a whole is moving.

n-shot statistical measurement: because an AI’s answer varies from attempt to attempt even for the same question, appearance rates are calculated from a sufficient number of attempts, with a Wilson 95% confidence interval.

Mention and citation kept apart: whether a name came up and whether it was presented as a source are recorded as separate measures, so the retention of citation can be tracked month on month.

Daily API measurement with weekly real-UI measurement: two layers, using fixed prompts, regular repetition and a Japanese domestic environment, detect model changes and faults as breaks in the data, separating external causes from your own work.

LLMOcheki https://llmocheki.com/

Free LLMO audit https://llmocheki.com/llmo