Maya AI Visibility Methodology

Prof. Dr. İbrahim Kırçova — Head of Research

Prof. Dr. İbrahim Kırçova is a faculty member in the Department of Business Administration, Faculty of Economic and Administrative Sciences, at Yıldız Technical University. His official academic profile lists marketing and management among his research areas.

His research leadership at Maya brings an academic perspective to work on brand visibility in a marketing context. How often a brand appears matters, but so does the customer need and competitive setting in which it appears.

Academic expertise does not replace the need to explain a study’s methods and evidence. A particular study should be assessed through its sample, calculations, and limitations. The university affiliation identifies Kırçova’s academic position; it is not a statement of institutional endorsement or partnership.

Explore Prof. Dr. İbrahim Kırçova’s academic profile and publications.

AI visibility describes how a brand appears in AI answers to a selected set of questions. It includes several observations that should not be reduced to an interchangeable set of scores.

A visit from an AI bot, a human website visit, and a sale are separate observations. Connecting them to answer visibility requires evaluating the relevant data sources separately.

A report is defined by its question set, date range, platforms, languages, and geographic scope. The same brand may appear differently across different customer needs.

Branded questions help examine how a brand is recognized and how its product information is described. Unbranded category questions examine discovery and consideration. These groups do not represent the same customer behavior.

For illustration, “What does Maya AI do?” and “Which tools should I consider for measuring AI visibility?” are different questions. The first already introduces the brand. The second examines whether the brand emerges among the options.

Adding a new product or customer segment changes the scope of a question set. Before comparing periods, check whether the question list and other conditions remain comparable. A selected set does not represent every possible question from every user.

The measurement process can be understood through five connected stages:

1. Scope: Establish the brand, questions, and comparison conditions.
2. Response: Evaluate the platform result within the relevant collection context.
3. Matching: Examine brand names, competitors, and source links.
4. Calculation: Apply the counting unit and denominator of the relevant report.
5. Interpretation: Read the result alongside the answer’s context and the limits of the measurement.

A result collected from one platform should not be presented as the exact experience of another platform or every individual user. Collection conditions, search behavior, and accessible information may differ. An unknown model version or personalization setting should not be treated as known.

The unit of measurement matters when comparing Maya reports. A platform response is one answer from a platform. An evaluation can combine results from multiple platforms in the relevant workflow.

Suppose, for illustration, that a question runs on four platforms and the brand appears in only one answer. Its response-level rate is one out of four. An evaluation-level measure asking whether the brand appeared on at least one platform can record that evaluation as containing a mention.

Different percentages across two reports therefore do not automatically establish a contradiction. First check whether the unit, scope, and denominator are the same.

Brand mention rate

The response-level calculation follows this principle:

For illustration, if a brand appears in 28 of 100 eligible responses, its rate is 28%. The response-level calculation excludes unsuccessful platform results and results without content from its denominator.

This rate does not count the real people who saw a brand. It describes the brand’s presence in the response sample. Evaluation-level reports use the relevant evaluation records as their counting unit instead.

Competitive share of voice

In the competitor leaderboard workflow, share of voice follows this calculation:

For illustration, suppose the brand appears in 30 evaluations, competitor A in 45, and competitor B in 25. The brand’s share in that comparison is 30%. More than one brand can appear in an evaluation, so the total mention count need not equal the number of evaluations.

Competitor selection and brand grouping affect the denominator. Comparisons should retain a consistent competitor scope.

Brands can appear as abbreviations, alternative names, or names within a brand family. Maya’s competitor evaluation workflow uses alternative names and brand-family grouping. Repeated appearances of the same competitor within one evaluation do not count as additional evaluations in that leaderboard.

Similar names do not necessarily identify the same organization. Context is particularly important when a brand name is also an ordinary word. For example, “maya” in a Turkish cooking recipe is not a mention of the Maya AI platform.

User-defined grouping and display choices can also affect a comparison. Tracking a sub-brand independently creates a different scope from grouping it under its parent brand.

Citation analysis examines the domains and pages linked in an answer. A domain view helps compare source websites; a page view helps examine particular pieces of content.

Citation records, unique sources, and responses containing a source are different measures. An answer can contain several links, so these counts should not be used interchangeably.

A source appearing in an answer is also different from a source supporting a specific claim. A linked URL does not establish that every sentence came from that page. The source content and the relevant statement should be examined together.

A page cited in the past may have changed or become unavailable. The historical observation and the page’s current content or availability are separate pieces of information.

Sentiment classifies the evaluative direction of the language used about a brand. Positive language is not the same as an explicit purchase recommendation.

The recommendation indicator in the sentiment report is based on evaluations in which the brand is mentioned and the adjusted sentiment value meets the positive threshold. It should be interpreted as a positive-classification-based indicator, preserving the distinction from a direct preference recommendation.

An answer might describe a brand’s strengths positively while finding another option more suitable for a particular use case. A positive answer can also contain incorrect product information. Perception and factual accuracy therefore require separate consideration.

An expression associated with a main question may be a provider-reported search, a related question, or a suggestion. Its origin determines what can be concluded from it. It is not appropriate to treat every related question as a search that actually occurred.

Maya’s fanout frequency calculation excludes results saved again through cache replay. Displaying the same collection event again should not create a new observation.

A sub-question and source URL appearing in the same answer do not prove that the URL was retrieved through that particular search step. Without a direct relationship in the available data, the finding is an observation of co-occurrence within the answer.

Retrospective traceability is about examining a past observation together with its supporting evidence. The question, platform, date, relevant answer passage, and source links provide the context needed to assess a finding.

The following table illustrates an evidence review. It is not a real customer record.

Reviewing a historical record is different from running the question again. A new answer may differ because the platform or the web has changed. Traceability explains a historical observation; it does not guarantee an identical future output.

On this page, SEO refers to work on discoverability in search engines; AEO refers to providing clear, reliable answers to questions; and GEO refers to work on representation and source visibility in generative AI answers. Their practical applications can overlap.

Maya data can help teams investigate the customer needs where a brand is absent, the sources that appear, and the existing pages that may need improvement. Not every gap calls for a new blog post. Product information, documentation, or technical access may need attention instead.

Google states that foundational SEO practices remain relevant to AI Overviews and AI Mode, and that special AI files are not required. Content structure or markup alone does not guarantee citation. Google Search Central guidance.

Question selection, language coverage, platform updates, and answer variability can affect results. A single answer should not be generalized to an entire market.

When a rate rises from 20% to 27%, the change is seven percentage points. Interpreting that change requires checking the question sets, platforms, and denominators in both periods. An increase after a content update does not establish that the update alone caused the increase.

Maya’s measurement approach is intended to help teams evaluate findings alongside their evidence and scope. Visibility results should not be presented as traffic, conversions, or commercial market share.

To evaluate where your brand appears, how it is described, and which pages are cited, explore Maya AI visibility tracking. For competitive context, learn about Maya’s benchmarking approach.