This study is grounded in an analysis of 12 financial-sector companies operating in U.S. and international markets over the 2024–2025 period. The sample comprises three equal groups of four companies each:

• Traditional banks: Citibank, Wells Fargo, HSBC, Barclays.
• Neobanks and digital financial platforms: Axos Financial, SoFi Technologies, Nubank, PayPal.
• Brokerage platforms: Robinhood, IBKR, Webull, Coinbase.

A total of 31 mention peaks were identified (defined as the month with the highest volume of online mentions for a given company within the observation period). Each peak was further classified along two dimensions:

• Intensity of emotional expression (ranging from minimal to extreme) — reflecting the volume of expressive mentions relative to the company’s baseline.
• Balance of emotional expression (ranging from very strong positive to very strong negative imbalance) — the ratio of positive to negative mentions within the expressive share, excluding neutral content.

For each peak, we analyzed changes in user search behavior (transactional and informational queries) and stock price dynamics (from t−2 to t+6 months relative to the peak, including the post-earnings period). A contextual analysis of the nature of negative peaks was also conducted.

A note on sample size. The sample (31 peaks across 12 companies) is sufficient to identify pronounced behavioral patterns and formulate well-grounded hypotheses; however, it does not support claims of statistical significance for all observed relationships. The findings of this study should be treated as exploratory—substantiated preliminary conclusions subject to verification on a larger sample. This caveat applies particularly to subgroups formed through segmentation by expression balance levels, where the number of observations per subgroup is limited.

The relationship between media sentiment and financial market behavior is an actively researched area. A body of recent work confirms that news sentiment is a statistically significant, though weak, predictor of market volatility. Specifically, studies employing GARCH models on 2024 data demonstrated a statistically significant negative relationship between negative news sentiment and market stability. Aggregated news sentiment has also shown predictive power for short-term fluctuations in market returns, particularly during periods of economic or political uncertainty.

A key finding from the academic literature is that market behavior tends to be anticipatory rather than reactive: implied sentiment explains a substantially larger share of return variation than explicit textual assessments. This is important for interpreting our data: the observed effects of media mention peaks on stock prices may stem not from the mentions themselves, but from underlying events that simultaneously generate both mentions and market reactions.

Our research complements this literature by focusing not on the aggregate news flow, but on the structure of emotional content within mention peaks for specific brands—an approach that reveals subtler patterns invisible in aggregate sentiment analyses.

Search Query Segmentation

All search queries were classified into two categories:

• Transactional: queries associated with direct action—“buy,” “download/install” (app), “invest,” “open account,” “sign up,” “apply for card,” etc.
• Informational: all remaining queries reflecting curiosity, comparison, or news-seeking behavior.

For most brokerage platforms, transactional queries were almost entirely limited to “download” or “install” (mobile application).

Data Normalization

Because raw search volume changes were heavily skewed toward brokerage platforms (outliers), we applied Z-normalization within each of the three sectors and then calculated weighted averages for each intensity and balance level. Z-values exceeding 1 were treated as anomalous (noteworthy) changes in search response.

Equity Analysis

Stock price changes were measured as the percentage change between monthly closing and opening prices. The post-earnings-report period was analyzed separately.

Based on preliminary observations, the following hypotheses were formulated:

1. Sector sensitivity — brokerage platforms exhibit significantly stronger search responses to mention peaks than banks and neobanks, owing to shorter investment decision horizons and heightened risk sensitivity.

2. Funnel effect — for high-risk financial products, mention peaks first trigger a surge in informational search (curiosity), which then converts into transactional activity with a one-to-two-month lag.

3. Sentiment conflict over dominance — maximum behavioral activation (search, then purchase) occurs not under clearly positive or negative sentiment, but under a strongly mixed expression balance (see the final section of this article for a discussion of possible causes).

4. Two classes of negative peaks:

• *Formal (strong/very strong)* — driven by verifiable events (regulatory fines, data breaches, layoffs). These are difficult to reverse; reducing overall expressiveness through neutral content (“dilution”) is more effective.
• *Diffuse (moderate)* — rooted in user dissatisfaction or low-credibility signals. These evolve through shifts in the positive/negative balance but frequently revert to rising negativity at the peak.

5. Capital movement — in the financial industry, mention peaks are weakly associated with immediate stock volatility, except in cases of *moderate imbalance* (both positive and negative), which may result from companies’ attempts to mask prior-month negativity.

1. Brokerage Platforms: The Most Sensitive Category

The absolute change in search activity (transactional + informational) among brokerage platforms was 4–9 times higher than among banks and neobanks. Even low-intensity peaks produced a sustained search increase of approximately +40% above baseline for brokerages. This confirms that the brokerage sector is the most susceptible to the influence of brand mention peaks.

This finding is consistent with earlier stages of our research, where significant periodic purchases with high prices and substantial financial risk demonstrated the highest conversion of media peaks into search activity. Brokerage financial services represent a classic example of such a “high-risk” category.

[Figure 1]

2. The Curiosity-to-Action Funnel Operates Only at Moderate Expression Intensity

At moderate expressive intensity, a clear sequential pattern emerges:

• t0 (peak month)—sharp increase in informational queries.
• t+1—growth in transactional queries.
• t+2—peak of transactional activity.

No such systematic funnel was observed at low, minimal, high, or extreme expression intensities. High and extreme intensities typically produce unpredictable or absent responses.

Interpretation. This finding may be explained by several factors. First, moderate expressiveness (sufficient to capture attention but not so high as to trigger distrust) creates an optimal balance between emotional engagement and rational evaluation. Second, extreme expression in the context of financial services is perceived as a signal of deviation: for products associated with financial risk, excessive emotionality connotes manipulation or panic rather than objective information. A similar effect was observed in our earlier research on non-financial product categories, suggesting its universality.

This observation is also consistent with academic findings showing that news content is predominantly composed of neutral information, and that explicit textual sentiment possesses limited predictive power for market behavior.

3. Sentiment Conflict Matters More Than Dominance

The highest search activation (Z-score up to 1.47) was recorded in the “strongly mixed expression balance” segment—where positive and negative mentions are nearly equal, rather than when one side clearly prevails. This suggests that for financial services, ambiguity is perceived not as noise, but as a signal prompting independent investigation.

Moreover, the analysis of dynamics reveals a characteristic “layered cake” pattern: positive sentiment, apparently initiated by the company itself, is layered onto negativity from the preceding period. When either sentiment dominates (even positive), the search response is significantly weaker.

[Figure 2]

In three observed cases, the share of negativity among expressive mentions declined in the month preceding the peak (drops of 7% for IBKR, 20% for Webull, and 5% for PayPal). However, at the time of the peak, the share of negativity rebounded by 13%, 8%, and 10%, respectively—effectively making negativity more conspicuous against the positive backdrop.

Alternative explanations. Beyond the direct attempt by companies to “drown out” negativity with positive content, this dynamic may also be driven by: (a) the natural cyclicality of news flow (negativity subsides on its own before being renewed by fresh developments); (b) reactions to quarterly earnings, which may coincide with periods of “positive” PR; or (c) competitor or media activity that redirects attention to negative themes. These alternative explanations do not preclude our primary hypothesis but underscore the need for additional contextual analysis in future research.

4. Two Fundamentally Different Types of Negative Peaks

Of 31 peaks, 84% exhibited a negative imbalance (moderate to very strong). Two distinct types emerged:

• Strong and very strong negative peaks are almost invariably triggered by *formal events* (lawsuits, data breaches, layoffs, regulatory actions). Their defining characteristic is that the share of negativity in the *total volume of mentions* is minimal at the peak itself, owing to dilution by neutral content. Reversing such negative dynamics is extremely difficult; further “dilution” is the more effective strategy.
• Moderate negative peaks evolve differently: they arise from user dissatisfaction or low-credibility accusations, and their trajectory shows a decline in the share of negativity among expressive mentions in the month prior to the peak (driven by rising positive content), followed by a resurgence of negativity at the peak itself. This approach (attempting to “outweigh” negativity with positive content) proves ineffective, as negativity returns and becomes more conspicuous.

5. Capital Movement: A Weak Link with One Exception

For most segments, stock price changes following mention peaks are less pronounced than those following official earnings reports. However, moderate imbalance (both positive and negative) demonstrates a stronger immediate market reaction (t0…t+2) than the post-earnings period. An illustrative example is Nubank (November 2025): with 60% negativity and 40% positivity, shares rose 7.4% in the peak month, as the situation was perceived as an *improvement* relative to the preceding 80% negativity.

While a causal relationship between reduced negativity share and stock price appreciation cannot be established from a single observation, this example illustrates a mechanism that merits verification on a larger sample. The overall finding of a weak link between mention peaks and immediate stock volatility is consistent with the consensus in the academic literature.

When a wave of negative feedback hits a company on social media, the ideal scenario—identifying and addressing the root cause—is more often the exception than the rule. Most companies seek a quick and inexpensive solution. This gives rise to a standard toolkit of media-field manipulation tactics that are directly relevant to the “layered cake” dynamics we observe in the financial sector.

The most common method is the artificial alteration of the negative-to-positive mention ratio. The scale of this problem is substantial: by various estimates, approximately 30% of all online reviews are inauthentic, and consumer losses from purchases made on the basis of fake reviews are estimated at hundreds of billions of dollars annually at the global level. According to an NBER study (Hahn & Metcalfe, 2023), fake reviews cost consumers approximately $0.12 for every dollar spent.

Mass procurement of positive reviews remains one of the primary instruments. Despite active countermeasures by major platforms (Amazon, for example, blocked over 275 million suspicious reviews in 2024), the scale of the problem continues to grow: the volume of fake reviews is increasing 12% faster than the volume of genuine ones.

SEO “washing” — the publication of thousands of neutral or positive materials designed to push negative links out of search results — is also widely practiced. This tactic directly corresponds to the “dilution” effect we observe in the context of formal negative peaks.

The prevalence and oversight of these tactics vary significantly by region. In developed economies, where regulatory pressure is higher (the U.S. FTC adopted rules prohibiting fake and incentivized reviews in 2024), anti-manipulation efforts are institutionalized. In China, the well-organized “water army” (shuijun) industry offers services ranging from mass generation of positive reviews to coordinated attacks on competitors; the cost per post starts at fractions of a yuan. In India, organized groups on messaging platforms offer cashback in exchange for five-star reviews. In Brazil, sellers on marketplaces reportedly aim for “zero negativity,” resulting in an even more pronounced “sterile” review landscape.

Connection to our findings. The practices described above create precisely the “layered cake” of sentiments that, according to our data, serves as the most potent trigger of search activity. When artificial positivity is layered onto organic negativity, a zone of “strongly mixed expression balance” emerges—the very zone where we recorded the maximum behavioral activation. The paradox is that the attempt to “mask” negativity does not reduce but rather amplifies audience attention.

For financial companies confronting factual negativity (regulatory fines, security concerns, service outages), a *volume control* strategy—increasing the share of neutral mentions (PR-neutral events, informational materials)—is preferable to attempting to “drown out” negativity with positive content.

This is not the first study in our series to demonstrate the harm caused by the “layered cake” of mixed-sentiment reviews—a byproduct of companies’ attempts to mask organic negativity with manufactured positivity. For publicly traded companies, this practice increases the risk of price volatility and erodes investor confidence. In the financial sector, this tendency is even more pronounced.

Practical recommendations for market participants:

1. For Investor Relations teams: monitoring expression balance (rather than mention volume alone) can serve as a leading indicator of investor behavior. The “moderate imbalance” zone is a signal for heightened vigilance.
2. For PR departments of financial companies: in the case of formal negative events, a “dilution” strategy using neutral content is more effective than a positive counterattack. For diffuse (user-driven) negative peaks, a direct response to specific grievances is advisable.
3. For risk management: “moderate imbalance” in sentiment represents the zone of least predictable market reaction. This is where search activity and stock price movements are most volatile.

This study is subject to several limitations that should be taken into account when interpreting the results:

• Sample size: 31 peaks across 12 companies are sufficient to identify patterns but insufficient for statistically robust generalizations. Expanding the sample to 40–50 companies and 100+ peaks would substantially strengthen the conclusions.
• Monthly granularity: a one-month time unit may obscure intra-month dynamics, particularly for high-frequency trading decisions.
• Confounders: stock prices are determined by a multitude of factors beyond media mentions (macroeconomics, sectoral trends, quarterly earnings). Isolating the effect of mentions would require controlled studies or the application of instrumental variable models.

In subsequent research, we plan to (a) expand geographic and sectoral coverage, (b) increase granularity to weekly intervals, and (c) apply Granger causality analysis to test the directionality of observed relationships.