How an AI assistant learns about your company
Before an AI answer shows your company’s name with a link, six links of a chain must hold. The left half happens days to weeks before the question; the right half in the seconds after it.
Point at (or tap) a number in the picture to read that step.
- FoundOfficial documentation
A crawler first has to learn the page exists. The clues are links from other sites, your sitemap, and registering with webmaster tools such as Search Console. Google describes all three officially, and does not promise to crawl what you submit. - ReadOfficial documentation
Each AI company’s crawler comes to read the pages robots.txt allows. OpenAI and Anthropic say that refusing their search crawlers keeps your site out of their answers, or makes it less visible. - IndexedOfficial documentation and outside reports
Being read does not mean being indexed. Each AI company has its own index, and being in one does not put you in the others. Which AI uses which search index is partly unconfirmed (for example, how ChatGPT relates to Bing). - SearchedHypothesis
The assistant is thought to look for candidates in its own index rather than across the whole web. - Chosen and trustedOutside reports and our observation
From the candidates, the AI picks sources: sometimes your own site, sometimes what others wrote about you. Finding something is not the same as believing it correctly. - Answered with a linkHypothesis
If the answer carries a link to your site, that answer is a chance to win a visit.
The dotted line is a side path: while answering, an assistant sometimes opens your page directly because the user asked (ChatGPT-User, Claude-User and others). It does not go through the index, and being opened does not mean being indexed (hypothesis).
What this explanation does not show
- "Hypothesis" means it does not contradict what we see; it is not a confirmed fact.
- We have not yet measured Google’s AI Overviews or AI Mode.
Where does your site break in this picture?
Parallel World measures where your site breaks in these six links (3 and 4 need access such as Search Console). Ask us through the contact form.
Information about the whole site
The content of the other pages is collected here. Each heading links to the original page.
Home
Original page: https://pw.migurei.ai/en/
Parallel World — Your website, as AI agents see it
A website built for search engines may "not exist" for AI agents.
LLMO, AIO and GEO (AI search optimization) aim for a site that AI reads correctly. Parallel World measures and monitors whether your site is in that state. It is a measurement service, not an optimization agency or consultancy.
The page people see and the page AI sees are not the same.
From training crawlers to shopping agents, Parallel World measures how your site looks to AI across 5 stages, and monitors how it changes.
Parallel World is available now: it measures and monitors how a website looks to AI agents.
5 stages
| Stage | Who touches your site | What they look at |
|---|---|---|
| A. Training | GPTBot, ClaudeBot, Google-Extended, etc. | robots.txt, Content-Signal |
| B. Search index | Googlebot, Bingbot, OAI-SearchBot, etc. | Sitemap, structured data |
| C. Fetch at answer time | ChatGPT-User, Claude-User, etc. | Whether key information is in the raw HTML |
| D. Browsing and actions | Browser agents | Accessibility tree, forms |
| E. Transactions | Shopping agents | Whether they stop before payment |
Share of sites that reach AI correctly at each stage (expected)
We expect fewer sites to reach AI correctly at each later stage.
| Stage | Expected |
|---|---|
| A. Training | High |
| B. Search index | Fairly high |
| C. Fetch at answer time | Medium |
| D. Browsing and actions | Low |
| E. Transactions | Very low |
These are expectations before testing, not measurements. Measured values will be published after testing.
Plans and free trial
Please contact us for pricing.
| Plan | Price | Sites | Scan frequency | Includes |
|---|---|---|---|---|
| Starter | Contact us | 1 | Weekly | 5-stage report, conflict detection, Japanese PDF |
| Business | Contact us | 5 | Weekly | Everything in Starter, anonymous industry comparison (planned), change alerts (planned) |
| Enterprise | Contact us | Unlimited | By arrangement | Everything in Business, import of your own logs (planned), dedicated support (planned) |
Free trial: Planned. We deliver the results of one scan of one site. Contact us to request it.
How an AI learns about a company and answers
What happens before your company shows up when someone asks an AI to compare suppliers. Six links of the chain in one picture.See the picture
What Parallel World does not do
- We do not create our own scoring criteria. We gather the public criteria of Cloudflare, Google and each AI provider, and show where they conflict and when they change.
- Scans are read-only. They never submit forms, add to cart or log in.
About
Original page: https://pw.migurei.ai/en/about
What Parallel World does
We check how AI agents read your site, against both the published criteria and how real AI agents actually behave.
Parallel World is available now: it measures and monitors how a website looks to AI agents.
LLMO (also called AIO, GEO or AI search optimization) is work aimed at getting AI answers to use the correct information from your site. Parallel World measures and monitors how AI reads your site, using published criteria and each provider's official tools. It does not do that work for you, and it does not offer consulting.
What we do
- Gather the criteriaWe collect the criteria published by Cloudflare, Google, OpenAI, Anthropic and public specifications. We do not create our own.
- Measure, read-onlyFor each of the 5 stages, we record the results of each provider's official tools and APIs as they are.
- Find conflicts and changesWe find where criteria contradict each other, and when they change.
- Check against real behaviorWe run real AI agents on our own sites to see which criteria actually matter.
Who it helps
- E-commerce and booking site operatorsWho want to know whether prices, stock, shipping and return terms reach AI correctly.
- Web and marketing teamsWho are unsure which criteria to follow.
- Business ownersWho want to be ready for customers and transactions that come through AI.
Value
- No criteria of our ownOnly the results of official tools and published criteria.
- Conflicts made visibleWhere criteria contradict each other, with their sources.
- Changes you can seeWhen a criterion changes, past results are kept as results under the old criterion, never rewritten.
- Checked against real behaviorExperiments show whether a criterion really changes what AI does.
- SafeRead-only. We never submit forms, add to cart or log in.
What you get
- A results report for each of the 5 stages, as a Japanese PDFEach item names the source and version of its criterion.
- A list of what to fixFor each conflict, a hint based on the source's own documents.
- Notice of criteria changes (planned)We tell you when a criterion changes, without rewriting your results.
The aim: when someone asks an AI, the answer uses the correct information from your site.
Pricing
Original page: https://pw.migurei.ai/en/pricing
Plans
Please contact us for pricing.
| Plan | Price | Sites | Scan frequency | Includes |
|---|---|---|---|---|
| Starter | Contact us | 1 | Weekly | 5-stage report, conflict detection, Japanese PDF |
| Business | Contact us | 5 | Weekly | Everything in Starter, anonymous industry comparison (planned), change alerts (planned) |
| Enterprise | Contact us | Unlimited | By arrangement | Everything in Business, import of your own logs (planned), dedicated support (planned) |
Questions
Is there a free trial? Planned. We deliver the results of one scan of one site. Contact us to request it.
Does a scan put load on my site? At most 50 requests per scan that the scanner sends itself (images and scripts the page loads while rendering are counted under a separate cap), and at least 2 to 3 seconds between requests to the same domain. Its User-Agent is migurei-ParallelWeb and it obeys robots.txt.
Docs
Original page: https://pw.migurei.ai/en/docs
Docs
The scanner is running. Below is its specification.
About the scanner
The Parallel World scanner identifies itself with this User-Agent.
migurei-ParallelWeb/0.1 (+https://migurei.ai/ja/scanner; research@migurei.ai)Use the robots.txt token migurei-ParallelWeb to allow or disallow it. If it is disallowed, the scan is recorded as "blocked" and no content is fetched.
At most 50 requests per scan that the scanner sends itself (images and scripts the page loads while rendering are counted under a separate cap), and at least 2 to 3 seconds between requests to the same domain.
In addition, it requests the top page once with each published AI fetcher User-Agent (6 to 8 of them) and records only the status code, the challenge type and the response time. It fetches no content, does not retry and does not change IP address.
What we measure
- robots.txt Whether it allows each major AI bot (about 15)
- llms.txt Whether it exists, and its format
- Sitemap Consistency (noindex, redirects, 404)
- Structured data Per page type (Product, Offer, Organization, etc.)
- Raw HTML Share of key information (price, stock, shipping, returns) present in it
- Fetcher User-Agents Responses to the published ones (403, 429, challenges)
- Accessibility Tree quality, autocomplete attributes on forms
Reports
Each item links to the source and version of its criterion and to the snapshot taken at measurement. When a criterion changes, past results are shown as "measured under the old criterion" and are never rewritten.
Information for AI agents
- llms.txt: an overview of this site and its pages (in Japanese)
- Markdown Request any page with
Accept: text/markdownto get it as Markdown. - MCP MCP server (read-only, in Japanese):
https://pw.migurei.ai/mcp. It returns only what this site states.
Company
Original page: https://pw.migurei.ai/en/company
Company
Parallel World is a service of migurei.ai. migurei.ai is a subsidiary of 株式会社エイチビーラボジャパン.
Parent company
| Name | 株式会社エイチビーラボジャパン |
|---|---|
| Address | 〒105-0012 東京都港区芝大門1丁目3番4号 グランファースト芝大門6階 |
| Founded | November 28, 2017 |
| Representative | グエン フィ タン |
| Capital | JPY 30 million |
| Phone | (+81)-3-6281-9068 |
| Certification | PrivacyMark (obtained August 2024) |
Source: https://hblab.co.jp/about-hblab/ (as of October 1, 2026). Name and address as published, in Japanese.
For questions about the service, please use the contact form.
Contact
Original page: https://pw.migurei.ai/en/contact
Contact
Free trial, pricing and getting started: write to us here.
Contact form fields:
- Company (required)
- Name (required)
- Email (required)
- URL of the site
- Topic：Free trial / Pricing / Other
- Questions or requests