Research Limitations

1. The corpus is not the entire internet

The Full Research Inventory contains:

1,779,805 unique discussion URLs

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

That is a large research environment.

It is not a census of the entire web.

The corpus does not claim to represent:

all websites,

all forums,

all consumer discussions,

all industries,

all countries,

or every Brand Mention available online.

The findings describe the environment that was actually analysed.

2. The source environment is specific

The analysed material comes from publishing infrastructure that has belonged to and been operated within the FunkyMEDIA ecosystem for approximately ten years.

These are not randomly selected websites collected from the open web.

That creates useful continuity.

It also creates a specific source environment.

Patterns observed in this environment should not automatically be assumed to describe every publishing ecosystem.

3. Different datasets answer different questions

The research uses several layers.

The Full Research Inventory contains:

1,779,805 unique discussion URLs

Representative Validation contains:

5,000 randomly sampled threads

and

8,708 posts

Gold Dataset v1 contains:

1,000 threads

4,447 posts

794 outbound links

The Macro Subset contains:

1,493,085 offline threads

These populations are not interchangeable.

A finding from one layer should not be silently generalized to another.

4. Gold Dataset percentages are calibration results

Gold Dataset v1 supports detailed Brand Mention classification.

It contains:

744 explicit Brand Mentions

including:

416 linked explicit Brand Mentions

and

328 unlinked Brand Mentions

The unlinked share is:

44.1%

That percentage belongs to the Gold Dataset calibration.

It should not be presented as the prevalence of unlinked Brand Mentions across all 1,779,805 discussion URLs.

The same scope rule applies to repetition, intent and semantic calibration results.

5. Brand Mention counts are not discussion counts

A discussion URL is not a Brand Mention.

A thread is not a Brand Mention.

A post is not a Brand Mention.

One discussion can contain:

one Brand Mention,

many Brand Mentions,

or none.

Therefore:

1 discussion ≠ 1 Brand Mention

The research keeps those units separate.

6. Links are not equivalent to Brand Mentions

Gold Dataset v1 contains:

794 outbound links

But link count and Brand Mention count are different units.

The taxonomy separates:

Explicit Brand Mention — Linked

Unlinked Brand Mention

Linked Entity Reference

A link can exist without equivalent visible brand naming.

A Brand Mention can also exist without a link.

Therefore:

link ≠ Brand Mention

7. Unique-like does not mean organic

The repetition calibration contains:

63 exact formulaic mentions

21 exact formulaic clusters

681 unique-like mentions

The term unique-like is intentionally cautious.

It does not mean:

organic,

independently authored,

human-written,

or original in an absolute sense.

Therefore:

unique-like ≠ organic

The classification describes observed wording repetition.

It does not establish provenance.

8. Repetition does not prove manipulation

Repeated wording can result from:

templates,

republishing,

editorial reuse,

copying,

or other repeated-language processes.

The repetition analysis detects textual patterns.

It does not determine why they occurred.

Therefore:

repetition ≠ manipulation

Additional evidence would be required to support a claim about motive or coordination.

9. Question does not mean purchase intent

The intent calibration contains:

248 question / consideration signals

A question can indicate that the brand entered the decision process.

It does not prove that the user intended to purchase.

Therefore:

question / consideration ≠ purchase intent

The category describes conversational role.

It does not record conversion intent.

10. Recommendation does not mean purchase

The calibration contains:

179 recommendation signals

A recommendation means the wording suggests or endorses a brand.

It does not prove:

that the recipient bought,

that the recommendation changed behavior,

or that a transaction occurred.

Therefore:

recommendation ≠ purchase

11. First-hand language does not prove the experience occurred

The current intent calibration contains:

8 strong first-hand experience signals

These signals show stronger direct-use language.

They do not independently verify:

ownership,

use,

duration,

or outcome.

Therefore:

strong experience evidence ≠ verified experience

The text is evidence of a claim.

It is not independent proof of the underlying event.

12. Extracted relationship does not mean verified fact

Semantic Relationships v1 contains:

536 relationship records

across:

276 unique Brand → Relation → Entity pairs

These relationships are derived from the analysed text.

If a source connects a brand with:

a product,

a result,

a feature,

or a problem,

the relationship can be structured.

But the extraction itself does not verify the underlying real-world claim.

Therefore:

extracted relationship ≠ verified fact

13. Result does not mean causality

The semantic calibration contains:

23 Result relationship records

A Result relationship records an outcome claimed in the text.

It does not prove that:

the brand caused the outcome,

the outcome was measured correctly,

or the result would reproduce elsewhere.

Therefore:

Result ≠ causality

14. Co-Mention does not mean competition

The semantic calibration contains:

64 Competitor / Co-Mention records

Two brands can appear together because they are:

alternatives,

complementary,

used together,

compared,

or simply part of the same category discussion.

Therefore:

Co-Mention ≠ competition

Competitive interpretation requires the surrounding context.

15. Product relationship does not automatically mean ownership

A Brand → Product relationship can exist because the brand:

manufactures,

sells,

services,

compares,

or is otherwise associated with the product.

The relationship records textual association.

It does not automatically prove official product ownership.

16. Location relationship does not automatically mean physical presence

A Location relationship may describe:

service context,

market discussion,

customer location,

or another geographic association.

It does not automatically establish:

headquarters,

office presence,

or service coverage.

17. More relationships do not mean more authority

A brand can have many semantic relationships.

That can reflect:

breadth,

complexity,

varied discussion,

or multiple product and use-case contexts.

It does not automatically mean:

greater authority,

greater quality,

or greater commercial value.

Therefore:

relationship volume ≠ authority

18. Broader semantic territory is not automatically better

A broad semantic territory can reflect many contexts.

A narrow semantic territory can reflect strong concentration.

Neither is automatically superior.

The graph layer describes structure.

It does not rank brands by semantic breadth.

19. Framework levels are not universal standards

The research uses calibrated frameworks such as:

BCD0–BCD5

Recommendation Strength RS0–RS5

Experience Evidence 0–5

and

Brand Mention Complexity

These frameworks are research tools.

They are not presented as universal industry standards.

They are designed to keep analysis consistent within this research program.

20. Framework dimensions should not be collapsed into one master score

Brand Context Depth,

Recommendation Strength,

Experience Evidence,

link structure,

repetition,

and semantic relationships

measure different properties.

Combining them automatically into one number would hide those differences.

The methodology therefore does not present one universal Brand Mention score.

21. Macro title analysis is not Brand Mention content analysis

The Macro Subset contains:

1,493,085 offline threads

and includes structural findings based on:

titles,

paths,

categories,

and repeated-title patterns.

For example, the current analysis contains:

39,203 exact duplicate-title groups

and

39,866 normalized repeated-title groups

These are title-level structural findings.

They are not explicit Brand Mention counts.

The two layers should not be merged.

22. Historical persistence remains a research question

The publishing infrastructure spans approximately ten years.

That creates the possibility of studying how long Brand Mentions and associated contexts persist.

But the current study does not yet establish:

how long individual mentions remain influential,

whether old mentions affect current AI outputs,

or how temporal decay operates.

Persistence is a future research question.

23. External brand context does not imply brand control

Many Brand Mentions exist on pages the brand does not own.

That makes them useful for studying external brand context.

But the brand does not necessarily control:

the page,

the surrounding wording,

the URL,

the publisher,

or the long-term availability of the content.

This is why the research can describe Brand Mentions as distributed brand assets only in a structural sense.

The term does not imply legal ownership or editorial control by the brand.

24. Brand presence does not establish commercial relationship

The corpus contains real brands.

A brand's presence in the inventory does not, by itself, establish:

client status,

partnership,

campaign relationship,

or a specific commercial model behind the publication.

The research can document observed presence.

Commercial relationship requires separate evidence.

25. Human discussion patterns do not automatically determine AI output

This is the most important boundary for AI Search interpretation.

The research measures human-readable and publisher-hosted brand context.

It does not prove that:

AI systems ingest every page,

AI systems weight every mention equally,

a Brand Mention causes an AI recommendation,

or a specific graph relationship changes model output.

Therefore:

Brand Mention presence ≠ proven AI influence

The current research creates variables that can be tested against AI-generated outputs.

It does not skip that validation step.

26. Correlation and co-occurrence are not causality

If a brand appears frequently around a product, problem or use case, that pattern can be measured.

If the same brand later appears in an AI answer, both observations may be interesting.

But co-occurrence across systems is not, by itself, proof of causality.

A causal claim would require a research design capable of isolating the effect.

The current study does not make that leap.

27. Dataset size does not remove interpretation risk

Large datasets reduce some forms of uncertainty.

They do not eliminate:

measurement error,

classification ambiguity,

sampling limitations,

context loss,

or overinterpretation.

Scale is valuable.

It is not a substitute for disciplined interpretation.

28. Public methodology is intentionally narrower than internal operations

The public methodology documents:

the research units,

the dataset layers,

the classification concepts,

and the interpretation boundaries.

It does not publish every internal operational procedure used to prepare, review or maintain the research environment.

That is intentional.

The goal is reproducible interpretation of the published findings, not disclosure of internal production operations.