1.7 Million Conversations. 10 Years of Brand Mentions.

Brand Mention volume alone does not tell you what happened

For years, Brand Mentions have often been reduced to a number.

How many times was the company mentioned?
How many links pointed to the brand?
How many recommendations appeared?

Those numbers are useful, but they are not the whole story.

A brand can be named with a link. It can be named without a link. A page can link to a company without visibly naming it. A user can ask about a brand before recommending it. A recommendation can repeat the same formula used in dozens of other places. One mention can contain only a brand name, while another can connect the brand with a product, problem, feature, experience and result.

All of those records may look similar in a spreadsheet.

They are not the same kind of evidence.

That is the problem this research program was built to measure.

A decade of FunkyMEDIA-owned publishing infrastructure

The source material is not a random selection of unrelated websites collected from the internet.

The discussion URLs in this research come from publishing infrastructure that has belonged to and been operated within the FunkyMEDIA ecosystem for approximately ten years.

That matters because the dataset has a known operational history.

We can observe Brand Mentions that were published years ago and remained part of the public web. We can examine repeated wording, long-lived references, recurring brand contexts and the difference between a publication event and the years of visibility that may follow it.

This does not mean that the corpus represents the entire internet.

It means that the research is based on a defined, long-running publisher environment whose history is known to FunkyMEDIA.

That makes persistence itself a research question.

A link is not always a Brand Mention

One of the first things the Gold Dataset showed was that a link to a company website and an explicit Brand Mention are not the same thing.

In Gold Dataset v1 we separated:

• 416 Explicit Brand Mentions — Linked
• 328 Unlinked Brand Mentions
• 378 Linked Entity References

The first two classes explicitly name the brand.

The third class records a link to an entity-owned destination even though the visible text does not necessarily name the brand.

This changes how link data should be interpreted.

If we counted only links, we would miss unlinked Brand Mentions.

If we counted every company-owned link as an explicit Brand Mention, we would overstate visible naming.

What this means in practice

Brand visibility can exist without a link, and a link can exist without explicit brand visibility.

The two signals overlap, but they measure different things.

Read the study →

Repetition changes the meaning of volume

A high mention count can create the impression of broad visibility.

But if part of that volume repeats the same language, the number of mentions and the diversity of information are no longer the same thing.

Among 744 explicit Brand Mentions in Gold Dataset v1:

• 63 were classified as exact formulaic mentions
• they formed 21 exact formulaic clusters
• 681 remained unique-like under the current method

The important word here is unique-like.

It does not mean organic.

It means only that the current repetition method did not classify the text as an exact or near-text duplicate.

Why this matters

Two brands can have the same number of mentions while producing very different information environments.

One may be surrounded by repeated wording.

Another may appear in a much wider range of questions, experiences, products and use cases.

Volume alone cannot show that difference.

Read the study →

Brands appear before people recommend them

Brand visibility does not begin only when somebody says:

“I recommend this company.”

In the Gold Dataset calibration layer we identified:

• 248 question / consideration signals
• 179 recommendation signals
• 19 negative / warning signals
• 8 strong first-hand experience signals
• 4 comparison signals

These are calibration results, not full-corpus prevalence estimates.

But they show something important.

A brand can enter the decision process while somebody is still asking whether the company is worth considering, whether anyone has used it, how it compares with another option or whether it solves a particular problem.

The research implication

Recommendation is only one stage of brand visibility.

Questions and consideration can place a brand inside the decision journey earlier.

Read the study →

One Brand Mention can contain an entire semantic network

A Brand Mention is not always one piece of information.

One passage can connect a brand with a product, feature, service, problem, use case, location, result and another brand.

The semantic calibration currently contains:

• 536 semantic relationship records
• 276 unique Brand → Relation → Entity pairs
• 139 brands with at least one detected semantic relationship
• 98 entities or relationship terms

The most frequently observed relationship family in the current calibration is Use Case, followed by other relationship types including Product, Feature, Service, Problem, Result, Location and Co-Mention.

Why this changes the analysis

A mention count tells us that a brand appeared.

A semantic relationship tells us what the brand appeared with.

That difference is the foundation of the Brand Context Graph.

Explore Brand Context Graph →

Brand Mentions can behave like distributed brand assets

There is another property of Brand Mentions that is easy to overlook.

The brand benefits from the visibility, but it often does not own the page on which that visibility exists.

Once a Brand Mention is published on a third-party publisher, the company may gain exposure, discoverability and context, but it does not fully control the domain, URL, surrounding editorial environment, internal linking structure or future maintenance of that page.

The publisher controls the infrastructure.

The brand benefits from the presence.

This makes Brand Mentions fundamentally different from owned media.

They can behave like distributed brand assets: created for the benefit of the brand, but hosted and maintained outside the brand's own website.

The long history of FunkyMEDIA's publishing infrastructure makes this particularly interesting, because some mentions can remain public for years after the original publication event.

The corpus contains real brands, not abstract placeholders

The analysed publishing inventory includes discussions referencing both specialist companies and widely recognised consumer and B2B brands.

Examples observed in the research material include:

Allegro · Samsung · BMW · Toyota · IKEA · Lidl · Media Expert · Medicover · Volvo · Castorama

These names are shown as examples of brands observed in the research inventory.

Their presence demonstrates that Brand Mentions inside the FunkyMEDIA publishing infrastructure have been used around real commercial entities and can persist as part of the public web.

The research corpus by itself does not establish the exact contractual model behind every individual publication.

Scale first. Deep analysis second.

The project uses several separate research layers because one dataset cannot answer every question.

Full Research Inventory

1,779,805 unique discussion URLs

This is the scale layer.

It defines the source pool and supports sampling and macro structural analysis.

It is not a count of Brand Mentions.

Representative Validation

5,000 threads / 8,708 posts

This layer is used to validate structure, chronology and the shape of the wider inventory.

Gold Dataset v1

1,000 threads / 4,447 posts / 794 outbound links

This is the deep-analysis layer.

It supports the Brand Mention taxonomy, repetition analysis, intent calibration and semantic relationship extraction.

Semantic Calibration

536 relationship records / 276 unique pairs

This layer converts richer Brand Mention context into structured Brand → Relation → Entity records.

The layers are deliberately kept separate.

A number from one layer should not silently become a claim about another.

Evidence has boundaries

The research does not claim that Brand Mentions cause AI recommendations.

It does not claim that unlinked mentions are automatically more valuable than links.

It does not claim that repeated wording is automatically spam or artificial.

It does not claim that unique-like text is proven organic.

It does not claim that brands appearing together are always competitors.

It does not claim that a Result relationship proves causality.

Those are separate questions.

The purpose of the research is to measure the structure of Brand Mentions before making stronger claims about their effect.

From agency practice to public research

FunkyMEDIA works in AI Search and Brand Mentions.

That creates a responsibility to separate commercial opinion from measurable evidence.

FunkyMEDIA Research Lab was created to publish the underlying research publicly: datasets, findings, graphs, limitations and methodology that can be inspected independently of a sales presentation.

The intention is not to make every reader learn the full classification system.

It is to make the evidence visible enough that a journalist, marketer, researcher or AI system can understand where a claim came from and how strong that claim actually is.

Methodology and research interpretation: Rafał Cyrański
Publisher: FunkyMEDIA / FunkyMEDIA Research Lab
Service operator: BIGTECH IT Sp. z o.o.