Methodology

1. Inventory layer

The Full Research Inventory contains:

1,779,805 unique discussion URLs

from approximately ten years of publishing infrastructure operated within the FunkyMEDIA ecosystem.

The inventory defines the source environment.

Its main purpose is scale and structure.

It is not itself a Brand Mention count.

A discussion URL can contain:

one Brand Mention,

many Brand Mentions,

or none.

This distinction remains fundamental throughout the methodology.

2. Representative validation

A smaller validation layer is used to inspect the wider inventory structure.

The Representative Validation layer contains:

5,000 randomly sampled threads

and

8,708 posts

Its role is to check the shape of the source material before deeper classification.

Validation is not the same as semantic analysis.

It is an intermediate layer between the Full Research Inventory and Gold Dataset v1.

3. Gold Dataset v1

Gold Dataset v1 is the deep-analysis layer.

It contains:

1,000 threads

4,447 posts

794 outbound links

This layer is used for:

Brand Mention taxonomy,

repetition calibration,

intent calibration,

and semantic relationship extraction.

The Gold Dataset provides depth rather than full-corpus coverage.

Findings derived from this layer remain attached to this layer unless separately validated elsewhere.

4. Brand Mention taxonomy

The public Brand Mention taxonomy separates three structures:

Explicit Brand Mention — Linked

Unlinked Brand Mention

Linked Entity Reference

The current calibration contains:

416 linked explicit Brand Mentions

328 unlinked Brand Mentions

378 Linked Entity References

This creates:

744 explicit Brand Mentions

and:

1,122 taxonomy records

The purpose of the taxonomy is to prevent two common errors:

counting only links and missing unlinked mentions,

or

counting every link as an explicit Brand Mention.

5. Repetition calibration

The repetition layer analyses wording inside the 744 explicit Brand Mentions.

The current calibration contains:

63 exact formulaic mentions

21 exact formulaic clusters

681 unique-like mentions

The distinction separates volume from wording diversity.

The term unique-like does not mean organic.

It means only that the wording was not classified as exact or near-formulaic repetition under the current calibration.

6. Intent calibration

Brand Mentions can perform different roles inside a discussion.

The current calibration includes:

248 question / consideration signals

179 recommendation signals

19 negative / warning signals

8 strong first-hand experience signals

4 comparison signals

These categories are used to describe conversational role.

They do not create a mandatory funnel.

A question does not prove purchase intent.

A recommendation does not prove purchase.

An experience statement is still a claim contained in the text.

7. Semantic relationship extraction

The semantic layer converts contextual relationships into structured:

Brand → Relation → Entity

records.

The current calibration contains:

536 relationship records

276 unique pairs

139 brands with at least one detected semantic relationship

98 entities or relationship terms

The public relationship families are:

Use Case,

Competitor / Co-Mention,

Product,

Feature,

Service,

Problem,

Result,

and Location.

These records are the foundation of the graph layer.

8. Macro structural analysis

A separate Macro Subset contains:

1,493,085 offline threads

This layer is used for large-scale structural analysis of:

titles,

paths,

categories,

and repeated title patterns.

Macro structural findings are not merged with Gold Dataset Brand Mention findings.

They use different populations and different units.

The central methodological rule

The most important rule is:

a conclusion should not be broader than the dataset that produced it.

A Gold Dataset percentage is a Gold Dataset result.

A Macro Subset percentage is a Macro Subset result.

An inventory count is an inventory count.

A semantic relationship count is not a Brand Mention count.

A title-structure finding is not a content-level Brand Mention finding.

This separation is what keeps the research interpretable.

Why the methodology avoids one universal score

The research does not collapse all signals into one master score.

A linked mention and an unlinked mention are different structures.

A recommendation and a question are different conversational roles.

A repeated phrase and a unique-like phrase are different wording patterns.

A Product relation and a Result relation are different semantic relationships.

Combining them into one universal number would remove the distinctions the research is designed to preserve.

Calibration frameworks

The research program also uses calibrated conceptual frameworks for:

Brand Context Depth,

Recommendation Strength,

Experience Evidence,

and Brand Mention Complexity.

These frameworks help organize analysis.

They are not presented here as universal scoring systems.

Their role is to create consistent analytical language for research comparisons.

Detailed scale definitions belong in the technical appendix rather than on the main research pages.

Why methodology and findings are separated

Methodology explains:

what was measured,

which population was used,

and what the analytical boundaries are.

Findings explain:

what the data showed.

Keeping those functions separate reduces the risk of turning methodological choices into conclusions.

It also makes individual findings easier to audit.

Conservative interpretation rules

Several rules apply across the entire research program.

Extracted relationship ≠ verified fact

A relationship record reflects what appears in the analysed text.

Result ≠ causality

A claimed outcome does not prove that the brand caused it.

Co-Mention ≠ competition

Two brands appearing together do not automatically form a competitor pair.

Unique-like ≠ organic

Non-repeated wording does not prove origin or authorship.

Question ≠ purchase intent

Consideration is not conversion.

Recommendation ≠ purchase

An endorsement does not prove a transaction.

These boundaries are part of the methodology, not optional disclaimers.

Why this matters for AI Search research

The methodology creates measurable variables that can later be compared with AI-generated outputs.

Those variables include:

linked versus unlinked explicit mentions,

repetition,

conversational role,

semantic breadth,

relationship frequency,

and co-mention structure.

The current methodology does not assume that any one of those variables directly causes AI visibility or recommendation.

It creates a controlled vocabulary and dataset structure for testing those questions.

What the methodology does not claim

The methodology does not claim to represent the entire internet.

It does not claim that Gold Dataset percentages apply to all 1,779,805 discussion URLs.

It does not claim that extracted relationships are verified real-world facts.

It does not claim that more Brand Mentions automatically produce stronger AI visibility.

It does not claim that one framework score can summarize brand performance.

And it does not treat correlation, co-occurrence or textual association as causality.

Where to go next

Use Brand Mention for the core unit definition.

Use Brand Mention Taxonomy for linked, unlinked and entity-reference distinctions.

Use Formulaic Language for repetition.

Use Semantic Relationships for Brand → Relation → Entity structure.

Use Research Limitations for the boundaries that apply across the entire program.