Case study

Repetition Changes Mention Volume

How repeated wording changes the interpretation of visible brand volume.

This case study starts with 744 explicit brand mentions and then separates repeated, formulaic wording from the rest. The result shows why raw volume alone can overstate diversity of expression and why repetition needs its own analytical layer.

Abstract architecture composition
REPETITION
VOLUME
PATTERN
INTERPRETATION
RAW
COUNT
CLUSTERS
MEANING
744Explicit brand mentionsAll explicit mentions in the calibrated dataset layer
63Exact formulaicMentions that reuse the same wording pattern
21Exact clustersClusters of repeated formulaic language
681Unique-likeMentions that remain more varied in wording
REAL CONVERSATIONS.RIGOROUS METHODS.LASTING INSIGHTS.
Key finding

Raw mention volume can hide repetition.

A single total may look rich, but the wording underneath can repeat. Separating exact formulaic patterns from the broader set shows whether the volume reflects many distinct voices or repeated language structures.

63exact formulaic mentions detected
Mention comparison

Repeated vs. unique-like wording

63
Exact formulaic mentions
681
Unique-like mentions
21
Exact clusters

Most explicit brand mentions remain unique-like, but the repeated layer is large enough to matter for interpretation, benchmarking and campaign analysis.

Data & evidence

Repetition evidence layers

LayerCountInterpretation
Explicit Brand Mentions744The full calibrated visible naming layer
Exact formulaic mentions63Mentions using repeated wording patterns
Exact clusters21Grouped repetition patterns behind the repeated layer
Unique-like mentions681The remainder after repeated formulas are separated
What this means

Volume matters, but structure matters more.

Repetition is not fake visibility, but it is a different kind of visibility. A mention total becomes more informative when we know how much of it is repeated wording and how much represents broader expression diversity.

◌

Repeated language still matters.

Repeated formulas may signal strong framing, campaign consistency or shared narrative.

▥

Unique-like language signals range.

The non-formulaic layer shows how much variation exists in the way brands are described.

◫

Better interpretation supports better comparison.

Repetition-aware measurement makes it easier to compare brands and campaigns more responsibly.

Our conclusion

A bigger number is not always a richer conversation.

Once repetition is measured separately, mention volume becomes a more careful and more useful research signal.

Methodology

How the case was structured

01

Begin with explicit mention inventory.

The source is the calibrated set of 744 explicit brand mentions.

02

Detect exact formulaic wording.

Repeated text patterns are identified and grouped into clusters.

03

Interpret the layers separately.

The repeated layer and the unique-like remainder are then read as distinct evidence structures.

Extended analysis

Full case study notes

This case study begins with a number that looks complete:

744 explicit Brand Mentions.

If we stop there, the interpretation is simple.

The dataset contains 744 explicit mentions.

But that number does not tell us whether those mentions express 744 different formulations.

Repetition analysis changes the picture.

Inside Gold Dataset v1:

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

The total stays the same.

The meaning changes.

01

The raw view: 744 explicit Brand Mentions

Before repetition is measured, the 744-record layer looks uniform.

Every record contributes one occurrence.

That is useful for frequency.

It tells us how many explicit Brand Mentions were identified in Gold Dataset v1.

But frequency alone cannot answer a second question:

How much new language does the dataset contain?

A repeated sentence and a completely different sentence both add one to the count.

That is where a raw total becomes insufficient.

02

The classification step

The repetition layer separates records that reuse the same wording pattern from records that are not classified as duplicates under the current method.

The result is:

63 exact formulaic mentions

These are repeated formulations detected under the normalization rules.

21 exact formulaic clusters

These clusters group recurring formulations so that repetition can be studied as structure rather than as isolated duplicates.

681 unique-like mentions

These records were not classified as exact or near-text duplicates by the current method.

The word unique-like is important.

It does not mean organic.

It does not prove independent authorship.

It does not prove that the wording was unpaid, spontaneous or created by different people.

It means only that the current repetition method did not classify the wording as a duplicate.

03

What changed after classification

Before repetition analysis:

744 Brand Mentions

After repetition analysis:

63 exact formulaic mentions

plus

681 unique-like mentions

with

21 separate exact repetition clusters

The dataset did not shrink.

No records disappeared.

The difference is interpretive.

The total still measures frequency.

The repetition layer now also tells us something about language diversity.

That makes the same dataset more informative without changing the underlying count.

04

Why one percentage is not enough

The exact formulaic mentions represent 8.5% of the 744 explicit Brand Mentions.

The unique-like layer represents 91.5%.

Those percentages are useful, but the 21 clusters matter too.

Why?

Because 63 repeated mentions could come from one dominant template or from many smaller recurring formulations.

The cluster count tells us that repetition is distributed across multiple patterns.

That creates a different analytical question:

Which formulations repeat?

How large is each cluster?

Which brands, products or contexts appear inside those repeated patterns?

The current case study does not answer every one of those questions.

It shows why they become possible only after repetition is classified.

05

A simple example

Imagine 10 Brand Mentions.

Eight use exactly the same sentence.

Two use different wording.

The raw report says:

10 mentions.

That is correct.

But if the goal is to understand how much distinct language surrounds the brand, the raw total is incomplete.

After repetition analysis, the same data can be described as:

10 total mentions,

8 repeated formulations,

2 other formulations.

Nothing was removed.

The report simply became more precise.

That is the logic applied to Gold Dataset v1.

06

Why repeated mentions still matter

Repetition is not automatically a negative signal.

A repeated formulation can still create visibility.

It can still name the brand.

It can still appear on a public page.

It can still persist over time.

The research does not classify repeated mentions as worthless.

It classifies them as repeated.

That distinction matters because the analytical question changes depending on the goal.

If the question is frequency, every mention counts.

If the question is language diversity, repetition must be separated.

If the question is semantic breadth, repeated contexts may need to be distinguished from genuinely different contexts.

The same record can therefore remain valuable while being interpreted differently.

07

What the case tells us about volume

The key lesson is simple:

volume is not the same thing as diversity.

A Brand Mention total answers:

"How many occurrences were identified?"

Repetition analysis adds:

"How many of those occurrences reuse language already seen elsewhere?"

Those two questions should not be collapsed into one number.

08

The macro layer shows repetition at a different scale

The wider macro subset contains 1,493,085 offline threads.

Its repetition analysis works at the title level, not at the explicit Brand Mention wording level.

The macro layer identified:

  • 39,203 exact duplicate-title groups
  • 39,866 normalized repeated-title groups
  • 156,339 threads in exact duplicate-title groups
  • 168,663 threads — 11.30% — in normalized repeated-title groups

These figures are not an extension of the 63 formulaic Brand Mentions.

They describe a different phenomenon.

The Gold Dataset repetition layer looks at repeated wording inside explicit Brand Mentions.

The macro layer looks at repeated title framing across threads.

The two layers support the same general observation:

repetition exists and can be measured.

They should not be merged into one repetition rate.

09

Why this changes campaign interpretation

Imagine two campaigns that each produce 500 Brand Mentions.

Campaign A generates many repeated formulations.

Campaign B generates mentions across a wider range of language, questions, products and use cases.

Both campaigns can truthfully report:

500 mentions.

But that total hides a meaningful difference in the information environment.

One campaign may be creating frequency.

The other may be creating both frequency and broader language diversity.

The current research does not rank those outcomes by value.

It shows that they are different and should be reported separately.

10

Why this matters for AI Search research

AI systems may encounter repeated public information across many pages.

It is tempting to assume that repetition helps, hurts or becomes irrelevant.

The current study does not make that claim.

The research first establishes a cleaner measurement layer:

which mentions repeat,

which do not appear to repeat under the current method,

and how repetition clusters are distributed.

Only after those variables are measured can future experiments test whether repetition, diversity or persistence are associated with different AI outcomes.

Without that separation, the experiment would mix several different information environments together.

11

What this case does not prove

The case does not prove that repeated language is spam.

It does not prove that formulaic mentions are artificial.

It does not prove that unique-like mentions are organic.

It does not prove that more diverse language is automatically more valuable.

It does not prove that repetition causes or prevents AI recommendations.

The conclusion is narrower:

the same Brand Mention total can represent different levels of language repetition, so volume and diversity should be measured separately.

12

The practical takeaway

If a report says:

"this campaign generated 1,000 Brand Mentions,"

the next useful question is:

How much of that volume repeats the same wording?

Then:

How many repetition clusters exist?

How much of the total sits inside those clusters?

Does the repeated language reproduce the same product, use case or recommendation?

Do other mentions introduce new contexts?

Those questions turn a volume report into an information report.

13

Related research

The Repetition Problem
The main research finding behind this case.

Not All Brand Mentions Are Equal
Why mention structure also changes the meaning of raw totals.

Formulaic Language v1
The derived dataset behind the repetition classification.

Gold Dataset v1
The deep-analysis dataset containing the 744 explicit Brand Mentions.

Research Limitations
The boundaries that apply to repetition analysis.

Explore more case studies →