Fintech, Brokerage and Crypto Stocks: What AI Recommendation Momentum Shows So Far

Research status: Exploratory longitudinal research. AI recommendation momentum has not been validated as a predictor of account growth, originations, trading activity, assets, crypto volumes, revenue, analyst revisions, valuation, or stock returns.

Initial observation window: July through September 2026

Fintech, brokerage and crypto research slice: 7 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 fintech, brokerage and crypto research slice shows a broad decline in AI recommendation coverage from July to September 2026, but the result should not be interpreted as a sector forecast.

Among seven mapped public-company parents, Webull was the only company with a positive aggregate recommendation change, increasing 4.4 percentage points, from 68.5% to 72.9%. Its exploratory interval crossed zero, so the row remains Mixed / neutral under the V0 rules.

The other six companies all met the V0 Negative AI divergence candidate criteria:

• Ally Financial: -5.2 percentage points
• Upstart: -7.3 points
• Happen / LendingClub: -8.2 points
• Coinbase: -9.1 points
• Chime: -9.4 points
• Goldman Sachs / Marcus: -11.3 points

The sector median recommendation change was approximately -8.2 percentage points, and the simple mean was approximately -6.6 points.

Cross-platform breadth strengthened the descriptive pattern. Happen / LendingClub and Goldman Sachs / Marcus declined on all six measured platform families. Ally, Coinbase and Chime declined on five of six. Upstart declined on four of six. Webull improved on four of six but declined on Google AI Mode and Perplexity.

These are measurements of AI recommendation behavior only. They do not establish weaker future lending volumes, lower brokerage account growth, reduced trading activity, declining crypto revenue, weaker deposits, or negative stock performance.

The prospective investor question is narrower: if unbranded AI recommendation visibility changes persist, do they precede measurable changes in branded demand, account acquisition, funded accounts, originations, trading activity, assets, transaction volume, analyst expectations, or reported revenue?

That question is part of the AI Commercial Momentum Hypothesis. The underlying measurement rules are documented in How We Measure AI Commercial Momentum, while the complete 25-company baseline is preserved in Initial Findings From 25 Public Companies.

Fintech, Brokerage and Crypto AI Recommendation Momentum at a Glance

This is a research-defined sector slice, not a formal fintech index. The companies have different revenue models, regulatory exposures, customer journeys, and relationships between the measured consumer-facing entity and the listed public parent.

Several companies also belong naturally in other sector views. Ally, Chime and Goldman Sachs / Marcus appear in the Bank Stocks and AI Search analysis. Upstart and Happen / LendingClub overlap with consumer lending. Coinbase is represented through Coinbase Wallet exposure. Webull is a brokerage platform. Those differences matter when later financial validation begins.

Questions This Section Answers

• Which fintech, brokerage and crypto companies gained or lost AI recommendation visibility?
• Which declines were broad across multiple AI systems?
• Does the negative sector median mean these companies are financially weakening?

What the Company-Level Signals Show

Webull: positive aggregate movement, but not a directional candidate

Webull increased recommendation coverage from 68.5% to 72.9%, a gain of 4.4 percentage points.

The exploratory interval ranged from approximately -1.3 to +10.2 points, so it crossed zero. Webull also improved on four of six platform families rather than meeting the stronger combination of magnitude, interval direction, and platform breadth required for a positive candidate classification.

Platform changes were:

• ChatGPT: +7.14 pp
• Gemini: +36.36 pp
• Google AI Mode: -5.08 pp
• Google AI Overviews: +11.43 pp
• Microsoft Copilot: +4.55 pp
• Perplexity: -7.41 pp

Webull is especially useful for separating recommendation coverage from simple presence. Recommendation coverage increased 4.4 points, while presence increased only about 0.5 points, from 80.8% to 81.3%.

That pattern means Webull was not simply appearing in substantially more matched responses. It was being classified as a valid recommendation in a larger share of those eligible cells.

The row remains mixed because the increase is not sufficiently directional under the current V0 thresholds.

Ally Financial: broad decline across five of six platforms

Ally Financial declined from 46.3% to 41.0% recommendation coverage, a change of -5.2 percentage points.

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

The platform changes were:

• ChatGPT: -13.64 pp
• Gemini: +4.00 pp
• Google AI Mode: -7.04 pp
• Google AI Overviews: -2.08 pp
• Microsoft Copilot: -10.53 pp
• Perplexity: -8.57 pp

Ally receives a High V0 AI-measurement confidence classification. The row has 268 matched cells, 223 prompt clusters, six measured platform families, a directional exploratory interval, and sensitivity results within the published high-confidence thresholds.

This is confidence in the measurement of the AI-side change, not confidence in a future financial outcome.

Upstart: recommendation coverage fell much more than presence

Upstart declined from 81.0% to 73.6% recommendation coverage, a change of -7.3 percentage points.

The exploratory interval was approximately -12.9 to -1.8 points, and four of six platforms declined.

Platform changes were:

• ChatGPT: +3.12 pp
• Gemini: -16.13 pp
• Google AI Mode: -14.71 pp
• Google AI Overviews: -4.55 pp
• Microsoft Copilot: +12.50 pp
• Perplexity: -27.27 pp

Upstart is another strong example of why recommendation coverage should not be substituted with mention or presence metrics. Its recommendation coverage fell 7.3 points, while presence fell only about 0.7 points, from 91.6% to 90.8%.

In other words, Upstart remained present in nearly the same share of matched responses while becoming a valid recommendation less often.

That difference is central to the framework developed in AI Recommendations vs. Mentions vs. Citations.

Happen / LendingClub: all six platforms declined, with a rebrand caveat

Happen / LendingClub declined from 13.2% to 5.0%, a change of -8.2 percentage points.

The exploratory interval remained below zero at approximately -11.0 to -5.3 points. All six platform families declined:

• ChatGPT: -44.44 pp
• Gemini: -12.50 pp
• Google AI Mode: -2.08 pp
• Google AI Overviews: -2.44 pp
• Microsoft Copilot: -6.90 pp
• Perplexity: -10.53 pp

The row also showed presence declining from 16.4% to 7.3%, a change of -9.1 points.

This is one of the most important entity-mapping caution cases in the full public-company panel. The row is a legacy LendingClub entity during the Happen rebrand transition. A name transition can affect how a model recognizes, retrieves, or recommends the entity even if underlying consumer demand is unchanged.

For that reason, later financial validation should treat this row as a rebrand-sensitive observation rather than assuming all measured movement represents a change in commercial momentum.

Coinbase: five-platform decline, measured through Coinbase Wallet exposure

Coinbase declined from 47.0% to 37.8%, a change of -9.1 percentage points.

The exploratory interval was approximately -14.5 to -3.7 points. Five of six platform families declined, while Google AI Mode increased slightly.

Platform changes were:

• ChatGPT: -5.88 pp
• Gemini: -13.89 pp
• Google AI Mode: +1.96 pp
• Google AI Overviews: -5.68 pp
• Microsoft Copilot: -5.41 pp
• Perplexity: -28.00 pp

Presence declined 8.1 points, from 58.8% to 50.7%.

The company mapping is important: the tracked entity is Coinbase Wallet, which is a product within Coinbase Global rather than a measurement of every Coinbase product or business line. The observed AI movement therefore should not be treated as a direct proxy for total Coinbase trading volume, exchange activity, subscription revenue, institutional services, or consolidated company performance.

Chime: high starting coverage, then a broad decline

Chime had the highest base recommendation coverage in this seven-company slice at 84.9%. September coverage was still high at 75.5%, but the change was -9.4 percentage points.

The exploratory interval remained below zero at approximately -17.1 to -1.7 points. Five of six platform families declined:

• ChatGPT: -22.22 pp
• Gemini: -25.00 pp
• Google AI Mode: -6.38 pp
• Google AI Overviews: -5.00 pp
• Microsoft Copilot: +7.14 pp
• Perplexity: -50.00 pp

Chime illustrates why absolute visibility and momentum should be reported separately. A company can remain highly visible while its matched recommendation coverage is declining from an even higher starting point.

That distinction matters for longitudinal research. Static level and change are different variables.

Goldman Sachs / Marcus: the largest decline in the slice, and six-platform agreement

Goldman Sachs / Marcus declined from 36.5% to 25.2%, a change of -11.3 percentage points.

The exploratory interval remained below zero at approximately -15.8 to -6.9 points. All six platform families moved negatively:

• 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

Presence also declined materially, by approximately 11.6 points, from 49.4% to 37.7%.

The parent mapping combines Goldman Sachs Group and Marcus by Goldman Sachs. This is therefore not a pure measurement of Goldman Sachs' institutional franchise, asset management business, investment banking operation, or all client segments. The measured exposure is partly consumer-facing Marcus visibility.

The row also has the largest cleaning sensitivity among the seven companies. The no-dedupe sensitivity differs from the primary estimate by about 2.0 points, while the capture-average sensitivity differs by about 2.2 points. The direction remains negative, but the measurement sensitivity should remain visible rather than being hidden.

Why This Sector Needs Business-Model Separation

Questions This Section Answers

• Why should Webull, Upstart, Coinbase and Chime not be treated as one homogeneous sector?
• What downstream outcomes should differ by company type?
• Why can the same AI signal have different financial relevance across these companies?

The seven companies in this article share digital consumer-finance exposure, but the path from an AI recommendation to a financial outcome differs substantially.

For a brokerage such as Webull, a relevant chain could be:

AI recommendation -> branded search -> app or site visit -> funded account -> assets or trading activity -> revenue

For a lending platform such as Upstart, the chain could be:

AI recommendation -> branded search -> application -> approval and funding -> originations or fee revenue

For a digital bank or banking brand such as Chime, Ally or Marcus, the chain could involve:

AI recommendation -> branded search -> account opening -> deposits, card usage, lending, or engagement -> revenue

For Coinbase Wallet exposure, the measured entity sits inside a broader crypto platform. The path could involve wallet adoption, exchange usage, asset balances, transaction activity, or other behaviors, but the mapping from wallet recommendation to Coinbase Global financial outcomes is indirect.

This is why later validation should not use one generic financial outcome for every company.

The common question is whether the AI signal contains incremental, time-ordered information. The downstream variable should be matched to the actual business model.

AI Is Already Part of Financial Research, but Trust Is Uneven

The case for testing AI recommendation visibility does not depend on assuming consumers blindly follow chatbot advice. Current research suggests the opposite: AI use in finance is increasing, while trust and final decision authority remain constrained.

Plaid's Spring 2026 State of Intelligent Finance report says 55% of consumers surveyed had used AI for money-related tasks in the prior 12 months, and 86% of AI users said it helped them better understand money. Plaid also reported that half of respondents believed managing money without AI would soon feel outdated. See Plaid, The State of Intelligent Finance.

HSBC reported in June 2026 that 73% of roughly 10,000 affluent and high-net-worth investors across 10 markets used AI for finance and investment, but only 12% said AI was the most influential factor in their most recent investment decision. The same research found human professionals and institutions remained the leading source of investment ideas. See HSBC, The Trust Threshold.

YouGov's 2026 U.S. research similarly found that among DIY investors, about 26% agreed or strongly agreed that AI is a valuable tool for creating and updating an investment strategy, while a large share remained neutral, unsure, or skeptical. See YouGov, DIY Investing in 2026.

These external findings do not validate the LLM Authority Index investor signal. They support a narrower reason to study it: AI is increasingly present in financial discovery and research, but usage, trust, and action are not the same thing.

That distinction mirrors the measurement hierarchy in this research:

AI visibility is not the same as recommendation. Recommendation is not the same as consumer action. Consumer action is not the same as company revenue. Revenue is not the same as stock return.

Cross-Platform Portability Matters in This Slice

The seven-company panel provides several examples of why aggregate recommendation change should be interpreted alongside platform breadth.

Happen / LendingClub and Goldman Sachs / Marcus are especially important because all six platform families moved in the same direction.

Webull shows the opposite type of case. Its aggregate result was positive, but Google AI Mode and Perplexity declined.

The broader framework is documented in Does Cross-Platform AI Visibility Matter?. That article treats platform breadth as a separate signal-quality dimension rather than assuming that a six-platform move is automatically more financially meaningful.

The correct future test is empirical: does multi-platform agreement improve prediction of later commercial outcomes after controlling for aggregate recommendation change itself?

Recommendation Coverage and Presence Tell Different Stories

This sector contains several useful metric-divergence examples.

This is why the research program does not collapse mentions, presence, citations, recommendation status, rank, sentiment, and downstream financial outcomes into one composite score.

For investors, that separation may eventually matter because a company that remains frequently mentioned but is recommended less often represents a different state from a company that is disappearing from responses altogether.

Whether either state predicts anything financially relevant remains unproven.

Recommendation Share vs. Market Share Is a Separate Future Test

The current sector article measures recommendation coverage, not real-world market share.

A company can be recommended frequently because the prompt set aligns strongly with its consumer proposition, because the AI systems rely on a favorable source ecosystem, because the brand is prominent in a particular use case, or for other reasons unrelated to current economic share.

The separate AI Recommendation Share vs. Market Share framework asks whether a standardized competitive recommendation share can eventually be compared with a properly matched real-world market-share denominator.

That comparison would need to be business-model specific.

For example:

• brokerage market share should not be compared with consumer-lending share;
• funded brokerage accounts are not comparable to loan originations;
• crypto wallet usage is not equivalent to exchange trading volume;
• deposit share is not equivalent to recommendation coverage;
• a consumer-facing product such as Coinbase Wallet cannot automatically stand in for the full public parent.

For that reason, this article does not publish a universal AI Share Index for the seven companies.

Investor Research Matrix

This matrix is a research agenda, not an investment recommendation.

What This Does Not Mean

The initial fintech, brokerage and crypto findings do not establish that:

• the six negative candidates will report weaker revenue;
• Webull will report stronger revenue;
• customer acquisition is already changing in the same direction as AI recommendations;
• AI systems are causing any change in consumer behavior;
• a negative AI candidate is overvalued;
• a positive or mixed company is undervalued;
• any company should be bought, sold, held, shorted, or avoided;
• the seven-company median represents the entire fintech, brokerage or crypto industries;
• one consumer-facing brand or product perfectly represents its public parent.

The labels are deliberately narrow.

A Negative AI divergence candidate means the measured recommendation change was at least -5 percentage points, the exploratory interval remained below zero, and at least four platform families worsened.

It is an AI measurement label only.

Methodology

The sector table is drawn from the V0 public-company panel documented in How We Measure AI Commercial Momentum.

Matched-panel design

For each company or public parent, the primary comparison uses the same normalized prompt on the same AI platform family in the base month and September 2026.

All seven companies in this slice use July 2026 as the base month.

Primary metric

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

Recommendation change = September recommendation coverage - base-month recommendation coverage

The change is reported in percentage points.

Denominator

Explicit not-mentioned company rows remain in the denominator when the underlying response is otherwise eligible. This prevents the analysis from considering only responses in which a company happened to appear.

Failure handling

Explicit extraction failures are excluded. A parsing failure is not treated as a legitimate zero-visibility response.

Duplicate handling

Response-identical cross-vertical duplicate exports are collapsed in the primary panel. The workbook separately reports a no-dedupe sensitivity result.

Public-parent mapping

Known consumer-facing brands, product entities, and entity variants are mapped to public parents where appropriate.

Important mappings in this article include:

• Ally Bank -> Ally Financial
• LendingClub legacy entity -> Happen, with a rebrand-transition flag
• Coinbase Wallet -> Coinbase Global
• Goldman Sachs Group + Marcus by Goldman Sachs -> Goldman Sachs Group

These mappings are analytical conveniences. They do not mean every business unit of the public parent has identical AI exposure.

Exploratory interval

The primary point estimate is cell-weighted. For uncertainty, matched-cell changes are averaged within normalized prompt clusters, and a normal 1.96 multiplier is applied around the primary estimate using the prompt-cluster standard error.

The interval is exploratory. It is not a causal confidence interval.

Candidate rules

A negative AI divergence candidate requires:

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

A positive candidate requires the corresponding positive thresholds.

Confidence labels

High confidence requires at least 200 matched cells, all six platform families, a directional exploratory interval, and limited sensitivity to the no-dedupe and capture-average alternatives.

Medium confidence requires at least 100 matched cells and at least five platforms.

These confidence labels measure the strength of the AI-side observation, not confidence in a financial prediction.

Sensitivity and Data Quality

The seven-company slice is generally stable to the current cleaning alternatives, with one notable exception.

• Webull: no material difference in the reported cleaning sensitivities.
• Ally: no-dedupe sensitivity about +1.35 pp and capture-average sensitivity about +1.52 pp relative to the primary estimate.
• Upstart: sensitivities of approximately -0.58 pp and -0.90 pp.
• Happen / LendingClub: no material difference in the reported cleaning sensitivities.
• Coinbase: both sensitivity alternatives differ by approximately -0.34 pp.
• Chime: no material difference in the reported cleaning sensitivities.
• Goldman Sachs / Marcus: no-dedupe sensitivity approximately +2.00 pp and capture-average sensitivity approximately +2.24 pp.

The Goldman row therefore deserves extra caution when discussing exact magnitude, even though the direction remains negative across all six platform families.

Limitations

1. Short observation window. The current panel covers July through September 2026 and is too short to establish persistence over multiple business cycles.
2. Research-defined sector. The seven companies do not form a standardized financial index and have substantially different business models.
3. Prompt population. The prompt set is designed for commercial relevance and comparability, not weighted by search volume, revenue contribution, account value, transaction volume, or market capitalization.
4. Entity mapping. Coinbase Wallet represents only part of Coinbase Global. Marcus represents a consumer-facing portion of Goldman Sachs. LendingClub/Happen is affected by a rebrand transition.
5. Platform changes. AI platforms can change retrieval, ranking, model, interface, or answer behavior during or after the research window.
6. Recommendation is not behavior. A model recommendation does not prove the user acted on it.
7. Behavior is not revenue. Even measurable downstream clicks or account visits may not translate into revenue.
8. Revenue is not stock return. Market expectations, valuation, interest rates, regulation, risk appetite, and other factors can dominate equity performance.
9. Crypto-specific exposure. Coinbase Wallet recommendation behavior should not be generalized to the entire crypto market or to every Coinbase revenue stream.
10. No causal identification. The current analysis is descriptive and longitudinal, not a causal experiment.

What We Will Test Next

The sector will become more useful only as later outcomes accumulate.

Potential business-model-specific validation targets include:

Brokerage

• branded search;
• app traffic;
• funded-account growth;
• assets under custody or administration where comparable;
• trading activity;
• revenue expectations.

Lending fintech

• branded search;
• application volume;
• approval and funding activity;
• loan originations;
• fee or net-interest revenue where relevant;
• analyst estimate revisions.

Digital banking

• account growth;
• deposits;
• direct-deposit adoption;
• card activity;
• engagement;
• relevant revenue lines.

Crypto

• branded search;
• wallet or app engagement;
• retail trading or transaction activity;
• assets on platform where comparable;
• revenue expectations.

The research design for these forward tests is described in Can AI Search Visibility Predict Revenue Growth?, while the statistical and investment testing architecture is described in How Investors Could Backtest AI Search Signals.

The current signals will remain frozen in the AI Investor Signal Tracker so later outcomes can be compared against what was actually measured at the time.

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

Company research

• Webull
• Ally Financial
• Upstart
• Happen / LendingClub
• Coinbase
• Chime
• Goldman Sachs / Marcus

External Context and References

1. Plaid. The State of Intelligent Finance. Spring 2026. https://plaid.com/state-of-intelligent-finance-report/
2. HSBC. The Trust Threshold: AI Makes Investors Bolder, but They Want Human Judgement to Make Decisions. June 24, 2026. https://www.hsbc.com/news-and-views/news/media-releases/2026/the-trust-threshold-ai-makes-investors-bolder-but-they-want-human-judgement-to-make-decisions
3. YouGov. DIY Investing: Why Americans Are Skipping Financial Advisors in 2026. 2026. https://yougov.com/en-us/articles/55143-diy-investing-why-americans-are-skipping-financial-advisors-in-2026

Research Disclosure

LLM Authority Index is affiliated with CiteWorks Studio and related AI-search research and services. Commercial relationships do not alter the preserved source corpus, matching rules, company mappings, candidate thresholds, or published measurements.

No company can purchase a positive investor-signal classification, ranking, recommendation, score, or inclusion outcome in this research.

Citation ≠ mention ≠ recommendation ≠ purchase ≠ revenue.