Retrieval
A source may enter the system's context. From the outside, it is recorded only when evidence is observable; often it is not.
GEO methodology
Majotek GEO methodology: seven layers, from ACCESS to MEASUREMENT
The framework Majotek uses to audit and implement GEO: what must hold for an AI search engine to reach a company, understand it, cite it and, where relevant, recommend it, and how that is measured.
Direct answer
How does Majotek approach GEO?
With a seven-layer framework: ACCESS, ENTITY, OWNED SEMANTIC CAPTURE, EXTERNAL CORROBORATION, RETRIEVAL & CITATION, RECOMMENDATION and MEASUREMENT. Each layer has a question, signals that are reviewed and a way to measure it.
It is our own working framework, not an industry standard, and it promises no specific outcome in any engine.
The framework
The seven layers of the Majotek framework
The layers order the diagnosis and its technical dependencies; they do not describe a single causal chain. After checking access, entity, first-party assets and external corroboration, we measure separately what each interface makes observable.
ACCESS
Can engines reach the pages and read them?
- Crawling and indexing
- HTML retrievable without JavaScript
- Canonicals and redirects
- robots.txt and search bots
- XML sitemap
ENTITY
Is it clear who the company is, what it does and for whom?
- Name and category
- Services and products
- People and authorship
- Verifiable recognitions
- Consistency between site and profiles
OWNED SEMANTIC CAPTURE
Does the own site cover the buyer's questions with citable pages?
- Commercial landings
- Resources and definitions
- Cases
- Comparisons
- Internal linking without cannibalisation
EXTERNAL CORROBORATION
Do third-party sources confirm what your own site says?
- Media and press
- Directories and profiles
- Third-party comparisons (Publisher Capture)
- Product references
- Independent mentions
RETRIEVAL & CITATION
Which sources enter the engine's context and which end up cited?
- Retrieval: sources fetched
- Citation: URLs shown as source
- The difference between them
- Competitors' sources
RECOMMENDATION
Is the brand named, recommended and in what relative position?
- Mention
- Recommendation
- Relative position
- Primary pick where relevant
MEASUREMENT
How do you know whether something changed?
- Frozen prompt sets
- Multiple runs when needed
- Engines kept separate
- Longitudinal analysis
Observable outcomes
Retrieval, citation and mention are separate outputs
We record independent signals from the answer and the sources exposed by the interface. One does not automatically prove another.
Citation
The interface presents a URL as a source. A citation may appear without naming the brand in the answer text.
Mention
The answer text names the brand. This can happen without a link or recommendation.
Recommendation and primary pick are coded as answer attributes when the context expresses them; they are not inferred from a citation or mention.
Definitions
Retrieval, citation, mention, recommendation and primary pick are not the same
| Concept | Working definition | How it is observed | Limit |
|---|---|---|---|
| Retrieval | A source enters the context the engine uses to answer. | Only indirectly or when the interface shows it. | Not always observable from outside. |
| Citation | A URL appears as a linked source in the answer. | Reading the sources the interface shows. | Depends on the engine showing sources. |
| Mention | The brand is named in the answer text. | Reading the text of each answer. | There can be a mention without a link, and a link without a mention. |
| Recommendation | The brand is presented as an option to consider or choose. | Reading the context of the mention. | Needs an explicit coding criterion. |
| Primary pick | The brand is presented as the first or most recommended option. | Position and emphasis within the answer. | Only meaningful when the answer ranks options. |
How it is applied
From the framework to the audit and to implementation
AuditImplementation Vocabulary
Diagnosis by layer
Every audit finding is assigned to a layer (A–G), so the roadmap tackles causes, not symptoms.
Priority by dependency
What blocks later layers is fixed first: access, entity and architecture before external authority.
GEO glossary
Short definitions of the terms the framework uses.
How to measure without fooling yourself
- Frozen prompt sets: the prompt set is documented and does not change between measurements meant to be compared.
- Multiple runs when needed: generative answers vary, so a single attempt is not evidence.
- Engines kept separate: each engine behaves differently and they are analysed without mixing.
- Longitudinal analysis: the aim is a trend over time, not a fixed position.
- No invented data: if something has not been measured, we say so.
What this methodology does not promise
It does not promise that a specific engine will cite, mention or recommend a company, nor that it will do so consistently. Allowing a bot or publishing schema does not guarantee citations.
Nor does it replace SEO: the ACCESS layer and much of OWNED SEMANTIC CAPTURE are technical and editorial SEO. GEO extends that base towards understanding and citability in generative answers.
FAQ
Frequently asked questions about the methodology
Is this framework an industry standard?
No. It is Majotek's own working framework to organise a GEO audit and implementation. It uses common industry concepts, but grouping them into seven layers is ours.
Why separate retrieval, citation, mention and recommendation?
Because they are different outcomes that can occur independently: a page can be retrieved without being cited, and a brand can be mentioned without being recommended. Measuring them together hides where the problem is.
Are results published with this method?
Not yet. There is a study in progress, but no figures are published until they are verified and can be reviewed together with their methodology and limitations.
Does applying the framework guarantee appearing in ChatGPT?
No. The framework orders what to review and in what sequence. It cannot guarantee mentions, citations or recommendations, because they depend on systems Majotek does not control.
