When ChatGPT picks the bank. SEO for banks in AI answers

The short version

• AI already helps people pick a bank. In McKinsey’s 2025 survey, 23% of consumers used gen AI for money tasks at least monthly, and 33% of those used it to compare providers.
• The answer is built mostly from other sites. Visa’s consultants estimate that bank and brand websites supply about 12% of the sources Google’s AI Overviews cite.
• Fewer people click. In Pew’s tracking of 900 US adults, people clicked a search result on 8% of visits when an AI summary showed, against 15% without one.
• Classic SEO is still the way in. A page that cannot be crawled never enters the pool the model writes from. An average AI Overview cites 6 to 14 sources, per Visa.
• Pages can be tuned for answers. The original GEO paper found that adding sources, quotes and statistics raised a page’s visibility in AI answers by up to 40%.
• Agents come next. McKinsey thinks an AI agent could become the main banking interface within 3 to 5 years.

1. The model reads other people’s pages before it reads yours

When someone asks ChatGPT or Google which savings account to open, the model does not start from your product page. It starts from whatever it trusts on the topic.

Visa Consulting and Analytics put rough numbers on this for Google’s AI Overviews. An average overview cites 6 to 14 sources. Brand and institutional websites make up only about 12% of those citations. The rest, by Visa’s directional estimate: about 42% from Google’s own properties such as YouTube and Maps, about 33% from editorial and news publishers, and about 13% from forums such as Reddit and Quora.

Pew Research Center saw the same pattern from the user’s side. In its sample, Wikipedia, YouTube and Reddit alone made up 15% of the sources listed in Google’s AI summaries.

McKinsey’s banking team adds two points that should worry a bank more. First, in its tests, AI models were much more likely than a web search to suggest digital lenders, fintechs and credit unions for a query like “best savings account”. Second, its Global Banking Annual Review warns that banks can no longer count on affiliates such as NerdWallet and Credit Karma to send customers their way, because the AI answers the question before anyone clicks through.

BCG and FT Partners describe the same shift from the fintech side. In a search-led world, big media budgets and keyword bidding carried a lot of weight. In a world where models shape the shortlist, the advantage moves to product clarity and credibility. BCG thinks that may favor smaller fintechs over large banks.

We see the same thing in our own data. We track 675 prompts about our own agency across ChatGPT, Perplexity and Google’s AI answers. In September 2026, ChatGPT opened our website in about 2% of its answers, against about 34% on Perplexity and Google. For startup and fintech questions, the engines mostly read other agencies’ “top 10” lists, not our pages.

What to do in 2027. Treat the sources the model trusts as your real storefront.

• Run a citation audit. Visa’s playbook is simple: ask your target questions in Google AI Overviews, Perplexity and ChatGPT, then note which competitors get cited, for which questions, and from which pages.
• Earn third-party mentions on purpose. McKinsey calls this “sourcing”: credible media coverage, reviews (it names Reddit, social media and LinkedIn), and research citations that a model can find and repeat.
• Make your facts match everywhere. Visa notes that institutions whose names, addresses and product details agree across trusted sources are more likely to be picked as a source.

2. Ranking first is worth less when people stop clicking

Pew tracked the browsing of 900 US adults in March 2025, across 68,879 Google searches. 58% of them ran at least one search that produced an AI summary. When a summary appeared, people clicked a regular search result on 8% of visits. Without one, they clicked on 15%. They clicked a link inside the summary itself on just 1% of visits. And they ended their browsing session on 26% of pages with a summary, against 16% without.

Source: Pew Research Center, “Google users are less likely to click on links when an AI summary appears in the results”, July 2025.

Visa’s report collects similar numbers from the SEO industry. Ahrefs, in February 2026, measured a 58% average drop in clicks on the first result when an AI Overview shows. Seer Interactive found about 65% for informational queries.

Google reads the same trend differently. In its July 2026 earnings remarks, Sundar Pichai said AI Mode had passed 1 billion monthly users and that AI features in Search send billions of clicks to websites every week.

The two claims fit together. Total clicks can stay large while the share going to any one bank page shrinks.

Meanwhile the AI-first tools keep growing. OpenAI says ChatGPT reaches 1.2 billion people each week, as of October 2026. BCG’s fintech report cites data that 77% of US ChatGPT users treat it as a search engine, and that 58% of US shoppers recently used an answer engine while researching a purchase.

The visits that do come through look better. Visa cites HubSpot and SimilarWeb data that AI-referred traffic converts at 4% to 8%, against 2% to 4% for regular organic traffic. BCG reports that in July 2025, traffic from AI tools to US retail sites rose 4,700% year over year, and those visitors spent 32% more time on site and bounced 27% less. Retail is not fintech, as BCG notes, but the direction is the same.

What to do in 2027. Change what you measure.

• Add “citation inclusion” to the SEO dashboard: how often your bank is named or cited for your priority questions, engine by engine. Visa lists it next to rankings and traffic as a core metric.
• Track AI-referred sessions and their conversion rate apart from other organic traffic.
• Stop judging content by visits alone. A page can win the decision without ever being visited.

3. Classic SEO is still the ticket in

It would be easy to read all this and decide SEO is over. The reports say the opposite.

Visa’s team explains the mechanism. Most AI search tools retrieve pages first and write the answer second. Traditional SEO decides whether your page makes it into that retrieval pool at all. If a page cannot be crawled or indexed, no amount of clever wording helps. Google’s AI Overviews lean mostly on Google’s own index and authority signals. Perplexity leans more on live web search and on fresh, well-organized pages.

So Visa proposes two layers. The foundation is the SEO a bank should already have: technical health, site speed, mobile, content that covers the topic in depth, links from credible sites, and clean internal linking. The second layer, which Visa calls answer engine optimization, adds:

• direct answers to questions people actually ask,
• structured data (schema markup) for the organization, products, FAQs, articles and breadcrumbs, in JSON-LD,
• a short, clear summary paragraph at the top of each important page,
• consistent brand, product and author names,
• third-party coverage and expert attribution.

McKinsey’s technical advice is close to Visa’s. Pages should be fast, well structured and unambiguous, with consistent product names, a clean page hierarchy and machine-readable feature data where possible. It also reminds banks that each model behaves a little differently, so the work has to cover several of them.

What to do in 2027. Fix the plumbing before the copy.

• Audit crawling, indexing and Core Web Vitals on every product page first.
• Add schema to the page types that matter, and check it with Google’s Rich Results Test, as Visa suggests.
• Pick one name for each product and use it everywhere: site, app, press, rate tables.

4. Pages that answer with sources and numbers get picked

The term “generative engine optimization” comes from a 2024 research paper by Pranjal Aggarwal and colleagues at Princeton and IIT Delhi, presented at KDD, a major data mining conference. It is still the best evidence we have on what moves a page into an AI answer.

The team built a test set of 10,000 queries and rewrote source pages in different ways. Their methods raised a page’s visibility in AI answers by up to 40%. Three changes worked best: citing sources, adding quotations from relevant experts, and adding statistics in place of vague claims. Each gave a relative gain of roughly 30% to 40% on the paper’s main measure. Keyword stuffing, the old SEO habit, did poorly. On Perplexity, a live engine, the gains reached 37%.

One finding matters for any bank that does not rank first. Citing sources lifted the visibility of the fifth-ranked page by 115.1%, while the top-ranked page lost 30.3% on average. In the paper’s tests, a lower-ranked page with better evidence could take ground from the leader.

Two caveats. This was a lab benchmark built on 2023 and 2024 engines, and the engines have changed a lot since. And a bank cannot invent statistics to look credible. It can, however, put its real rates, fees and eligibility rules on the page as numbers.

The consulting reports agree. McKinsey’s advice for banks is decision-making content, not marketing: plain-language product explanations, transparent pricing and eligibility, and scenario guides that answer a real question, such as how a frequent traveler should handle foreign exchange. Its line is that quality beats volume. Visa adds that AI systems favor original, experience-based content, which matters in banking, where product pages often read the same across institutions.

The results claimed so far are early but large. McKinsey reports that one financial institution saw sixfold growth in monthly organic traffic after it started GEO work. BCG says companies with GEO see about 7 times higher AI visibility. Neither firm names the companies or the method, so I read both as a direction, not a benchmark.

What to do in 2027. Rewrite the product pages that matter most.

• Open each product page with a two- or three-sentence answer: who it is for, what it costs, how to qualify.
• Put rates, fees and eligibility in plain text on the page. A PDF or an image is not enough.
• Write headings as the questions customers ask, as Visa suggests.
• Back claims with sources and real numbers. That is what the GEO paper measured, and it is also simply good banking content.

(This is where design meets SEO. A clear product page that a person can scan is usually also a page a model can read.)

5. A wrong answer is a compliance problem

The answer also has to be right. Visa’s report spends a full section on this. Google treats financial content as “Your Money or Your Life” and applies its strictest quality standards to it. When an AI answer quotes an old rate, the wrong eligibility rule or a product that no longer exists, the bank can lose trust and attract regulators. Visa notes that supervisors in key markets are paying closer attention to AI-made misstatements of product terms. Outdated rate data, vague eligibility criteria and terms that differ between channels all raise the risk.

McKinsey gives banks four questions to ask of every AI answer about them:

1. Presence and prominence. Are we in the answer, and do we look like a real option?
2. Sentiment. Is the tone confident, or hedged and doubtful?
3. Accuracy and relevance. Are our products described correctly and matched to the right customer?
4. Depth. Does the answer help someone decide, or stay on the surface?

What to do in 2027. Give accuracy an owner.

• Name one person accountable for how the bank appears in AI answers, with compliance in the loop.
• Check your top questions across the main engines every month, using McKinsey’s four questions.
• Make sure every rate change reaches the website, the app and partner pages the same day. Visa calls this content accuracy governance.

Where the reports disagree

How big is AI discovery today? Visa puts AI-native platforms at only about 1% to 3% of discovery interactions. It also cites analysts, Gartner among them, who expect traditional search traffic to fall by up to 25%. Against that, OpenAI reports 1.2 billion weekly ChatGPT users, and Google reports more than 1 billion monthly users for AI Mode inside Search. My reading: AI tools still handle a small share of searches, yet McKinsey finds that 51% of consumers already use gen AI.

Who wins: fintechs or banks? McKinsey’s tests suggest models favor fintechs, digital lenders and credit unions. BCG argues that trust and regulation give fintech more protection than retail or travel, and that the B2C fintechs most exposed are the ones that live on paid search and comparison sites. The two views meet at one weak spot: acquisition that depends on comparison sites.

Do clicks fall? Pew says yes, sharply. Google says AI features still send billions of clicks a week. The two numbers measure different things, and neither tells you about your own pages. Only your own tracking does.

What I’m watching in 2027

The agent that picks for the customer. McKinsey thinks an AI agent could become customers’ main banking interface within 3 to 5 years. Visa expects agents to drop institutions whose product data is not structured or current before any human sees the shortlist.

Banks moving into the assistants. BBVA put its app inside ChatGPT in Germany and Italy. NatWest became the first UK bank to launch an app in ChatGPT, for mortgage guidance, in April 2026. Accenture’s advice is to join these platforms only if the brand stays visible, and to publish products, pricing and service through partner APIs and pages built for AI agents.

The interface fight. In McKinsey’s UK survey, 71% of banking AI users turn to general tools such as ChatGPT and Gemini for financial tasks, and 45% use bank chatbots. Yet in McKinsey’s global survey, 62% of consumers most trust their main bank to offer AI financial services, against 19% for a big tech company, and 57% would consider a third-party AI agent if their bank had none. For now, people trust banks more and use the general tools more.

Ads next to the answers. OpenAI is expanding ads in ChatGPT and states that advertising does not influence the answers. If that holds, paid placement will sit beside the answer, and banks will still have to earn a place inside it.

If I had to pick one first step for 2027, it would be a small one. Ask the assistants your customers use the questions your customers ask, and read what comes back.

Then fix what they get wrong, one page at a time.

Sources

• Visa Consulting and Analytics, “SEO in the Era of AI-Mediated Discovery”, 2026, https://corporate.visa.com/content/dam/VCOM/corporate/services/documents/seo-in-the-era-of-ai-mediated-discovery-accessible.pdf
• McKinsey, “How gen AI agents threaten retail banks’ customer relationships”, April 2026, https://www.mckinsey.com/industries/financial-services/our-insights/how-gen-ai-agents-threaten-retail-banks-customer-relationships
• McKinsey, “Global Banking Annual Review 2025: Why precision, not heft, defines the future of banking”, October 2025, https://www.mckinsey.com/industries/financial-services/our-insights/global-banking-annual-review-2025
• McKinsey, “The decade to redefine UK banking”, September 2026, https://www.mckinsey.com/industries/financial-services/our-insights/the-decade-to-redefine-uk-banking
• BCG and FT Partners, “Global Fintech 2026: From Recovery to Resurgence”, June 2026, https://www.bcg.com/publications/2026/from-recovery-to-resurgence-in-global-fintech
• BCG, “How AI-First Banks Are Rewriting the Rules of Retail Banking” (Executive Perspectives), June 2026, https://www.bcg.com/publications/2026/how-ai-first-banks-are-rewriting-the-rules-of-retail-banking
• Accenture, “Top Banking Trends 2026: Unconstrained Banking”, December 2025, https://www.accenture.com/en/insights/banking/accenture-banking-trends-2026
• Pew Research Center, “Google users are less likely to click on links when an AI summary appears in the results”, July 2025, https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/
• Google, “Q2 2026 earnings call: Remarks from our CEO”, July 2026, https://blog.google/company-news/inside-google/message-ceo/alphabet-earnings-q2-2026/
• OpenAI, “Building advertising for the way people use AI”, October 2026, https://openai.com/index/new-chatgpt-ads-format-and-measurement/
• Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan, Ameet Deshpande, “GEO: Generative Engine Optimization”, KDD 2024, https://arxiv.org/abs/2311.09735