Ask an AI assistant which bank a newcomer to Canada should open an account with, and you will get a confident answer. Ask it which firm a mid-sized pension plan should trust with its fixed income mandate, and you will get a confident answer too. The difference is that the first answer is nearly identical across every major AI model, and the second one changes depending on which model you ask.
That gap is the defining feature of how AI evaluates financial services today. We analysed the most recent edition of fourteen financial services categories tracked by QuadrantX, from retail banking and insurance to ETFs, asset management, and M&A advisory. Each category is scored by five production AI models: Claude Sonnet 4.6, GPT-5, Gemini 2.5 Flash, DeepSeek, and Perplexity's Sonar Pro. What emerged is not one financial services industry but two, evaluated in fundamentally different ways.
Where AI Agrees: Consumer Banking and Insurance
In the categories closest to everyday consumers, AI models speak with one voice. All five models named Royal Bank of Canada as the leading retail bank in Canada. All five named State Farm for home and auto insurance in Texas, and all five named Kaiser Permanente for health insurance in California. In each of these categories, the leader scored a Narrative Dominance of 100 and a Sentiment above 90, and the same firm has held the top position in every edition of the report we have generated this year.
The agreement extends beyond the top spot. In Texas home and auto insurance, four out of every five vendors that appear in one model's top five also appear in the others'. State Farm, Allstate, Progressive, USAA, and GEICO form a shortlist that every model reproduces with only minor reordering. Manchester's home and auto insurance market looks similar: Aviva leads for four of the five models, with Admiral, Direct Line, and LV= close behind.
Model Consensus — The degree to which independent AI models, asked the same category question, produce the same shortlist and the same leader. High consensus means a buyer will hear the same recommendation regardless of which assistant they use. Low consensus means the recommendation is effectively random with respect to the buyer's choice of tool.
Why the unanimity? These brands have enormous, consistent footprints in the public text that AI models learn from: decades of press coverage, regulatory filings, comparison sites, consumer reviews, and advertising. Every model has seen the same story told the same way thousands of times. There is nothing to disagree about.
Where AI Fragments: Advisory, Asset Management, and Capital Markets
Move up the value chain toward institutional and advisory services, and the picture changes completely.
In the investment advisory firms category, the five models named five different leaders: Vanguard Personal Advisor, UBS Global Wealth Management, Merrill Lynch Wealth Management, BlackRock, and Vanguard's broader brand. Across the five models' top-five lists, eighteen distinct firms appeared. On average, any two models shared only about one firm in ten. Nearly half of all firms mentioned in the category were mentioned by a single model and no other.
Canadian fixed income specialists showed the same fragmentation. RBC Capital Markets led for two models, while the others named PIMCO Canada, Fiera Capital, and Franklin Templeton. Seventeen distinct firms filled the combined top-five lists, and 63% of all firms mentioned came from a single model. Canadian asset management split three ways at the top, between RBC Global Asset Management, BlackRock Canada, and Sun Life.
Even where a leader is clear, the field beneath it can be chaotic. Four of five models agreed that Canaccord Genuity leads Canadian small-cap M&A advisory, but their top-five lists shared almost nothing else: sixteen distinct firms across five lists. Software M&A advisory looked the same, with Goldman Sachs on top for four models and a scattered field of boutiques and bulge-bracket banks below.
In consumer finance, AI has memorised the answer. In institutional finance, AI is improvising, and every model improvises differently.
The numbers behind the split
Across all fourteen financial services categories, all five models named the same leader in about three categories out of ten. At least four of five agreed in roughly two-thirds. But in nearly three categories out of ten, the five models' top-five lists had no firm in common at all. On average, a single financial services category produced eleven distinct top-five firms across five models. If every model agreed, that number would be five.
Consensus also tracks stability over time. The categories where models agree today are the same categories whose leader has not changed all year. Broad market index ETFs have named Vanguard first in every edition we have generated. Investment advisory, by contrast, has changed its top firm in roughly half of consecutive editions, cycling between Vanguard and Morgan Stanley Wealth Management with several others taking turns.
Three Patterns Financial Services Leaders Should Understand
1. Retrieval-based models see a different market
Perplexity's Sonar Pro is the one model in our production set that searches the live web before answering. Across all categories we track, it is also the model most likely to name a leader that no other model names. In financial services, its picks were consistently the freshest and the least conventional: Sun Life for Canadian asset management, Ventum Financial for small-cap M&A advisory, Franklin Templeton for Canadian fixed income. When a firm has recent, well-indexed content about a specific mandate, retrieval-based models notice. Training-data models, which learn from a snapshot of the web taken months or years earlier, do not.
This is why single-model measurement is dangerous in financial services. A firm that checks only ChatGPT will see one market. A firm that checks only Perplexity will see another.
2. AI splits your brand into sub-brands
Large financial institutions do not appear in AI answers as a single entity. They appear as a family. Our Canadian categories contain RBC, RBC Global Asset Management, and RBC Capital Markets as separate recommendations, each with its own score. BlackRock appears alongside iShares and the iShares Bitcoin Trust. Vanguard appears alongside Vanguard Personal Advisor Services and individual Vanguard funds.
For a marketing team, this has a practical consequence. The question is not just "does AI recommend us?" but "which of our brands does AI recommend, for which need, and does it connect them to each other?" A bank whose capital markets arm is invisible to AI while its retail brand is dominant has a discoverability problem that a single brand-level score will hide.
3. Sentiment is not the differentiator. Presence is.
Among financial services firms that AI does recommend, sentiment is uniformly positive. Leaders in these categories typically score between 85 and 95 on Sentiment; models describe banks, insurers, and asset managers in careful, respectful terms. The variation that matters is in Narrative Dominance: whether a firm is mentioned at all, how early, and how often. In the fragmented categories, dozens of well-regarded firms are simply absent from most models' answers. They are not being criticised. They are being ignored.
If your firm competes in a consumer-facing category, your AI visibility is largely a function of brand scale, and the shortlist is difficult to break into. If your firm competes in advisory, institutional, or specialist categories, the shortlist is unsettled, models disagree, and specific, well-structured content about specific mandates can move you onto it. The second situation is harder to measure and far easier to change.
What This Means for Compliance and Reputation
Financial services is a regulated industry, and AI recommendations carry risks that a software category does not. When a model names a specific ETF or a specific advisory firm in response to a question about someone's retirement savings, it is producing something adjacent to advice. Different models handle this differently: some hedge heavily, some name products directly, and retrieval-based models may surface a fund launched last quarter alongside one that closed last year.
For firms, this creates two obligations. The first is to know what AI says about you, across models and over time, because a hallucinated fee, a misattributed product, or an outdated regulatory status can reach clients before your compliance team ever sees it. The second is to make the authoritative version of your story easy for models to find and hard to get wrong: consistent entity descriptions, clear product naming, and structured, citable content on the pages models actually retrieve.
How to Read Your Own Category
The QuadrantX Explore page lets you view any of our financial services categories and see where each firm sits on Narrative Dominance and Sentiment. When you look at your own category, ask three questions:
- Is the leader unanimous? If every model agrees on the top firm, the category is memorised and the shortlist is sticky. Your goal is to be securely in the top five, not to displace the leader.
- How many distinct firms appear across models? A category with fifteen or more distinct top-five firms is fragmented. Models are guessing, and the firm that gives them the clearest, most consistent signal will win the guess.
- Which model is the outlier? If the web-retrieval model names firms the others do not, recent content is being rewarded. If the training-data models agree and the retrieval model diverges, the market narrative may be shifting faster than the models' memory.
For a step-by-step approach to running this audit yourself, see our guide to measuring whether AI recommends your brand.
The Bottom Line
Financial services buyers, from retail customers to institutional allocators, increasingly ask AI before they ask anyone else. In the consumer half of the industry, AI already has a settled answer. In the institutional and advisory half, it does not, and the firms that understand this have a window that will not stay open. Consensus, once it forms, is very hard to break.