GEO Methodology
The Magna GEO Framework is the five-layer methodology Magna Dijital uses in its GEO and AI visibility work. We don't look only at the outcome of how a brand is described in ChatGPT, Gemini, Perplexity and Google AI Overviews answers; first we make clear who the brand is, tie its claims to proof, make its information accessible, examine its relationship with outside sources, and measure the result regularly under the same conditions.
On this page we explain the method behind our work. If you want us to apply this method to your brand, choose our GEO Services; if you first want to see only where you stand today, start with an AI Visibility Analysis.
Which Questions Does Our Methodology Keep Apart?
We don't see GEO work as limited to asking questions, producing content, adding schema or tracking a visibility score. In every project we keep two questions apart: How ready is your brand's information and technical foundation? and How is your brand actually described in AI answers? The first is readiness, the second is observed visibility. They are related but not the same; a technically ready site doesn't automatically appear in AI answers, and a brand that doesn't appear doesn't always have a technical problem.
The Five Layers of the Magna GEO Framework
| Layer | The question we ask | What we do | Output |
|---|---|---|---|
| Entity Clarity | Who is the brand, and is it described everywhere with the same facts? | We tie brand and service information to a single source of truth and check names and page roles. | Consistent brand and service definitions |
| Proof Integrity | Which verifiable proof are the important claims based on? | We link every claim to its primary proof page and clear out contradictions in figures and scope. | Claims matched with proof |
| Retrieval Readiness | Can the information be found and read? | We examine technical access, HTML structure, internal links and the clarity of content sections together. | An accessible, readable information architecture |
| Citation Layer | Which outside sources can confirm the brand's information? | We look at consistency across company pages, research, success stories, outside profiles and third-party sources. | A verifiable source network |
| Measurement Loop | How is the brand described in AI answers, and how is that changing? | We measure regularly with defined question sets and platforms, under recorded conditions. | Comparable visibility measurement |
Entity Clarity and a Single Source of Truth
In the Entity Clarity layer, we make sure your brand is described with the same company, the same service structure and the same core facts across different pages and outside sources. Brand name, legal name, founding information, location, service lines, and product and tool names all belong to this layer. When the same brand is described with different services or different facts on different pages, entity drift sets in and it becomes harder for AI engines to recognize the brand correctly.
To prevent this, we use a single source of truth (SSoT): we define in one document which information is authoritative. The goal is not to repeat the same text on every page, but to have the service page, company page, success story and research support the same core facts without contradiction, even though each does a different job. We run our own site with this method too.
Proof Integrity
A claim written on a site is not proof by itself. In the Proof Integrity layer, we set a primary proof page (canonical proof surface) for every important claim: a success story, research, a measurement output or a trusted outside source. If a client result has been verified in a success story, we don't use the same result on another page with a different figure, period or scope. We don't present an SEO result as a GEO result, or traffic from AI platforms as AI visibility. We explain how we build this link in content on our GEO-Ready Content page.
Retrieval Readiness
In the Retrieval Readiness layer, we look not just at whether your information exists, but at whether it can be found, read and understood in the right context. robots.txt, sitemaps, heading hierarchy, schema, internal links and canonical setup are the technical side of this layer. On the content side, we look not at length but at which question each section answers and whether it makes sense when read on its own. We audit the technical side with Magna GEO Auditor and build the Q&A structure as part of our AEO Services.
Citation Layer
AI engines don't read only your own site about your brand. In this layer, we examine whether company profiles, industry publications, business directories, partner pages and news describe your brand with the same facts. Our goal is not to collect as many links or mentions as possible, but to have different sources support the same correct information. We run this outside source work as part of our GEO Services.
Measurement Loop
The Measurement Loop turns GEO work from a one-off check into an ongoing process. We measure the same question set, on the same platforms, under the same language and market conditions, at regular intervals. In each measurement, we record which questions, platforms, competitors and sources the brand appears with, whether its information is correct and what has changed since the previous measurement. We run the measurement with Magna AI Analyzer.
Our Process: Seven Stages
The five layers show what we look at; the seven stages show how we run a GEO project:
| Stage | What we do | Output | The limit we set ourselves |
|---|---|---|---|
| 1. Scope | We define the target market, language, industry, competitor set, AI platforms and commercial questions. | The scope of work and measurement | We don't interpret results from different markets or languages as a single dataset. |
| 2. Baseline measurement | We record current visibility, technical readiness, entity consistency and proof structure. | A baseline table | Without knowing the baseline, we don't speak definitively about the cause of later change. |
| 3. Diagnosis | We classify problems according to the five layers. | A problem map | We don't assume every visibility gap has the same cause. |
| 4. Prioritization | We rank actions by impact, strength of evidence, feasibility and measurability. | A priority action plan | We write out the reasoning behind decisions instead of invented scores. |
| 5. Implementation | We make the technical, content, entity, proof, internal link, schema and outside source changes. | Updated information and technical structure | We don't claim that a single action produces a result on its own. |
| 6. Measurement | We measure visibility and readiness again under the same conditions. | A period-over-period comparison | We don't treat a single answer or a single measurement as a permanent state. |
| 7. Improvement | We set the next actions based on the new findings. | A new cycle of work | As AI platforms change, we reassess the scope as well. |
In this process, tools are not the starting point. First we define which problem we are measuring, then we bring in the right tool; that way tool outputs become evidence for the methodology rather than scores cut off from their context. We explain the difference between our tools on our AI Visibility Tools page.
How Do We Measure AI Visibility?
In measurement we keep three classes apart. Readiness is the technical and structural state of your site and information sources. Observed visibility is how your brand appears in AI answers. Downstream outcomes are results that come afterward, such as traffic from AI platforms to your site. A brand being mentioned doesn't mean it is recommended, and a recommendation doesn't guarantee traffic or sales. That is why we measure each signal separately:
| Class | Signal | What it tells you | What it doesn't tell you |
|---|---|---|---|
| Readiness | Technical and source readiness | Whether the site can be accessed, crawled and read | That the brand appears in AI answers |
| Observed | Mention rate | The share of answers in the question set that name the brand | That it is cited or recommended |
| Observed | Citation rate | The share of answers that explicitly cite your site as a source | That the brand is preferred |
| Observed | Recommendation rate | How often the brand is among the recommended candidates in decision questions | A sales or lead result |
| Observed | Correctness | Whether the brand's facts and claims are described correctly in the answer | The level of visibility |
| Observed | Position | Where the brand ranks in a list | That it is described correctly or strongly |
| Observed, rubric-based | Preference | How well-reasoned the brand's presentation is compared with alternatives | An absolute score without predefined criteria |
| Observed, calculated | AI Share of Voice | Relative visibility against the competitor set | A comparison with other studies when the method isn't disclosed |
| Downstream | AI referral traffic | Sessions coming to your site from AI platforms | Total visibility in AI answers |
When we measure correctness, we don't ask a general "is the answer right?" question; we check core facts such as the brand name, founding information, service lines and tool names (canonical facts), and claims containing figures or client results (canonical claims), one by one. When we find wrong or outdated information, we go back not only to the AI answer but also to the sources of that information on the web.
How Do We Build Question Sets?
The question set (prompt set) defines which user intents, which language and market conditions and which commercial decisions we will track. Producing a large number of questions doesn't make a good measurement on its own; the questions have to relate to the brand's real commercial situation. We build sets from six question types:
| Question type | The user's intent | Example |
|---|---|---|
| Discovery | Understanding a topic or solution | "How does GEO work?" |
| Recommendation | Looking for a provider for a specific need | "Which agencies offer GEO services?" |
| Comparison | Comparing options | "What is the difference between GEO and AEO?" |
| Validation | Confirming a brand's expertise | "What does Magna Dijital do in GEO?" |
| Problem and solution | Solving a specific problem | "Why doesn't my brand show up in ChatGPT?" |
| Brand-focused | Asking directly about the brand, tool or method | "What is the Magna GEO Framework?" |
When we make a significant change to the question set, we don't interpret the old and new measurements as one continuous series. We keep the version of each set and the dates it was used in our project records.
How Do We Keep Measurements Comparable?
When the question set, language, country, platform or measurement period changes, reading the results as part of the same trend is misleading. That is why, in every measurement, we hold the following constant or record any change:
| What we record | The risk if it isn't recorded |
|---|---|
| Question set | A change in the questions can be read as a change in visibility. |
| Language and country | Results from different markets can be merged into a single visibility. |
| Platform | Different AI platforms can be interpreted as if they behave the same way. |
| Measurement date | An old observation can be presented as the current state. |
| Number of repetitions | A single answer can be mistaken for the platform's general behavior. |
| Web access | Answers produced with and without web search can be taken as comparable. |
| Raw evidence | If the answer text and sources aren't kept, the result can't be verified later. |
This discipline doesn't make measurement certain, but it shows clearly under which conditions each result was obtained and where the limits of comparison lie.
A Study Where We Applied the Methodology
We applied these rules in our research, AI Visibility of Turkish Automotive Suppliers in Germany 2026:
- Scope: We defined six German-language commercial questions a German buyer would ask; two were Europe-wide and four were Türkiye-focused.
- Comparability: We asked the same questions in ChatGPT, Gemini, Meta AI, Cerebras and Google AI Overviews, in three separate measurement rounds, and reported each platform's data coverage separately.
- Measurement method: We calculated how far the platforms recommended the same companies using the Jaccard overlap ratio.
- Finding: Overlap was 40.9% for Europe-wide questions but fell to about 3.5% for Türkiye-focused questions; 84.5% of the 103 distinct company names appeared on only one platform.
- Limit: Because of caching on one platform, we did not interpret repeated answers as model stability, and we stated this clearly in the report.
A measurement on a single platform or with a single question could not have shown this picture. You can find our other studies on our Magna Research page.
How Do We Prioritize Actions?
We don't try to solve every problem at once, and we don't use a fixed order or a single score for every project. We set priorities with four questions:
| Question | What we look at | Example decision |
|---|---|---|
| Impact | How much does the problem affect the brand's accuracy or visibility? | For a brand described with the wrong service, we fix the entity information before writing new content. |
| Strength of evidence | Is the problem supported by measurement, an audit or observed answers? | We prioritize a directly observed problem over an action based on assumptions. |
| Feasibility | Can the fix be made with the current site, team and technical setup? | We move a critical information fix that can be done quickly ahead of a long development task. |
| Measurability | Can the change be measured again under the same conditions? | We prefer a trackable change over a broad goal that can't be measured. |
The Limits of Measurement
AI answers are variable and sensitive to context. The same question can produce different answers at different times; platform, language, market, web access and model updates all affect the result. We consider stating these limits openly part of the methodology:
- A single question or a single measurement doesn't represent a brand's overall AI visibility.
- Mentions, citations and recommendations are not the same signal, and a recommendation is not a guarantee of sales, leads or traffic.
- Technical readiness is not a visibility result on its own, and organic search ranking doesn't count as an AI recommendation.
- We don't merge results from different platforms as if they were one behavior.
- We don't attribute a change to a single GEO action without sufficient evidence.
- No methodology can guarantee appearing, being cited or being recommended in AI answers.
How Do We Keep the Methodology Up to Date?
The Magna GEO Framework isn't tied to a particular tool or to the behavior of a single AI platform. When platform coverage, measurement conditions or the way AI engines access information change, we reassess the relevant sections. We record changes that affect measurement classes, metric definitions, question set rules or comparison rules, and we update this page.
Let's Assess Your Brand With This Framework
Let's look together, within the same framework, at your brand's entity and proof structure, technical readiness, outside sources and where it stands in AI answers today.
Fill in the contact form or call us at +90 850 333 80 91.
If you want to run the work with your own team and set up the methodology together with us, see our GEO Consulting page; if you are choosing an agency for your GEO work, see our GEO Agency page.
Frequently Asked Questions About GEO Methodology
What is a GEO methodology?
A GEO methodology is a working framework that looks at how a brand is described in AI answers not only through the outcome, but together with its entity information, proof, retrievability, outside sources and measurement conditions.
What is the Magna GEO Framework?
The Magna GEO Framework is the five-layer methodology Magna Dijital uses in its GEO and AI visibility work: Entity Clarity, Proof Integrity, Retrieval Readiness, Citation Layer and Measurement Loop.
Why does a single source of truth (SSoT) matter?
A single source of truth defines which information is authoritative for a brand's core facts, service definitions and claims. That way different pages and sources support the same information without contradiction, and AI engines recognize the brand more accurately.
What is a canonical proof surface?
A canonical proof surface is the primary source where an important claim is verified: a success story, research, a measurement output or a trusted outside source. By linking the same claim to this page everywhere, we prevent contradictory figures from being used on different pages.
What stages does a GEO project have?
Our projects consist of scope, baseline measurement, diagnosis, prioritization, implementation, measurement and improvement. The weight of each stage depends on the brand's problems.
How do you measure AI visibility?
We ask defined question sets on the selected platforms regularly, under recorded conditions. We measure mentions, citations, recommendations, correctness and position separately, and evaluate technical readiness and traffic from AI platforms in separate classes.
What is the difference between a mention, a citation and a recommendation?
A mention is when the brand's name appears in the answer. A citation is when the brand's site is explicitly used as a source in the answer. A recommendation is when the brand is presented as an option in a decision question. A brand can be mentioned without being cited, or cited without being recommended; that is why we measure all three separately.
How should AI Share of Voice be interpreted?
AI Share of Voice shows a brand's relative visibility against its competitors in a defined question set. There is no universal formula, so our reports state clearly which platforms, questions and competitors we included. We don't compare a share whose method isn't disclosed with other studies.
Can ChatGPT, Gemini, Google AI Overviews and Perplexity be measured the same way?
They can be measured with the same discipline, but they don't behave the same way. Because answer format, use of sources and the relationship with web search differ from platform to platform, we evaluate the results for each platform separately.
Does a GEO methodology guarantee results?
No. No methodology can guarantee appearing, being cited or being recommended on a particular AI platform, or an increase in traffic and sales. Our methodology helps diagnose problems correctly, solve them in the right order and measure the result in a comparable way.
Written by Mert Feke, Co-Founder, Magna Dijital · Last updated: November 3, 2026