For two decades, banks competed for visibility through branches, brokers, marketing scale, and online search rankings. In the span of a year, the sales funnel has moved into the AI era. Generative artificial intelligence has completely disrupted the existing model, changing not just how customers search and discover but how they decide and purchase. AI assistants increasingly influence the choice by shaping perception, driving traffic, and converting demand.

Brand discovery shifts from search to answer

AI has become a key tool for customers discovering brands and products, as search shifts from link-based results to AI-generated answers. During the discovery stage, AI is embedded in search engines such as Google’s Gemini and in standalone assistants such as ChatGPT and Claude. AI assistants provide synthesized responses to customer questions, which is especially valuable when the question requires explanation, guidance, or an analysis of trade-offs. As a result, a growing number of customers form a view and short list before or without ever visiting a brand’s website.

New research by Profound for Bain & Company, analyzing prompts and queries from opt-in consumer panels in Australia, paints a detailed profile of how consumers use AI assistants.

Within banking, customers are using AI assistants to discover brands, prompting assistants to interpret, advise, and even draft communications on their behalf (see Figure 1). Roughly one-third of queries involved home loans, with customers looking to find banks that match their selection criteria (such as fast preapproval). Credit card customers often pasted in a page of fees and asked the AI assistant to convert the fine print into understandable text.

Figure 1

In these cases, the AI assistant becomes the port of entry rather than the bank’s channels (its website, app, branch, or call center). While this analysis covers Australia, a growing share of consumers in many other countries show a similar willingness to rely first on AI assistants to discover the best banking products for their needs.

Across all types of queries, social media and content sites dominated as the sources used by AI assistants. For banking queries, AI assistants generally relied not only on the banks’ own websites but on other trusted sources as well, including comparison websites (see Figure 2). Recommendations are also influenced by customer feedback found in reviews and forums, in ways that traditional search engines do not capture. Brands can no longer gain awareness and relevance simply by driving customer traffic to site; they must position themselves for inclusion in AI-generated short lists and to win in the answer.

Figure 2

Some banks have already begun to optimize their brands for AI discoverability. For example, Wells Fargo partnered with Schema App to layer structured data across its website in order to reduce AI hallucinations and improve the reliability of answers.

AI assistants have also developed perspectives on each bank that inform share of voice. For example, in transaction banking, the major banks led in share of voice, but that metric does not reflect overall market share, highlighting an opportunity for smaller banks in the realm of AI brand discovery. Take home loans: While major banks did lead in share of voice, digital-first banks such as Unloan (owned by Commonwealth Bank) and Loans.com.au (owned by Firstmac) also ranked among the top-cited brands, showing how challenger brands are shaping customers’ consideration (see Figure 3).

We also observed that global banks with a limited Australian presence ranked high in share of voice in large language model (LLM) outputs. This was most pronounced in credit cards, where US competitors including Chase and Capital One, which do not offer consumer-focused cards in Australia, had a high share of mentions. Their share derived from the large amount of globally relevant content available through trusted websites that LLMs scrape.

Figure 3

Being included in the list is not enough, however. How the LLM behind an AI assistant frames a bank’s brand will influence consideration by customers. In our sentiment analysis, positive framing of a brand correlated with repeated signals, such as clear anchor facts, structured and comparable product language, and credible third-party validation. Negative framing tended to stem from regulatory scrutiny; articles and commentary about scams, fraud, and other harms to customers; and persistent clusters of complaints about fees, disputes, or service delays. This dynamic raises the importance of nurturing customer reviews from promoters of the bank and minimizing the number of detractors.

Moving to the purchase stage

Consumers are already graduating from using AI assistants for brand discovery and comparison to purchasing with the help of AI assistants. In banking, this could take the form of a prompt like “Compare three mortgage offers I qualify for and prepare my application information for the lowest-rate lender.”

Pioneers in this space are experimenting with tools for consumers to make purchases within AI assistants. In banking and personal finance, Intuit has embedded apps—including Credit Karma, which holds credit scores, spending patterns, and loan history for many US consumers—directly into ChatGPT. Users of the Credit Karma product recommendation engine can ask ChatGPT to surface credit card, personal loan, or mortgage options tailored to their actual credit profile and approval odds, without leaving the AI assistant.

In insurance, Insurify offers a comparison and quote engine within ChatGPT to help customers create a short list and decide on the best auto policy for them. Meanwhile, Tuio and Aviva have both built OpenAI-approved apps to provide personalized home insurance quotes within ChatGPT.

Today the purchasing capability is nascent. Scaling up adoption in consumer financial services will depend both on implementing the technology with appropriate safeguards and on building consumer trust so that AI purchasing feels as secure as traditional digital banking.

The trust factor

While consumers are beginning to use AI assistants, trust remains a constraint. Caution and skepticism tend to surface when the consumer advances from advice to money movement or commitment. Bain’s research shows that only one-quarter of US consumers are comfortable using AI to complete a purchase. Customers cite data security and encryption, privacy, and AI misunderstanding their needs as the primary concerns. Brand also matters, with 36% of respondents saying they’d feel comfortable making purchases using an AI assistant from a brand they already trust for digital shopping.

To be sure, even as adoption grows, many consumers will still prefer human interactions for complex, distressing, or high-value decisions, underscoring how companies must reinforce trust in the AI era.

Practical next steps

Banking executives committed to winning in this new era can benefit from the following steps.

• Manage the narrative to control brand framing; don’t just solve for visibility. Move beyond aiming to be mentioned. Make sure AI assistants describe you favorably across selection criteria such as transparency, low friction, and credibility, rather than being defined by complaint clusters. Track how you are framed in AI answers, then fix the underlying issues and reinforce the right signals with third-party proof.
• Build segmented value propositions that make the short list. LLMs flatten generic offers into average answers and tend to reinforce repeated patterns across sources. Sharpen propositions by segment and job to be done, whether that’s compare, understand, or switch, and express them as quotable, comparable facts so you show up as “best for me,” not just “one of many.”
• Understand your role in the AI-enabled sales funnel before the profit pools move. As assistants shift from giving advice to agent-led applications and checkout, value goes to the firms that own the decision flow and execution rails. Decide whether you will be the customer’s assistant, the trusted rails partner, or the product provider; understand who will play the other roles; and redesign distribution and economics to avoid disintermediation.
• Rewrite and technically structure content for the answer layer, not the click. Turn marketing pages into natural-language modules that directly address trade-offs, scenarios, fine print, and next steps to align content to how customers ask questions. Use structured formats, such as FAQs and tables, and verified first-party facts so that AI assistants can cite you accurately and hallucinate less. Ensure your site is technically built with raw code that LLMs can easily access and read. Build a strong third-party presence so that AI will trust your brand cues.
• Put in safeguards to protect customers. Trust drops the moment advice turns into commitment or money movement, and customers will only engage with brands through AI assistants that feel secure, private, and accountable. Make primacy explicit through permissioned delegation, transparent controls, human escalation for high-stakes moments, and clear recourse when something goes wrong.

Banks no longer compete only for customer attention but also for how machines interpret their offerings and act on customers’ behalf. That compresses the distance between being considered and being chosen, shifting advantage toward brands and value propositions that are easy for AI assistants to understand and recommend. Now that AI has started to mediate between banks and their customers, banks determined to thrive in acquiring and retaining customers will need to shape that layer instead of being shaped by it.