From a real entity to an answer an AI can cite.
This page explains how worldwiki pages are built, what each section means, and how the AI Visibility measurement works.
The knowledge graph
worldwiki is a graph, not a list of articles. Entities are the nodes; relations (provides, part of, competes with, treats…) are the edges. Topics group entities by field, questions attach to the entity they are about, and articles expand a topic in depth. Every node has one canonical URL and a machine-readable copy in the API.
Anatomy of an entity page
Definition
One or two sentences that say what the entity is, written to stand alone when quoted.
Quick answer
The most-asked question about the entity with a short, sourced answer. This is what AI assistants most often lift.
Key facts
Predicate–value pairs (founded, headquarters, approvals, prices…) each linked to a numbered reference. Values without a source are marked pending.
Evidence & claims
Statements made by the provider are kept separate from independently verified facts and carry an evidence level.
References
Numbered sources with type, publisher and access date. The independent-source ratio is shown for transparency.
How AI search engines use it
ChatGPT, Perplexity, Gemini and Claude answer with web search on by fetching pages, extracting facts and citing the URL. They prefer pages that answer directly, expose structured data, are stable and dated, and are allowed in robots.txt. worldwiki pages are built for exactly that: server-rendered HTML, JSON-LD, canonical URLs, sitemaps with real modification dates, llms.txt and an open API.
We track which AI crawlers fetch which pages (GPTBot, ClaudeBot, PerplexityBot, Google-Extended and others) and publish a crawl-readiness score for the site.
How AI Visibility measurement works
Prompts
For a business customer we agree on a fixed set of questions their customers ask AI assistants (for example "best clinic for X in Seoul").
Engines
Each prompt is sent to the public, search-enabled product of each engine on a fixed cadence: ChatGPT, Perplexity, Gemini, Claude; Grok on the top plan. Google AI Overviews and Copilot have no prompt API and are reported as not measured.
Record
We store the full answer, whether the organization or an alias is named, and every cited URL. Nothing is inferred or extrapolated.
Aggregate
Mention rate, citation rate and share of voice per engine and per prompt, compared with the previous period.
Act
Gaps become sourced knowledge pages and answers on worldwiki. Daily, weekly and monthly reports show crawler visits, citations gained or lost and next actions.
What we never do
- Change facts, sources or rankings because someone pays.
- Fabricate sources or structured data that describes content not on the page.
- Publish medical, legal or financial content without expert review.
- Measure with anything other than the engine's real, search-enabled product.
- Advertise language versions that do not exist.