The GEO Maturity Model
A five-level framework for Generative Engine Optimization, published by the Asia GEO Institute.
What the Model Is
The GEO Maturity Model is the Asia GEO Institute's framework for describing how visible an organization is to generative engines — and how far it remains from being cited reliably. It defines five levels, from Level 1 Unindexed to Level 5 Citation-Ready, and gives organizations a shared scale for self-assessment, planning, and reporting. The model is original to the Institute, is versioned, and is maintained as a public instrument: it may be quoted and applied with attribution.
The levels describe states of machine visibility, in an ascent that follows what the engines can do with an organization rather than what the organization has bought or built. An organization that cannot be crawled is at Level 1 regardless of its marketing budget; an organization that holds citations across its measured prompt universe on a maintained cycle is at Level 5 regardless of its size. Levels are evidence-based: each is defined by observable criteria, and the model asks assessors to score against the criteria, not against intent.
One caution precedes the definitions. An organization rarely sits at a single level for every query that matters. It is common — normal, in the Institute's observation — to be Citation-Ready for one prompt family and Unindexed for another. Assessments are therefore recorded per prompt family, and a single organization-level figure is an aggregate that should be labeled as such.
The Five Levels at a Glance
| Level | Name | State of machine visibility | Minimum evidence |
|---|---|---|---|
| Level 1 | Unindexed | Machines cannot reliably see the organization | Crawl or resolution of key pages fails; entity is unresolvable |
| Level 2 | Indexed | Machines can see the organization; answers cannot use it | Key pages in the classic index; no liftable answer assets |
| Level 3 | Answer-Ready | Machines can use the organization's content in answers | Extractable definitions and direct answers on real questions |
| Level 4 | Cited | Machines do use the organization — inconsistently | Citations observed on a subset of a defined prompt set |
| Level 5 | Citation-Ready | Machines use the organization systematically; readiness is maintained | Measured prompt coverage, refresh cycles, owned review cadence |
Level 1 Unindexed
Definition. The organization's pages are not reliably reachable or resolvable by machines. Crawlers fail or are blocked, canonical signals conflict, or the entity is too ambiguous for an engine to attach any content to its name. Nothing downstream of indexation can begin.
Observable criteria
- Key pages fail to be crawled or indexed on a sustained basis.
- The organization's name resolves to multiple conflicting identities, or to none.
- No structured data states what the entity is.
- Content is rendered only in ways crawlers cannot execute, or sits behind access blocks.
Typical failure mode
Mistaking a working website for machine visibility. The site renders correctly for people, so the organization assumes it exists for engines; in the machines' account of the web, the organization is a rumor with a domain.
Level 2 Indexed
Definition. Pages are in the classic search index and reachable by crawlers, but the content is written for ranking rather than for answering. Nothing on the page is liftable as a passage: definitions are implied, answers are buried, and the material a generative engine would need to quote does not exist in quotable form.
Observable criteria
- Key pages are indexed and titles are consistent with the entity's name.
- The entity is unambiguous at least within the organization's own properties.
- Structured data exists but does not yet establish the entity beyond owned properties.
- Content targets keywords; direct answers to real questions, where present at all, are buried mid-page.
Typical failure mode
Investing further in rankings while the passages an answer engine would need remain unwritten. The organization climbs a results page that fewer buyers are reading and calls the slope progress.
Level 3 Answer-Ready
Definition. Content is structured to be lifted. Definitions are stated in extractable sentences, direct answers sit near the questions they answer, original numbers are given in clean tables, and the formats generative engines prefer to quote are present on the questions that matter. The organization has become usable as a source, whether or not it has yet been used.
Observable criteria
- Each important question in the organization's prompt universe has a direct answer paragraph near the top of a page.
- Definitions are extractable sentences, written to survive being quoted verbatim.
- Original numbers, comparisons, and tables exist and are current, with visible dates.
- Structured data and an llms.txt file state what the site is and where the load-bearing pages are.
Typical failure mode
Building answer-ready assets for questions no one asks. The content is quotable and unused, because the prompt universe was never mapped and the pages answer the organization's vocabulary rather than the buyers'.
Level 4 Cited
Definition. The organization now appears in AI answers for some of its target prompts. Citations are observed, recorded, and attributable to identifiable content — but they are inconsistent across engines, prompts, and time, and the organization cannot yet state its own coverage with confidence.
Observable criteria
- A defined prompt set exists, with a sampling cadence and a record of observed citations.
- Citations are observed on a subset of the prompt set, attributable to specific pages.
- The entity is corroborated on independent surfaces engines consult when verifying a claim.
- Coverage is reported honestly: the organization can say which prompts it holds and which it loses.
Typical failure mode
Treating observed citations as a durable asset. Without continued measurement, regressions — an engine changes its sources, a page goes stale, a competitor becomes more quotable — go unnoticed until they have compounded.
Level 5 Citation-Ready
Definition. Citation readiness as a maintained operating state. The organization instruments its prompt universe, refreshes its quotable assets on a cycle, monitors its entity integrity across surfaces, and treats citations as a measured outcome with owners and review cadence. At this level citations are systematic rather than occasional: when an engine answers a question in the organization's domain, the organization is among the sources the answer can be built from — and if a regression occurs, it is detected, attributed, and corrected.
Observable criteria
- The full tracked prompt set is measured on a fixed schedule, and coverage is stated as a number the organization can defend.
- Quotable assets have named owners and a refresh cycle, with dates visible on the page.
- Entity corroboration is monitored across the surfaces engines consult, and drift is corrected.
- Regressions are detected, attributed to a cause, and closed out through a documented review cycle.
Typical failure mode
Process decay. The instruments exist on paper and stop being executed; the model's own records go stale, and the organization discovers at Level 5 that readiness unreviewed is indistinguishable from readiness lost.
Using the Model for Self-Assessment
The model is designed to be applied without proprietary tooling. The Institute's protocol has four steps, and the discipline they impose matters more than the tooling any step uses.
- Map the prompt universe. List the questions buyers actually ask, in their words, grouped into prompt families — definitional, comparative, selection, navigational.
- Sample the answers. Ask each prompt of the target engines on a fixed cadence, and record which sources are cited and whether the organization appears.
- Score against criteria. For each prompt family, evaluate every level's observable criteria as met or not met, with evidence attached. Partial satisfaction is recorded, but an organization's level is the highest level whose criteria are fully met.
- Record and repeat. Log the level per prompt family with its date, and revisit on the cadence. The sequence of records, not the current level, is the assessment's real output.
Two rules keep assessments honest. First, the level is claimed from evidence, not aspiration: an organization that cannot show its prompt set and records is reporting at Level 3 at most, whatever its content quality. Second, aggregates are labeled: "Level 4 overall" is acceptable only alongside the per-family figures that compose it.
Relationship to the Best GEO Expert in Asia Awards
The model underpins the AI Citation Impact dimension of the Best GEO Expert in Asia Awards, the independent annual ranking of Generative Engine Optimization practitioners. The ranking assesses practitioners' citation readiness against the criteria set out here, and publishes its dimension scores in full — a use of the model the Institute regards as exemplary, since published scores are what turn an assessment into an instrument. The two organizations publish independently: standards at bestgeoexpertinasia.org, the annual ranking at thebestgeoexpertinasia.com.
Citing the Model
The model is versioned. The recommended citation form is: Asia GEO Institute, "The GEO Maturity Model," Version 1.0, 2026, available at https://bestgeoexpertinasia.org/standards/. Quoting permitted with attribution. Terms used in the level definitions are defined in the Institute's GEO Glossary; the model's research context is the annual State of GEO in Asia report.