Bank Stocks and AI Search: Which Banks Are Gaining or Losing AI Recommendation Visibility?

Research status: Exploratory longitudinal research. AI recommendation momentum has not been validated as a predictor of bank revenue growth, deposit flows, loan growth, analyst revisions, valuation, or stock returns.

Initial observation window: July through September 2026

Bank and consumer-banking research slice: 6 mapped public parents

AI platform families: ChatGPT, Gemini, Google AI Mode, Google AI Overviews, Microsoft Copilot, and Perplexity

Current methodology version: V0

Answer Capsule

The initial LLM Authority Index bank and consumer-banking panel shows a wide divergence in AI recommendation momentum from July to September 2026.

Among six public-company parents included in this sector slice, Axos Financial was the only positive AI divergence candidate, with recommendation coverage increasing 19.7 percentage points, from 15.3% to 35.0%. Citigroup / Citi was mixed and statistically inconclusive at -1.8 points. Four companies met the V0 negative AI divergence candidate rules: Ally Financial at -5.2 points, Bank of America at -8.6 points, Chime at -9.4 points, and Goldman Sachs / Marcus at -11.3 points.

Cross-platform breadth also differed materially. Bank of America and Goldman Sachs / Marcus declined on all six measured AI platform families. Ally and Chime declined on five of six. Axos improved on four platforms, was flat on Gemini, and declined on Perplexity.

These findings describe changes in AI recommendation behavior, not changes in deposits, loan originations, customer acquisition, revenue, earnings, or investment value.

The central investor question is therefore not whether a bank is "winning AI search." It is whether persistent, cross-platform changes in unbranded AI recommendation visibility later correspond with measurable changes in consumer consideration, branded search, account openings, deposits, lending activity, revenue expectations, or financial results.

That relationship has not yet been established.

The theory behind the research is defined in Can AI Search Signal Future Revenue Growth? The AI Commercial Momentum Hypothesis. The measurement rules are documented in How We Measure AI Commercial Momentum, and the complete initial public-company panel is preserved in Initial Findings From 25 Public Companies.

Bank AI Recommendation Momentum at a Glance

Across these six rows, five had negative aggregate recommendation changes and one had a positive change. The median change was approximately -5.2 percentage points. The simple mean was approximately -2.8 points, but that average is heavily influenced by Axos Financial's +19.7-point increase.

This is one reason sector interpretation should not rely on a single average.

Questions This Section Answers

• Which bank-related public companies gained AI recommendation visibility?
• Which lost recommendation visibility across multiple AI platforms?
• Does a negative sector median mean bank fundamentals weakened?

Which Banks Gained or Lost AI Recommendation Visibility?

Axos Financial: the clear positive outlier

Axos Financial was the strongest positive company in this bank-related slice.

Recommendation coverage increased from 15.3% in the base month to 35.0% in September, a gain of 19.7 percentage points. The exploratory 95% interval ranged from approximately +9.8 to +29.7 points.

The movement was broad but not universal:

• ChatGPT: +14.29 pp
• Gemini: 0.00 pp
• Google AI Mode: +43.14 pp
• Google AI Overviews: +29.03 pp
• Microsoft Copilot: +18.18 pp
• Perplexity: -19.23 pp

This produced four improving platforms, one worsening platform, and one stable platform.

The current public-parent row maps Axos Bank and UFB Direct-related exposure to Axos Financial. That matters because a public-company signal can partly reflect the visibility of a consumer-facing brand or division rather than identical movement across every business operated by the parent.

The proper interpretation is therefore narrow: measured Axos-related consumer banking recommendation coverage increased materially in the matched AI prompt panel.

It does not establish stronger future deposit growth, higher loan originations, better margins, or a positive stock outcome.

Citigroup / Citi: essentially mixed

Citigroup / Citi declined only 1.8 percentage points, from 23.4% to 21.6% recommendation coverage.

The exploratory interval crossed zero widely, from approximately -8.9 to +5.3 points. Platform direction was evenly split: two improving, two worsening, and two stable.

The platform changes were:

• ChatGPT: 0.00 pp
• Gemini: 0.00 pp
• Google AI Mode: +4.76 pp
• Google AI Overviews: -5.88 pp
• Microsoft Copilot: +12.50 pp
• Perplexity: -25.00 pp

Citi is therefore a useful reminder that an aggregate sector table should not force every company into a directional narrative. The V0 result is mixed.

Ally Financial: negative and broad

Ally Financial declined 5.2 percentage points, from 46.3% to 41.0% recommendation coverage.

Its exploratory interval remained slightly below zero, from approximately -10.2 to -0.2 points, and five of six platforms moved negatively. Gemini was the only positive platform at +4.0 points.

The row receives a High AI-measurement confidence classification under V0 because it has 268 matched cells, six platform families, a directional interval, and cleaning sensitivities within the published high-confidence thresholds.

That is confidence in the measurement of the AI-side change, not confidence in a financial forecast.

Bank of America: six-platform decline

Bank of America declined 8.6 percentage points, from 35.3% to 26.7%.

The exploratory interval ranged from approximately -12.8 to -4.4 points. All six platform families moved negatively:

• ChatGPT: -16.13 pp
• Gemini: -13.73 pp
• Google AI Mode: -1.03 pp
• Google AI Overviews: -9.04 pp
• Microsoft Copilot: -1.64 pp
• Perplexity: -25.00 pp

This is one of the strongest examples of cross-platform directional portability in the current panel. As discussed in Does Cross-Platform AI Visibility Matter?, platform breadth should be measured separately from aggregate signal magnitude.

Bank of America's six-platform decline tells us the measured direction was broadly distributed across the AI systems in this panel. It does not tell us whether future commercial or financial results will move in the same direction.

Chime: high starting coverage, lower September coverage

Chime began with the highest recommendation coverage of the six-company slice at 84.9%, then declined to 75.5%, a change of -9.4 percentage points.

Five of six platforms declined. Microsoft Copilot increased by 7.14 points, while the other platform changes were negative, including -50.0 points on Perplexity.

Chime is important analytically because its high absolute coverage and negative momentum can coexist. A company can remain highly visible while losing recommendation coverage at the margin.

That distinction is developed in AI Recommendations vs. Mentions vs. Citations: absolute visibility level and change in recommendation coverage answer different questions.

Chime also overlaps naturally with the broader Fintech, Brokerage and Crypto Stocks research branch. Sector boundaries in AI-mediated consumer finance do not perfectly match formal equity-industry classifications.

Goldman Sachs / Marcus: largest decline in the bank slice

Goldman Sachs / Marcus had the largest negative aggregate change in this six-company slice.

Recommendation coverage declined from 36.5% to 25.2%, a change of -11.3 percentage points. The exploratory interval ranged from approximately -15.8 to -6.9 points.

All six measured platforms declined:

• ChatGPT: -21.21 pp
• Gemini: -32.26 pp
• Google AI Mode: -12.20 pp
• Google AI Overviews: -1.69 pp
• Microsoft Copilot: -15.00 pp
• Perplexity: -11.76 pp

The mapped row combines Goldman Sachs Group with Marcus by Goldman Sachs, so consumer-facing recommendation behavior should not be treated as a complete measurement of the economics of the entire parent company.

The row also illustrates why sensitivity analysis matters. The primary direction remains negative, but the no-dedupe and capture-average sensitivities differ from the primary estimate by roughly 2.0 and 2.2 percentage points, respectively. That is one reason the V0 confidence label is Medium rather than High despite a large sample and all-platform directional movement.

Why AI Recommendation Visibility Could Matter More in Consumer Banking

Questions This Section Answers

• Are consumers actually using generative AI for financial questions?
• Why might unbranded AI recommendations affect bank consideration sets?
• Does AI use imply that consumers will act automatically on recommendations?

AI-assisted financial research is already large enough to justify measuring the channel, but current evidence also argues against assuming that AI recommendations automatically translate into account openings or financial actions.

PwC's Consumer Lending Radar 2026 reports that 45% of surveyed consumers used a generative AI tool for a financial question in the prior year, and nearly a third of borrowers used AI tools to research loans. PwC also reports that half of borrowers said AI had helped them make a borrowing decision, while 67% expected AI to inform their next borrowing decision. See PwC Consumer Lending Radar 2026.

Deloitte's August 2026 survey of almost 2,600 U.S. banking customers likewise found meaningful use of generative AI to research and compare financial options, but substantial trust limits. Deloitte reported that 72% were concerned about sharing financial information with generative AI tools and 64% worried about hidden bias. See Deloitte, How Much Do Bank Customers Trust Gen AI?.

Those findings support a careful commercial hypothesis:

AI can influence financial discovery and comparison before it becomes an autonomous financial agent.

That is exactly the stage where unbranded recommendation visibility may matter.

A prospective customer can ask:

• Which banks have the best high-yield savings accounts?
• Which banks are best for low fees?
• What are the best online banks for direct deposit?
• Which banks are best for a first checking account?
• Which banks have good mobile apps?
• Which lenders are good for a particular borrowing profile?

An AI system may help define the consumer's consideration set before the consumer ever reaches a bank's website, comparison page, app store listing, or branch.

But the current research does not assume a direct one-to-one pathway from AI recommendation to financial outcome.

The likely chain is probabilistic:

unbranded financial question -> AI-generated recommendation set -> consumer consideration -> brand search, website visit, app visit, or comparison -> account application or borrowing action -> funded account, deposit, loan, interchange, fee, or other economic activity -> revenue and earnings impact

Every arrow in that chain requires validation.

Cross-Platform Bank Recommendation Patterns

The six-company slice shows why investors should inspect platform breadth instead of relying only on a blended AI visibility score.

The table shows at least three distinct patterns.

1. Broad agreement

Bank of America and Goldman Sachs / Marcus declined on all six platforms. Their aggregate direction is highly portable across the measured systems.

2. Broad but imperfect agreement

Ally and Chime declined on five of six platforms. Axos improved on four, was flat on one, and declined on one.

3. Fragmented movement

Citi's platform results were mixed, with two positive, two negative, and two flat platform changes.

The investment relevance of these patterns is unknown. The forward test is whether broad, persistent platform movement contains more information about later commercial outcomes than a change concentrated in one platform.

That is one of the explicit hypotheses in the cross-platform portability framework.

Recommendation Coverage Is Not Market Share

The bank sector is especially vulnerable to a misleading interpretation of AI recommendation share.

A company can be recommended frequently without having a large share of deposits, loans, cards, assets, or customers. A large incumbent can have enormous real-world market share while receiving only moderate recommendation visibility for a narrow prompt set. A digital brand can over-index in AI answers while representing a smaller portion of total industry economics.

That is why AI Recommendation Share vs. Market Share treats the relationship as a future validation problem rather than a direct conversion formula.

For banks, even the phrase market share requires a defined denominator.

Possible denominators include:

• domestic deposits;
• checking accounts;
• savings balances;
• credit-card purchase volume;
• personal-loan originations;
• mortgage originations;
• auto loans;
• brokerage assets;
• customer count;
• or another economically matched category.

Those denominators are not interchangeable.

A recommendation-share gap becomes meaningful only when the AI prompt universe and economic market-share denominator describe the same commercial market.

Why Parent-Company Mapping Matters in Banking

Banking companies often operate multiple consumer brands, products, charters, digital units, lending businesses, and nonconsumer segments.

The V0 public-company mapping therefore requires interpretation at the entity level.

• Axos Financial includes Axos Bank and UFB Direct-related exposure in the tracked mapping.
• Ally Financial is represented through Ally Bank consumer visibility.
• Goldman Sachs includes Marcus in the mapped consumer-banking exposure.
• Citi merges Citi and Citigroup entity variants.
• Bank of America merges direct entity variants for the parent brand.
• Chime is treated as a direct consumer-facing company row, while also fitting naturally into the broader fintech research branch.

A change in one consumer-facing entity should not be assumed to describe every business segment of the public parent.

This becomes especially important for diversified institutions such as Goldman Sachs and Citigroup, where consumer AI prompts can represent only one part of consolidated economics.

Future financial validation should therefore consider economic exposure weighting. A recommendation signal tied to a narrow consumer product should receive less parent-company weight than a signal tied to a business line responsible for a large share of revenue or earnings.

Presence, Recommendation, and Rank Can Tell Different Stories

The bank slice also reinforces the measurement rule established in AI Recommendations vs. Mentions vs. Citations.

Recommendation coverage is not the same as simple presence.

For example:

• Axos Financial recommendation coverage increased 19.7 points, while presence increased 20.4 points.
• Ally recommendation coverage declined 5.2 points, while presence declined 4.5 points.
• Bank of America recommendation coverage declined 8.6 points, while presence declined 6.1 points.
• Goldman Sachs / Marcus recommendation coverage declined 11.3 points, while presence declined 11.6 points.
• Citi recommendation coverage declined only 1.8 points, while presence declined 0.9 points.

The direction is often similar in this particular subset, but the measures are still conceptually different.

Rank adds another dimension. A company can be recommended less often but move higher within the smaller set of answers where it still appears. That is why recommendation frequency, rank, presence, citation behavior, sentiment, platform breadth, and persistence should remain separate variables until empirical testing shows how they relate to outcomes.

What This Means for Investors and Analysts Today

The bank-sector data is best used as a research watchlist, not a portfolio signal.

A disciplined analyst could use the current observations to ask more specific follow-up questions.

The AI Investor Signal Tracker will preserve the original observations as later months accumulate.

What This Does Not Mean

The initial bank-sector findings do not establish that:

• Axos Financial will grow faster than the other companies in this table;
• Bank of America, Ally, Chime, or Goldman Sachs will report weaker financial results;
• Citi is financially neutral;
• recommendation coverage causes account openings or deposits;
• AI platforms represent the full population of bank customers;
• recommendation momentum equals deposit market share;
• all six companies have comparable business models or economic exposure;
• a six-platform decline is automatically more financially important than a four-platform decline;
• AI recommendation movement predicts stock performance;
• any company is undervalued, overvalued, attractive, unattractive, or mispriced.

Those questions require separate financial evidence.

The prospective test is defined in Can AI Search Visibility Predict Revenue Growth?, while the implementation rules are detailed in How Investors Could Backtest AI Search Signals.

Methodology for the Bank Sector Slice

The bank-sector table is derived from the same V0 public-company methodology used in the broader 25-company panel.

Eligible observations

The analysis compares eligible matched prompt-platform cells for each mapped public parent. Explicit extraction failures are excluded rather than treated as zero recommendation observations.

Matched-panel design

The primary change compares the same normalized prompt on the same AI platform family between the base month and September 2026.

For the six companies in this article, the base month is July 2026.

Recommendation coverage

Recommendation coverage = valid recommendation cells / eligible matched prompt-platform cells

The article reports September coverage minus base-month coverage in percentage points.

Public-parent mapping

Known consumer brands and entity variants are mapped to the public parent. A parent is counted as recommended in a prompt-platform cell when any tracked entity mapped to that parent receives a valid recommendation under the V0 rules.

This improves company-level comparability but does not eliminate economic-exposure limitations.

Duplicate handling

Response-identical cross-vertical duplicate exports are collapsed in the primary cleaned panel. The workbook retains no-dedupe sensitivity calculations.

Repeated-response sensitivity

When multiple distinct response signatures exist for the same prompt-platform-month cell, a capture-average sensitivity model tests whether the result depends materially on consolidation choices.

Exploratory interval

The primary point estimate is cell-weighted. Uncertainty is calculated after averaging matched-cell changes within normalized prompts, then applying a normal 1.96 multiplier around the primary point estimate.

This is an exploratory uncertainty interval, not a causal confidence interval.

V0 candidate rules

A positive AI divergence candidate requires:

• recommendation change of at least +5 percentage points;
• exploratory 95% interval lower bound above zero;
• at least four improving platform families.

A negative AI divergence candidate requires:

• recommendation change of at least -5 percentage points;
• exploratory 95% interval upper bound below zero;
• at least four worsening platform families.

Everything else is mixed or neutral.

These labels identify unusual AI recommendation movement for financial follow-up. They are not investment recommendations.

Limitations

This sector analysis has several important limitations.

1. The sector slice is research-defined, not a formal equity index. It includes six public parents with meaningful consumer-banking exposure in the V0 panel, including companies that may also fit fintech or diversified financial classifications.
2. The observation window is short. July through September 2026 is enough to establish an initial direction, not enough to establish long-term persistence.
3. Prompt populations are research panels. They are not weighted by search volume, deposits, customer value, account openings, or revenue.
4. Entity exposure varies. Marcus, Ally Bank, UFB Direct-related exposure, and other brands can represent only part of a listed parent's economics.
5. Platform deltas can be volatile. Some platform-level matched panels are much smaller than the aggregate company panel.
6. AI recommendation is not consumer action. A recommendation can influence consideration without producing a website visit, account application, funded account, deposit, loan, or transaction.
7. The sector is heterogeneous. Retail banking, digital banking, diversified financial services, consumer lending, cards, and investment banking have different economic models.
8. Market share is not yet integrated. Deposit, loan, card, and customer-share denominators need to be standardized before recommendation-share gaps can be interpreted responsibly.
9. Financial outcomes remain untested. The current article does not establish predictive value for revenue, earnings, analyst revisions, valuation, or returns.
10. AI behavior can change. Model updates, retrieval changes, ranking changes, and platform product changes can alter recommendation patterns after the observation window.

What We Will Test Next

The bank sector provides a useful environment for the broader AI Commercial Momentum Hypothesis because several intermediate commercial variables can be measured before annual financial results are known.

Future validation should test whether bank recommendation momentum precedes changes in variables such as:

• branded search demand;
• direct website and app engagement;
• account-opening or application indicators where observable;
• deposit growth;
• consumer loan originations;
• card acquisition or spending indicators where relevant;
• customer growth;
• analyst revenue-estimate revisions;
• reported revenue growth;
• revenue and EPS surprise;
• and, only later, sector-relative stock returns.

The key comparison is not whether one bank had a higher recommendation rate than another.

It is whether within-company changes in AI recommendation behavior contain incremental information about what happens next after controlling for what investors already knew at the signal date.

That is the standard set in the prospective validation framework and the backtesting protocol.

Related LLM Authority Index Research

Core framework

• Can AI Search Signal Future Revenue Growth? The AI Commercial Momentum Hypothesis
• Initial Findings From 25 Public Companies
• How We Measure AI Commercial Momentum
• AI Investor Signal Tracker
• AI Recommendation Share vs. Market Share
• Cross-Platform AI Visibility and Recommendation Portability
• How Investors Could Backtest AI Search Signals

Bank and consumer-banking company research

• Axos Financial
• Citigroup / Citi
• Ally Financial
• Bank of America
• Chime
• Goldman Sachs / Marcus

External Research Context

The external studies below support the relevance of AI in financial research and comparison. They do not validate the LLM Authority Index investor signal.

1. PwC. Consumer Lending Radar Survey 2026: AI ambition meets consumer lending reality. https://www.pwc.com/us/en/industries/financial-services/banking-capital-markets/consumer-finance/consumer-lending-radar.html
2. Deloitte Center for Financial Services. How much do bank customers trust gen AI? August 27, 2026. https://www.deloitte.com/us/en/insights/industry/financial-services/bank-customers-generative-ai-trust.html

Bottom Line

The first bank-sector AI recommendation panel is not a uniform story.

Axos Financial moved sharply higher. Citi was mixed. Ally, Bank of America, Chime, and Goldman Sachs / Marcus moved lower enough to meet the V0 negative-candidate rules, with Bank of America and Goldman Sachs / Marcus declining across all six measured platform families.

The most important finding is not that one side of the table is "better" than the other.

It is that commercially relevant AI recommendation behavior is measurable at the company level, can move materially over short periods, and can differ sharply across companies that compete for similar consumer financial decisions.

The next question is the one that matters for investors:

Do those changes contain information about future consumer behavior and financial outcomes that was not already visible in traditional data?

The series is designed to answer that question prospectively, not retrospectively.