AI Visibility for Banks and Credit Unions

People ask AI assistants the same questions they used to type into a search box, but in full sentences and with local context. For a bank or credit union, these are the prompts worth testing first:

• "What's the current 12-month CD rate at [your institution]?"
• "Does [your institution] charge a monthly fee on checking accounts, and how do I avoid it?"
• "Is the [neighborhood] branch of [your institution] open on Saturday?"
• "Can I join [your credit union] if I don't work for [sponsor employer]?"
• "Which credit union near me has the best high-yield savings rate?"
• "What's the overdraft fee at [your institution] compared with [competitor]?"
• "Does [your institution] have a branch or ATM network in [city]?"

The first four are about you by name. The last three are comparison prompts where you may not appear at all. Test both kinds. A wrong answer to a named prompt costs trust. A missing answer to a comparison prompt costs the account.

No single engine covers the whole journey, and each one draws on a different index.

• ChatGPT searches the web at the model's discretion. Its search uses OAI-SearchBot, not GPTBot. See how often ChatGPT searches the web before answering.
• Google AI Overviews and AI Mode use Google's index. Google states a page must be indexed and eligible to show with a snippet to appear as a supporting link. There are no extra technical requirements.
• Gemini is a separate Google product, governed in part by Google-Extended for training and grounding.
• Perplexity searches by default on most queries and uses PerplexityBot for its index.
• Claude uses Claude-SearchBot when search is on. On claude.ai, search is a user toggle.
• Microsoft Copilot grounds on Bing. In Otterly's 2026 experiment, Copilot kept citing sites Google had deindexed because Bing still held them.

The practical point: measure each engine separately. A healthy Google index tells you nothing about Bing, and a good ChatGPT answer tells you nothing about Perplexity.

Every model has a knowledge cutoff. OpenAI's model documentation lists a Feb 16, 2026 cutoff for the GPT-5.6 family and Aug 31, 2025 for the Instant path that ChatGPT also routes to. When an assistant answers without searching, it can only repeat rates it saw before that date.

Banks are unusually exposed to this. A CD rate or savings APY can change several times between model releases. A model trained on last year's rate sheet will state last year's rate with the same confidence as a current one.

You cannot fix a retrieval-off answer this quarter. Nothing a publisher does reaches it until a new model is trained. What you can fix is the retrieval-on path: make sure that when an engine does search, it can fetch a page that states today's rate in plain HTML.

It helps to know which kind of error you are looking at. Hallucination vs outdated answers covers how to tell a fabricated number from a stale one. A stale rate that matches an old rate sheet points to retrieval-off. A rate that never existed points somewhere else.

These are the facts to check first, because they change often or are easy to misread:

• Rates and APYs. Stale values, promotional rates presented as standard, or rates from a different term.
• Account fees. Monthly maintenance fees, waiver conditions, overdraft and NSF fees. A fee you dropped can live on in old coverage.
• Branch hours and locations. Holiday hours, closed branches, drive-through versus lobby hours.
• Membership eligibility. Credit union field of membership is often summarized wrongly, for example as employer-only when community residents qualify.
• Ownership and charter. Mergers and name changes confuse models trained before the change.

Say you run a credit union that merged with a neighbor in 2025. A retrieval-off answer may still name the old institution, quote its fees and list its branches. The fix is not on the old brand's pages. It is a clear statement on yours, plus correct listings elsewhere.

State each fact as a specific sentence on a fetchable page: "The 12-month share certificate earns 4.10% APY as of October 1, 2026." That is a hypothetical figure, but the format matters. A specific stated fact gives an engine a sentence it can quote. Publishing your prices for AI answers covers the same idea for fees.

Bank security stacks are built to turn away automated traffic. That is the right default for login and account pages. The risk is that the same rules also block AI search crawlers from public rate, fee and branch pages.

OpenAI's help documentation states eligibility as a conjunction: allow OAI-SearchBot in robots.txt, and confirm your host or CDN allows traffic from OpenAI's published searchbot IP addresses. A robots.txt that says allow, behind a WAF that returns 403, is a silent and complete failure.

What to check

1. robots.txt, per bot. Training and search bots are separate. OpenAI states each setting is independent. You can block training and still allow search:

User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

2. The WAF and CDN response. Request your rate page and a branch page as each AI search crawler. A 403, a 429 or a JavaScript challenge page means the crawler cannot read it.
3. Cloudflare defaults. From 2026-09-15, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for new domains, new sites and free-tier accounts. Search stays allowed by default. Check your zone settings rather than assuming.
4. JavaScript rate widgets. Vercel's crawler study found ChatGPT and Claude crawlers fetch JavaScript but do not execute it. If your rate table loads from a script or iframe feed, those crawlers may see an empty table. Compare the raw HTML with the rendered page.
5. Snippet controls. A stray nosnippet, max-snippet:0 or data-nosnippet on a template keeps content out of Google and Bing AI answers.

There is no robots.txt token for AI Overviews or AI Mode. Googlebot is the control there. The AI Crawler Checker requests a page as each AI crawler and shows which ones your firewall turns away.

When engines cite other sites instead of yours, being named correctly on those sites is the route into the answer. The evidence shows an association there, not a guaranteed effect.

For banks and credit unions, check:

• Rate comparison sites that list your products. A stale rate there can outlive the one on your own site.
• Map and directory listings for every branch, including holiday hours.
• Review platforms where members describe fees and service.
• Local news coverage of mergers, branch closures and fee changes.
• Regulator and industry directories that list your charter and name.

Fix the listings you control directly. For the rest, request corrections from the site owner. The local business fact check gives a checklist for hours and location data that applies to branches. If you also offer home loans, the mortgage broker guide covers rate prompts in that category.

Week 1: baseline

Run your named prompts and comparison prompts on each engine, logged out with memory off. Record which facts are wrong, missing or stale. The free checker on this site gives a snapshot for three buyer questions.

Week 2: access

Work with your security team. Confirm robots.txt per bot, then test the WAF and CDN response for OAI-SearchBot, PerplexityBot, Claude-SearchBot, bingbot and Googlebot on rate, fee and branch pages. Allow verified search crawlers on public pages only. Keep login and account paths locked.

Week 3: facts on the page

Move rates, fees, hours and eligibility into server-rendered HTML. Add a dated, one-line statement for each rate. Check Google Search Console and Bing Webmaster Tools separately for index status.

Week 4: third parties and re-test

Correct directory and comparison listings. Re-run the week 1 prompts. Expect retrieval-on answers to move first. Retrieval-off answers will not change until a new model ships, and nothing you do controls that timing. For a wider view of regulated finance, see AI visibility for fintech companies.