A bank's money market account page ranks on the first page of Google. A customer who visits it sees the rate after entering a ZIP code, opens an accordion for the account terms and downloads a PDF for the fee schedule. The page's structured data still carries last quarter's rate. The page works for the customer, but a crawler that does not run scripts or type into forms may find no rate or fees. The figure in the code may no longer match anything on the page.
This guide is for the SEO managers and web teams who build and maintain those pages. It covers product pages, schema, what Perplexity in particular needs, and branch pages. It assumes the content follows the pattern in our article on how large banks can compete with fintechs in AI search. Our guide to running an AI visibility audit for a bank lists access settings for each crawler in a table.
Make each product page indexable and eligible for a snippet, with the facts a customer asks about in HTML text near the top. Keep one canonical page per product and confirm the site is included in Google's generative AI features in Search Console. Google's AI Overviews and AI Mode draw on its Search index, so most of the work is the same work that makes the page rank. The difference is that a single passage, rather than the whole page, can end up in front of the customer.
Google's generative AI optimization guide sets the conditions. A page has to be indexed and eligible to be shown in Search with a snippet, and the site has to be included in Search generative AI features through a setting in Search Console. The same guide says there is no need to break content into small chunks, because Google's systems can find the relevant part of a page on their own. Microsoft describes Copilot's process differently, as one in which pages are parsed into smaller pieces that are judged separately. A page built so that each section answers one question clearly works for both.
Rate, fee and eligibility pages are Your Money or Your Life content in Google's rater guidelines, so a stale or unreadable figure carries more risk than it would on most sites. The table below covers the main page elements, what to build for each and why. The audit guide explains how to find these problems on a live site.
| Page element | What to build | Why |
|---|---|---|
| Rates and fees | An HTML text table with an "as of" date, rendered in the page's initial HTML rather than inserted by a script | Google says important content should be available as text, and in published tests the crawlers used by ChatGPT and Perplexity did not run JavaScript |
| Rates that depend on location | A page whose crawlable version shows a rate a customer can actually get, such as a state-level page or a clearly qualified rate, with the wording approved by compliance | Google says it will not load content that requires typing, so a rate that appears only after a ZIP code is entered is unlikely to be seen by Googlebot, and crawlers that do not run scripts will not see it either |
| Account terms in tabs or accordions | Keep the answer to the main question visible on load, with supporting detail in expandable sections if needed | Google accepts tabs and accordions when the content is in the HTML, but Microsoft warns that AI systems may not render hidden content |
| Fee schedules and disclosures | An HTML page or section for the core terms, with the PDF kept as the formal document | PDFs are indexable, but Microsoft notes they often lack the headings and structure that help AI systems parse them |
| Product variants | One canonical page per product, with campaign, state and legacy versions redirected or pointed to it | Google treats redirects and canonical tags as strong signals, and Bing says duplicates can delay updates in AI answers |
| Dates | A visible "Rates as of" or "Last updated" date that matches any date in the structured data | Google asks for consistent dates and lists changing dates without real changes as a warning sign |
| Snippet controls | A review of nosnippet and max-snippet on product templates, and data-nosnippet only where compliance wants a passage kept out of answers, never on disclosures that must travel with a rate | Google applies these controls to AI Overviews and AI Mode, and Bing now supports data-nosnippet in its AI answers too |
| Interstitials | Overlays rather than redirects for eligibility or entry gates | Google says intrusive dialogs make content harder to understand and advises against redirecting for consent |
The location-based rate is the problem most specific to banks. When a bank prices deposits or loans by market, a crawler that cannot enter a ZIP code sees a page with no rate at all. The fix is a product decision as much as a technical one, because any rate a page shows must be one the bank actually offers. Our article on keeping bank content compliant while optimizing for AI search covers the advertising rules that apply.
Bing has said more about freshness signals than Google has. It calls the lastmod field in a sitemap a key signal for deciding which URLs to recrawl. Search Engine Roundtable reported an attendee's summary of a 2025 talk by Microsoft's Fabrice Canel, quoting his recommendation to use IndexNow to push updates as they are published. When a bank changes a rate, a sitemap update and an IndexNow ping help the change reach Bing and Copilot sooner.
Use a small set of types that describe the bank accurately, keep every value in them identical to what the page shows, and do not expect schema alone to change AI visibility. For a typical bank site, we recommend Organization on the homepage, BankOrCreditUnion for each branch, BreadcrumbList across the site, and Article markup with author and dates on educational content. Product-level types from schema.org are optional, because Google offers no search feature for them.
The table below sets out what each type does and the bank-specific caution that goes with it.
| Schema type | Where it goes | What Google does with it | Bank caution |
|---|---|---|---|
| Organization | Homepage or about page | Helps Google identify the bank and choose the logo in results and the knowledge panel | Use legalName for the registered name and list official profiles in sameAs. Google suggests the iso6523Code property with prefix 0199 for a Legal Entity Identifier |
| BankOrCreditUnion, a LocalBusiness subtype | Each branch page | Local business markup can describe hours, departments and location | Google asks for the most specific subtype but does not name bank types, so this choice rests on schema.org's hierarchy |
| BreadcrumbList | Every page | Shows the page's place in the site hierarchy | Keep it consistent with the visible navigation |
| Article or BlogPosting | Educational and insight content | Supports byline dates and authorship | Dates and author names must match the page |
| FinancialProduct types, such as DepositAccount, LoanOrCredit, MortgageLoan and CreditCard | Product pages, if used | No Google search feature exists for them | Rates in interestRate or annualPercentageRate must update with the page, and APY has no dedicated property, so labeling is a compliance decision |
| FAQPage | Any page with visible questions and answers | Google stopped showing FAQ rich results on 7 May 2026 | Google said in 2023 that unused markup causes no problems, but it no longer earns a search feature |
| Review or AggregateRating | Not on the bank's own branches or products | Google does not show review rich results for LocalBusiness or Organization when the entity controls its own reviews | Embedding a third-party review widget does not change that |
The evidence that schema drives AI citations is weak. Ahrefs tracked 1,885 pages that added JSON-LD between August 2025 and March 2026 against 4,000 matched control pages and found no meaningful lift in citations on AI Overviews, AI Mode or ChatGPT. The pages were already heavily cited and Perplexity was not tested, but it is the only study we found with matched controls. Search Atlas found that domains with extensive schema were cited no more often than domains with little. SE Ranking found visible FAQ sections associated with more ChatGPT citations, while FAQ schema on its own was not. AirOps found that 61% of pages cited by ChatGPT carried three or more schema types. Findings like this show correlation, which Ahrefs attributes to better-maintained sites doing many things well at once. A small searchVIU experiment found that ChatGPT, Claude and Perplexity ignored JSON-LD when fetching a page live and read only the visible HTML. The test covered live fetching only, so engines with a search index may still use schema when indexing.
The platforms themselves are more positive about schema than the citation studies are. Google's guide says structured data is not required for generative AI search but is still worth using for rich results. Microsoft's Bing team says schema helps AI systems interpret content. Google's John Mueller, offering a personal view rather than official guidance, said the answer to whether schema helps with large language models is "yes, no, and it depends," and that a comparison site will not rank better by adding insurance markup.
A bank also needs to check whether its markup contains outdated rates. A rate in JSON-LD that lags the rate on the page breaks Google's rule that structured data must match visible content. It also puts a rate where reviewers do not see it, so compliance may want to review the markup alongside the page. If the bank's systems cannot update the markup from the same source as the page, it is safer to leave rates out of the markup.
Banks get cited by Perplexity when its crawler can reach and read their pages, when those pages carry current passages that answer the question on their own, and when the outside sources Perplexity tends to cite describe the bank accurately. We found no Perplexity guide to earning citations, so the evidence here comes from its technical documentation and from vendor studies.
Perplexity runs its own index. It describes an index tracking more than 200 billion URLs and says its retrieval system splits documents into fine-grained units that are scored individually against the query. That means a well-structured section can be retrieved even when the rest of the page is about something else, and a page that buries its answer gives Perplexity little to work with.
Access deserves a check of its own. Beyond the robots.txt and user-agent settings listed in the audit guide, the CDN or firewall in front of a bank's site can block Perplexity. Cloudflare began asking new domains at sign-up in July 2025 whether to allow AI crawlers and blocks them by default. In August 2025 Cloudflare said it had seen Perplexity crawling under undeclared user agents and removed it from its verified bot list, a claim Perplexity disputed. Whatever the outcome of that dispute, a bank's CDN and firewall settings can decide whether Perplexity sees its pages. The security team may manage those settings, so the SEO team needs a named contact there.
Rendering and freshness come next. In Vercel and MERJ's analysis of crawler traffic from late 2024, PerplexityBot did not run JavaScript, so content that loads only through scripts is unlikely to be read. An Ahrefs study of 17 million citations found that AI assistants cited content that was 25.7% fresher on average than pages ranking in organic results, and that Perplexity listed its references from newest to oldest. A product page with a current, visible "as of" date gives readers and Perplexity a dated fact, which is what the freshness findings point toward.
Perplexity also spreads its citations widely. A study by Qwairy, a small vendor, counted an average of about 22 citations per Perplexity answer against about 8 for ChatGPT, and Profound found Reddit holding 46.7% of the citations among Perplexity's ten most cited domains, though only 6.6% of its citations overall. A bank's own pages are therefore competing with many outside sources in each answer. Our banking AI citation report shows how much of Perplexity's banking citations go to publishers, and our article on how banks earn citations beyond their website covers the outside channels.
Give every real branch its own crawlable page with the same name, address, phone number, hours and services as its Business Profile and other listings. Link each page from a plain HTML directory. Remove or redirect pages for branches that have closed or changed names. Local answers draw on a mix of the bank's own pages and listing data, and both have to agree.
The evidence for branch pages is clearer than for most technical work. Google says it compiles Business Profile information partly from a business's official website, and its guide to generative AI features says Business Profiles help businesses appear in AI responses as well as regular results. Yext, which sells listings software, named branch location pages and advisor profiles as a significant citation source in its study of AI answers to financial services queries, and our banking AI citation report sets out how the balance between bank sites and listings shifts on local questions.
AI assistants are still far less likely than Google's local results to recommend a specific branch. SOCi, a local marketing software company, found the average financial services brand, in a segment that also includes insurers, mortgage companies and tax firms, appearing in Google's local three-pack 44% of the time, but recommended by ChatGPT only 3.1% of the time and by Perplexity 9.9%. The gap supports checking the basics on every page and setting cautious expectations for results.
The table below covers what each branch page needs.
| Element | What to publish | Why |
|---|---|---|
| A unique URL for each branch | One page per real location, linked from a state and city directory built with ordinary HTML links | Google can only follow links that are anchor elements with an href, and crawlers rarely type into search boxes, so branches reachable only through a ZIP search or a script-drawn map may go unfound |
| Name, address and phone | The same details as the Business Profile, using a local number where possible | Google asks for a phone number or website that represents the individual location, and Google's own systems may call that number to confirm details |
| Hours | Lobby hours, drive-through hours and holiday hours, kept identical to the profile and to the signs on the door | Google's rule for banks is to use lobby hours where possible, and Google Maps now lets users suggest new hours from a photo of a storefront sign |
| Services and departments | What this branch offers, such as drive-through service, an ATM, safe deposit boxes or a mortgage desk, with separate hours where they differ | Google's local business markup supports departments with their own hours, and the description gives each page content that differs from its neighbors |
| People | Branch managers or loan officers where the bank chooses to show them, with mortgage loan originator NMLS IDs where its compliance team requires them | Regulation G requires a covered institution to make its registered mortgage loan originators' unique identifiers, their NMLS IDs, available to consumers in a manner and method practicable to the institution, and showing them on branch or loan officer pages is one way to do that, subject to the bank's compliance team |
| Structured data | BankOrCreditUnion markup with the branch's own URL, coordinates to at least five decimal places and opening hours, and departments nested where relevant | Google's documentation says the url property must be the specific location's page and that coordinates need at least five decimal places, and the markup must match the visible page |
| Closed and renamed branches | Redirect the old page to the nearest branch or the directory, and remove leftover copies of old pages | In a Search Engine Journal session on local pages, Whitespark's Darren Shaw warned that AI systems would pick up outdated details from orphaned copies of old pages |
Each branch page needs content specific to that location beyond its address. Google's doorway policy gives as examples pages targeted at specific cities that funnel users to one page, and Bing warns that near-identical localized pages give its systems fewer signals to match each page to a need. A page for a real branch with its own hours, services and staff is legitimate. A templated city page for a market with no branch that only funnels visitors elsewhere matches the examples in Google's policy.
Business Profiles follow their own rules, and several apply directly to banks. Google allows one profile per location, asks every location of a chain to use the same name, and bans marketing taglines in the name, using a US bank's slogan as its example of what not to do. An ATM attached to a branch may have its own profile with its own hours, named as the bank's name plus "ATM" without location details. Profiles for Bing Places and Apple Business Connect deserve the same care. Microsoft says registering with Bing Places helps keep details eligible for inclusion in AI-generated responses. When Apple launched Business Connect in 2023, it said place cards reach Apple Maps, Siri and other apps. The tool is now part of Apple Business.
A template change can break rendering, markup or canonical tags on every page that uses it, so these checks belong in each release of a product or branch template as well as in an audit. The table below lists each check and a suggested owner.
| Check | How to test | Owner |
|---|---|---|
| Rates, fees and the main answer appear in the initial HTML | View the page source or fetch it with scripts disabled, and search for the rate | Web team |
| AI crawlers are allowed | Confirm robots.txt and firewall rules for each platform's crawler, as listed in the AI visibility audit guide | Web and security teams |
| Google can index and render the page | Use URL Inspection in Search Console, and confirm the site is included in generative AI features | SEO team |
| One canonical URL per product and per branch | Check canonical tags and redirects on variants and old URLs | SEO team |
| Structured data matches the page | Run the markup through a validator and compare every rate, date, address and hour with the visible text | SEO team, with compliance for rates |
| Branch details match listings | Compare the page with its Business Profile, Bing Places and Apple Business Connect entries | Branch operations or local marketing |
| Changes are announced | Update sitemap lastmod values and send IndexNow pings for changed URLs | Web team |
CB/I Digital can turn these checks into tickets for a bank's web team, each with the evidence attached, such as the raw HTML a crawler received or the profile field that disagrees with the branch page. Each ticket names the template it affects, so one fix reaches every page built on it.
Our SEO & AI Search services include technical work of this kind. Share the address of one product page and one branch page, and we will show you what each AI crawler receives from them.
Should a bank put its rates in structured data?
Only if the markup and the page draw on the same rate source and update at the same moment. A team can test this by changing a rate in a staging environment and checking that the page and the JSON-LD change together. If they do not, mark up the product type and leave the rate to the page text.
How can a team see what an AI crawler sees?
Fetch the page with scripts disabled or view its raw source, since that is close to what crawlers that do not run JavaScript receive. Search Console's URL Inspection shows what Googlebot rendered, and Bing Webmaster Tools' AI Performance report shows which of the bank's pages Copilot cites. Server logs show whether each platform's crawler is reaching the pages at all.
Do ATMs need their own pages?
Not necessarily. Google allows an ATM attached to a branch, or one inside another business, to have its own Business Profile, and a branch page can list an attached ATM and its hours. For standalone ATMs, a bank can choose between profiles, an ATM locator with crawlable links and dedicated pages for the locations that warrant them.
Copyright © 2026 CB/I DIGITAL INC. All rights reserved | Privacy Policy