LLMOcheki launches the Improvement Auto-Loop, joining detection of a drop in AI search visibility through to verification of the fix

Itera, Inc. (head office: Shinjuku-ku, Tokyo; Representative Director: Takayuki Muto) has begun offering four new features in LLMOcheki, its tool for measuring and improving LLMO, led by the Improvement Auto-Loop.

Detecting a statistically significant drop, proposing what to do about it, drafting the article and measuring the effect after publication are joined into a single flow. Alongside it come a place to gather brand information, a content inventory, and WordPress support for measuring AI bot access.

Background: measuring alone does not produce results

In our own survey, 65.0% of those who used AI when choosing a B2B product put the AI’s comparison into internal approval or proposal documents (announced 25 September 2026). About half of AI users have, on the strength of an AI answer, either abandoned a purchase or switched to a different brand (announced 24 September 2026).

AI answers are already being carried into the room where decisions are made. What companies need is not measurement in itself, but the practice of noticing a drop, fixing it, and keeping a check on whether the fix worked.

That practice tends to rest on manual work. Without someone watching the numbers, someone working out the cause, someone rewriting the article and someone confirming the effect, you end up with a measurement tool in place and no improvement going round.

Feature 1: the Improvement Auto-Loop

When your exposure or citation rate in AI search falls by a statistically significant margin, the system detects it automatically. It then generates proposals grounded in the measurement data and asks the person responsible to approve them.

For an approved proposal, AI drafts the rewritten article. Once the article is published, it is recorded automatically in the activity log.

Two weeks after the work is carried out, an effect measurement runs automatically, comparing share of mentions and AI citation counts before and after, and the result is recorded and notified. Lines of work shown to be effective are given priority in later proposals.

Signals from conventional SEO are folded into the proposals as well, including falls in search rank and cannibalisation, where your own pages compete with each other on the same term.

Adopting a proposal and publishing an article both pass through the approval of the person responsible. Generated content is never published as it stands.

Feature 2: Brand Profile

The brand information AI refers to when generating an article is gathered on one screen: your official definition, your strengths, reader personas, and primary sources with attribution.

Feed in material such as a URL, a PDF or meeting notes, and AI extracts candidate facts automatically. Only the facts the person responsible has approved are used in generating articles.

Feature 3: content inventory

Every page on the site, up to 500 URLs, is matched against search performance, AI crawler access, search rank and AI citation data, and automatically sorted into four categories — keep, rewrite, merge, candidate for deletion — with a priority from one to five.

The criteria are fixed, so the same data always produces the same result. It exports to CSV together with the figures behind each judgement, ready to use in an agency’s or a production company’s proposal documents.

Feature 4: AI bot access measurement now supports WordPress

The feature that automatically measures which of your pages AI bots such as GPTBot come to read is now available by placing code on a WordPress site, in addition to the existing Cloudflare integration. No dedicated server configuration or manual log upload is required.

Comment from our Representative Director

“We hear a great deal that improvement does not go round even once a measurement tool is in place. Someone to notice the numbers falling, someone to work out why, someone to fix it, someone to confirm it worked — hardly any company has all four of those roles. We designed this so the tool takes on three of them and only the judgement is left to people.”

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, Claude, Perplexity, Grok, Google AI Mode, AI Overview and Copilot — and covers everything from carrying out improvements to verifying their effect in a single tool.

Its distinguishing features are a measurement method that puts the same question repeatedly and records the result together with its statistical spread, and a design that lets a team run measurement, article creation, publication and verification in house. It was in use at 180 brands within two months of launch (as of early September 2026).

LLMOcheki https://llmocheki.com/