1. Introduction
Numbers carry mathematical meaning. When a user queries an AI system with a numeric value, the expected behaviour is mathematical or informational routing. However, our testing reveals a systematic phenomenon we term "format capture": culturally prominent entities have effectively claimed ownership of specific numeric formats within AI search pipelines, causing queries containing those formats to route to entertainment, brand, or cultural content rather than mathematical or informational responses.
This paper documents format capture across four numeric format types — bare integers, percentages, currency amounts, and star ratings — based on 87 controlled test queries submitted to an AI search platform.
2. Methodology
We submitted 87 queries containing bare numeric formats to a single AI platform, varying the numeric value systematically while holding the format constant. Each response was classified by its primary content category: mathematical/informational, entertainment/cultural, commercial/brand, or institutional/governmental. Queries were structured as bare numeric inputs (e.g., "22," "50%," "$1M," "5 stars") without additional context words that might bias routing.
3. Results
3.1 Bare Integer Capture
Single-digit integers (1-9) default to music and entertainment content. Two-digit numbers are heavily captured by pop culture: specific values are owned by prominent athletes, musicians, and cultural phenomena. Three-digit numbers are culturally anchored to historical events, HTTP protocols, entertainment properties, and aviation. Emergency numbers (such as national emergency service numbers) are permanently owned by emergency services regardless of context.
| Format Range | Primary Routing | Cultural Capture Rate |
|---|---|---|
| Single digits (1-9) | Music / entertainment | High |
| Two digits (10-99) | Pop culture figures | Very high |
| Round numbers (10, 20, 100) | Institutional / listicle content | Moderate |
| Three digits (100-999) | Cultural / historical anchors | High |
| Emergency numbers | Emergency services | Permanent lock |
3.2 Percentage Format Capture
Percentage symbols provide almost zero content-type signalling. Pop culture entities override the mathematical meaning of percentages, with specific values routing to musicians or entertainers rather than mathematical or statistical content. In our testing, only one percentage value (75%) achieved clean mathematical routing without cultural collision. This suggests that the percentage symbol is effectively transparent to the AI routing system, and the underlying integer drives content selection.
3.3 Currency Format Routing
Dollar-sign queries do not lock to commercial content as might be expected. Instead, they route by magnitude through a predictable hierarchy: lower amounts route to government and consumer information, mid-range amounts to financial news and personal finance, and high amounts to luxury, entertainment, and speculative content. The currency symbol modifies but does not determine the content category.
3.4 Star Rating Differential
Star ratings near the maximum reveal a subtle processing distinction. Ratings of 4.9 trigger credentialing language (emphasis on quality verification and trust), while perfect 5.0 ratings trigger brand-naming language (emphasis on entity identity and recognition). This differential processing aligns with known consumer psychology findings about suspicion of perfection and is explored further in SIGI-2026-086.
3.5 Alphanumeric Code Resolution
Alphanumeric codes resolve to single dominant entities with minimal ambiguity. Short codes combining letters and numbers are claimed by the most culturally prominent entity using that code, with near-zero routing to alternative interpretations. This pattern suggests that mixed-format inputs trigger entity-recognition pathways rather than mathematical or informational pathways.
4. The Format Capture Model
We propose that format capture occurs through the following mechanism: AI systems trained on large corpora develop statistical associations between numeric formats and their most frequent contextual usage. When a bare numeric format is submitted without disambiguating context, the system defaults to the statistically dominant association — which is typically a cultural or entertainment entity rather than a mathematical concept. Cultural weight, measured by training-data frequency and contextual diversity, determines capture ownership.
This model generates several testable predictions: format capture should be stronger for numbers with singular dominant cultural associations, weaker for numbers with diverse associations, and vulnerable to displacement when new cultural phenomena claim a numeric value (e.g., a new entertainment franchise claiming a previously uncaptured number).
5. Limitations
These findings are based on a single AI platform tested at a single point in time. Cultural associations are temporally unstable: a number captured by a current cultural phenomenon may be released or recaptured as cultural salience shifts. The 87-query sample, while systematic, does not exhaustively cover the numeric space. Cross-platform validation has not been performed.
6. Conclusions
Format capture is a consistent phenomenon in AI search behaviour: bare numeric formats are systematically routed to culturally prominent entities rather than processed mathematically. This has implications for content strategy (numeric values in headings or content may trigger unintended routing), information retrieval research (numeric query understanding remains an unsolved challenge), and AI system design (the dominance of cultural associations over mathematical meaning in numeric processing warrants architectural attention).