This page is the short evidence layer of the research program.
It is not a sales case study and it is not a list of methodology definitions.
Its purpose is to answer a simpler question:
What has actually been measured, classified and published?
The research is based on approximately ten years of FunkyMEDIA-owned publishing infrastructure and a current inventory of 1,779,805 unique discussion URLs.
That inventory was not treated as one giant Brand Mention count.
Instead, smaller research layers were created so that individual mentions could be classified and interpreted in detail.
Five evidence layers define the public research portfolio
The numbers describe different research layers and should not be combined into one population.
1,779,805 unique discussion URLs
The full research inventory.
5,000 threads / 8,708 posts
A representative validation layer used to inspect structure and chronology.
1,000 threads / 4,447 posts
The deep-analysis layer used for classification.
794 outbound links
Audited inside Gold Dataset v1.
744 explicit Brand Mentions
The explicit naming layer inside Gold Dataset v1.
536 relationship records
Extracted from Brand Mention context.
276 unique Brand → Relation → Entity pairs
The normalized semantic relationship layer.
Five findings that change how Brand Mentions should be interpreted
Each finding comes from a specific calibrated layer rather than from one undifferentiated total.
Links and Brand Mentions are not the same signal
Gold Dataset v1 contains:
- 416 linked explicit Brand Mentions
- 328 unlinked Brand Mentions
- 378 Linked Entity References
This demonstrates that visible brand naming can exist without a link, while a link to an entity-owned destination can exist without explicit visible naming.
That distinction is the foundation of the public Brand Mention taxonomy.
Repetition changes the meaning of mention volume
Among 744 explicit Brand Mentions:
- 63 were classified as exact formulaic mentions
- they formed 21 exact clusters
- 681 remained unique-like under the current method
The practical result is simple:
two brands can have the same mention count while producing very different amounts of language diversity.
The research therefore treats frequency and repetition as separate measurements.
Brands appear before the recommendation stage
The Gold Dataset calibration identified:
- 248 question / consideration signals
- 179 recommendation signals
- 19 negative / warning signals
- 8 strong first-hand experience signals
- 4 comparison signals
These are calibration counts, not estimates of full-corpus prevalence.
They show that brand visibility can begin while people are still asking, checking and considering options.
Recommendation is only one stage of the decision process.
Brand Mentions can contain structured semantic context
The semantic calibration contains:
- 536 relationship records
- 276 unique Brand → Relation → Entity pairs
- 139 brands with at least one detected semantic relationship
- 98 entities or relationship terms
The observed relationship families include Use Case, Product, Service, Feature, Problem, Result, Location and Co-Mention.
This is the evidence layer behind the Brand Context Graph.
The source material contains real commercial brands, not synthetic examples.
Recognisable brands observed in the research inventory include:
Allegro · Samsung · BMW · Toyota · IKEA · Lidl · Media Expert · Medicover · Volvo · Castorama
These names are shown because they appear in the analysed publishing inventory.
Their presence demonstrates that the corpus contains real commercial brand context distributed through the publishing infrastructure.
The research inventory itself does not prove the exact contractual relationship behind every individual publication.
A Brand Mention can create value without being owned by the brand
One of the more interesting observations from the publishing infrastructure is that Brand Mentions behave differently from owned media.
A brand can benefit from visibility on a page that it does not own.
The publisher controls the domain, URL, surrounding editorial context and long-term maintenance of the page.
The brand benefits from the mention, but it does not fully control the asset.
This makes Brand Mentions useful to think about as distributed brand assets.
They exist outside the brand's own website but can remain part of the public information environment for years.
The research does not claim that this persistence causes AI recommendations.
It shows that external brand context can remain available long after the original publication event.
Why the ten-year infrastructure matters
The corpus is not a random web crawl.
The research inventory comes from publishing infrastructure that has belonged to and been operated within the FunkyMEDIA ecosystem for approximately ten years.
That makes it possible to study more than a one-time snapshot.
The infrastructure creates the conditions for future analysis of persistence:
- when a brand first appeared
- when it last appeared
- how many months or years it remained visible
- whether the same brand returned in different contexts
- whether semantic territory changed over time
Those analyses are not yet presented as completed findings.
They are the next research layer made possible by the existing data.
Evidence first. Stronger conclusions later.
The portfolio does not claim that Brand Mentions cause AI recommendations.
It does not claim that unlinked mentions are automatically more valuable than linked mentions.
It does not claim that repeated wording is spam or artificial.
It does not claim that unique-like text is organic.
It does not claim that co-mentioned brands are necessarily competitors.
It does not claim that a Result relationship proves causality.
The value of the research is in separating these structures before stronger conclusions are attempted.