LLMOcheki launches weakness analysis, pinpointing which AI you are passed over on and for which kind of question

Itera, Inc. (head office: Shinjuku-ku, Tokyo; Representative Director: Takayuki Muto) began offering, on 6 October 2026, a new feature in LLMOcheki, its tool for measuring and improving LLMO, that strengthens analysis of measurement prompts — the questions put to AI — by category.

Mention rates are shown as a matrix, one cell per combination of question type and AI engine. Cells with enough measurements behind them and no mentions at all are highlighted, so the matrix can be used to decide what to work on first. Analysis of the sites AI cites can now also be filtered by question type.

Background: an overall mention rate does not tell you where you are weak

As people increasingly research products and services through generative AI, more companies are measuring continuously whether their brand comes up in the answers.

But the more questions you measure, the less an overall mention rate shows. Two companies can have the same overall figure while one of them never comes up on a particular AI, and only when the question is a comparison. An average conceals a weakness.

Feature 1: a mention-rate matrix of question type by AI engine

LLMOcheki has long classified measurement prompts automatically into six question types and shown mention rate, share and tone for each. The six are awareness, competitor comparison, verification, accuracy, reputation and market perception.

We have now added a matrix showing the mention rate for each combination of question type and AI engine. It identifies the weak combination — that you do not come up on ChatGPT when the question is a comparison, for instance.

Cells with enough measurements behind them and yet no mentions at all are highlighted in red. Cells with too few measurements are held back from judgement, so a zero that happened by chance on a small sample can be told apart from consistently not being chosen. The matrix is usable as it stands for deciding what to work on first.

Feature 2: cited-source analysis by question type

Analysis of the sites AI referred to as the basis of its answers can now be narrowed by question type.

Extract the sites AI refers to on comparison questions and you have the media worth approaching about a listing. The sites referred to on verification and accuracy questions point to the pages on your own site that need strengthening. Identifying the weakness and identifying what to do about it now happen on the same screen.

Alongside the existing category analysis

In addition to this feature, we offer coverage analysis by CEP — category entry point, meaning the situations and purposes that bring a category to a customer’s mind — designed according to the industry, and mention-rate analysis by stage of purchase consideration.

When a weak category is found, you can go from a suggested representative question to registering it for measurement in one click. Finding the category, identifying the weakness and starting the work are supported as one sequence.

Availability

Available on every plan at no additional cost.

Comment from our Representative Director

“Being told your mention rate is 60% does not tell you what to do next. What you want to know is which AI you are falling down on, and with what kind of question. Only once it is broken down that far does the work become something concrete.

We hold back cells with too few measurements so that a zero which happened by chance is not mistaken for a weakness.”

About LLMOcheki

LLMOcheki is a Japanese-built tool for measuring and improving LLMO. It measures mentions and citations of a brand daily across eight AI search engines — ChatGPT, Gemini, Google AI Overview, AI Mode, Claude, Grok, Perplexity and Copilot — and carries through to the work of improvement. It puts the same question repeatedly and records the result together with its statistical spread.

A seven-day free trial and a free audit are available.

LLMOcheki https://llmocheki.com/