The short answer
Banks should measure AI visibility at the deposit-product and consumer-decision level—not as one institution-wide rank. Define eligible products and markets first, test unaided discovery separately from named-bank verification, preserve every answer, and compare any stated APY, fee, minimum balance, term, bonus, access condition, and insured-status claim with a dated source of truth. Report visibility, product fit, factual accuracy, source support, and provider disagreement as separate layers, then route material errors through product and compliance review before anyone changes public content.
Editorial method · evidence boundaries
This guide proposes a review method. Worked examples and pilot sizes are illustrations unless a dated study is explicitly identified. Provider documentation supports only the behavior and product surface it describes; it does not prove that an AIBL workflow improves recommendations or business outcomes.
This is a proposed sector measurement framework. The saved AIBL audit does not establish sector-wide performance, customer outcomes or the effectiveness of this workflow. Domain references support their stated facts, not an AEO benchmark.
Historical AIBL records have limitations: completed jobs can contain truncated answers, and native citation metadata was not retained uniformly. Verify answer usability and source availability before scoring or comparing results. Missing saved metadata is not proof that a provider supplied no sources.
What businesses notice
Common signs of the problem
- An AI answer recommends the bank but describes a retired account, stale APY, or wrong fee-waiver condition.
- A parent brand appears for a national question even though the relevant deposit product is unavailable in the buyer’s market.
- A comparison blends checking, savings, CDs, fintech apps, and non-deposit products as though they were interchangeable.
- Marketing reports a favorable mention while product or compliance teams cannot inspect the prompt, answer, source, or collection date.
01
Define the banking decision before the score
A consumer choosing a first checking account is not making the same decision as a saver comparing CDs. The measurement contract must name the decision, eligible products, and market before a provider answer can be graded.
Choose a consumer job
Start with a real task: everyday checking, emergency savings, a fixed-term deposit, branch access, cash handling, or a specific account feature. Record the audience and consequence of the decision. Do not turn a broad prompt such as “best bank” into a universal measure of deposit-product visibility.
Separate product families
Keep checking, savings, money market deposit accounts, and certificates of deposit in distinct strata. The CFPB’s Regulation DD overview treats these as deposit-account types and identifies APY, interest rates, minimum balances, account-opening disclosures, and fee schedules as comparison information. Those fields create a useful factual review frame; they are not an AIBL score formula.
Review the CFPB Regulation DD overview ↗Declare the market and channel
Record the state or eligible market, whether the account can be opened online, any branch or relationship requirement, and the collection date. A product available nationwide online and a branch-only regional account should not share an eligibility denominator merely because they use the same bank name.
Write the boundary in plain language
State that the study measures completed answers to a fixed question-provider-market sample. It does not estimate deposits, account openings, consumer preference, regulatory compliance, or market share. NIST distinguishes fixed-benchmark performance from claims generalized to a wider population; use that distinction to keep conclusions bounded.
Read NIST AI 800-3 ↗02
Build a bank and product identity ledger
Resolve the institution and account before grading visibility. Bank trade names, legal entities, branches, partner programs, product tiers, and similarly named accounts can otherwise turn an apparent mention into an entity error.
Anchor the institution identity
Record the legal institution name, public brand, primary domain, FDIC certificate number when applicable, operating status, and relevant branches or markets. FDIC BankFind lets the public locate current and former FDIC-insured institutions by name, certificate number, website, or location. Use it to verify identity and status—not as proof that every product or statement in an AI answer is correct.
Open FDIC BankFind Suite ↗Give every product a stable key
Use a ledger row for institution, product family, official product name, tier or version, eligible market, channel, authoritative URL, and effective or checked date. When a name is reused after terms change, version the row rather than overwriting the facts that supported an earlier observation.
Separate deposits from adjacent offers
Mark credit cards, loans, brokerage, insurance, wallets, and other non-deposit products separately. If an answer mixes them into a deposit comparison, capture the classification error and the exact language. Do not infer insured status from a brand relationship, logo, app, or partner name.
Verify insured status at the right entity
The FDIC directs consumers to BankFind to determine whether a bank is FDIC-insured and explains that coverage attaches to deposits at an FDIC-insured institution, subject to applicable rules. Review claims against current FDIC materials and the specific legal entity; do not convert a general brand mention into a promise about an account or balance.
Review FDIC deposit-insurance FAQs ↗03
Design a product × need × market question matrix
The question set should represent how eligible consumers discover and verify deposit products without rewarding the tracked bank with easier wording than its peers.
Start with unaided discovery
Ask neutral questions such as “Which banks should I consider for a checking account with no monthly maintenance fee in Raleigh?” Define eligibility before collection and do not name the tracked bank. Record whether each response presents the institution as a genuine option, a passing mention, an ineligible alternative, or not present.
Add constrained fit questions
Test real requirements one at a time: monthly-fee conditions, minimum opening deposit, branch or ATM access, online opening, term length, early-withdrawal considerations, or customer segment. Avoid a mega-prompt that lets the provider satisfy one constraint while silently ignoring the others.
Keep branded verification separate
Questions such as “What is the current monthly fee for Bank X Account Y?” test description accuracy after the institution and product are supplied. They do not measure unaided discovery. Keep their denominator, label, and conclusion separate from recommendation questions.
Use representative cases and human review
OpenAI evaluation guidance recommends task-specific objectives, representative cases, explicit criteria, logging, continuous evaluation, and human calibration. Adapt those principles to observable consumer AI answers, while recognizing that an API evaluation framework neither controls nor certifies a public answer product.
Review evaluation best practices ↗04
Preserve one evidence row per answer
A bank cannot responsibly investigate a percentage that has been detached from the answer, market, product version, and fact sources used to judge it.
Save the collection context
Store the question ID and version, exact prompt, consumer-need stratum, product family, market, provider and visible surface, timestamp, repeat number, completion state, retry reason, full answer, returned sources, and evaluator. Keep failures visible but outside the brand-performance denominator.
Classify the recommendation role
Distinguish absent, passing mention, considered option, qualified recommendation, first choice, caution, and rejection. If the answer gives an unordered list, do not convert display order into a rank. If the product is ineligible for the declared market or need, record ineligible instead of scoring a win or loss.
Extract material claims atomically
Create separate fields for product name, APY, interest rate, fee and waiver condition, minimum opening deposit, minimum balance, bonus and qualification, term, early-withdrawal statement, availability, branch or channel access, and insured-status wording. Grade only claims the answer actually makes.
Record citation support separately
For each returned source, note whether it is official, current enough for the claim, and actually supports the nearby statement. A citation to a bank homepage does not validate a product-specific APY, and repeated wording across providers does not turn an unsupported detail into fact.
Use the full brand-perception audit →05
Compare answers with a dated source of truth
Deposit terms can be conditional and time-sensitive. The fact ledger should preserve what was true, for whom, in which market, and as of what date—without pretending that the guide itself is a compliance determination.
Use the maintained product disclosure
Prefer the bank’s current product page, rate sheet, fee schedule, account disclosure, and official terms for the declared market. Save the URL and checked time. If those sources conflict, mark the answer unresolved and route the underlying source conflict to the product owner.
Check rate language precisely
CFPB Regulation DD says that when a deposit-account advertisement states a rate of return, it must state the rate as an annual percentage yield, subject to the regulation’s details. In answer review, distinguish APY from interest rate, capture balance tiers and time limits, and never “correct” an answer from an unlabeled rate fragment.
Read current § 1030.8 advertising rules ↗Keep fee qualifiers attached
Record the fee, the exact waiver condition, and the population or product version to which it applies. Section 1030.8 also restricts misleading or inaccurate deposit advertisements and the use of “free” or “no cost” when maintenance or activity fees may be imposed. A bank’s compliance team—not an automated grader—should decide how current law applies to bank-controlled communications.
Review insured-status wording with care
Current FDIC Part 328 materials cover official signs, advertising statements, insured-status representations, and distinctions involving non-deposit products. Flag omitted qualifiers, wrong entities, or mixed-product wording for qualified review. Do not assume that every independent AI answer is the bank’s advertisement, and do not publish legal conclusions from this measurement workflow.
Review the FDIC Part 328 Q&A ↗06
Report visibility, fit, and accuracy separately
A bank can be visible for the wrong account, recommended outside its eligible market, or accurately described but absent from unaided discovery. One blended score conceals those different operating problems.
Publish collection coverage first
Show planned attempts, completed answers, failures, refusals, and excluded cells. Use completed eligible answers for visibility denominators and keep the raw count beside every percentage. A failed provider call is a collection failure, not an omission by the bank.
Separate institution and product visibility
Report the share of completed eligible answers that mention the bank, name the correct product family, and recommend an eligible current product. A parent-brand mention should not pass a product-level test when the account is missing, retired, or unavailable in the declared market.
Publish accuracy by claim class
For each claim family, show correct, contradicted, stale, unsupported, unresolved, and not stated counts. Keep APY and rate, fees, minimums, bonus terms, availability, identity, and insured-status wording separate so a high-volume low-risk field cannot hide one consequential error.
Triage by consumer consequence
Escalate wrong institution identity, insured-status wording, product availability, fees, qualification conditions, and material rate or term claims before cosmetic differences. Route the evidence to the product, legal, compliance, branch, or web owner that can verify it; visibility analysts should not rewrite regulated facts from memory.
07
Work through a 48-observation pilot
This fictional example demonstrates the measurement shape only. It is not provider, bank, customer, regulatory, or market-performance evidence, and twelve questions are not a universal minimum.
Freeze twelve questions
Use three checking questions, three savings questions, two CD questions, two local access questions, and two branded fact checks. Run each against two providers with two fresh-context repeats: 12 × 2 × 2 = 48 planned observations. Predeclare products, markets, eligibility, and grading rules.
Show completion before outcomes
Suppose 46 answers complete and two fail. Collection coverage is 46/48, or 96% when rounded. Visibility and accuracy use only the completed eligible cells for each stated measure. The two failures remain visible and are never scored as absent brands or incorrect products.
Expose a product-level contradiction
Imagine the bank appears in 18 of 46 completed answers, but only 11 name an eligible current product. Among seven completed answers that state a monthly fee or waiver condition, five match the dated source, one is contradicted, and one is unresolved. Report 18/46, 11/46, and 5/7 rather than one “bank visibility” percentage.
Preserve provider and market splits
Break the same counts out by provider, product family, consumer need, and market. If the contradiction appears in one provider and both repeats while other cells are accurate, investigate that bounded pattern. Do not claim a universal provider defect, a compliance violation, or consumer harm from the fictional counts.
08
Turn the baseline into a controlled operating loop
The purpose of the measurement is to find reviewable evidence gaps and factual risks. It is not to chase one generated answer with unapproved rate or product copy.
Start with one neutral decision question
Use the free AI Visibility Snapshot to compare how ChatGPT and Gemini answer one genuine deposit-product question for a declared market. Inspect both complete answers, brands surfaced, product fit, and stated facts. This is a two-provider observation—not a bank-wide audit or compliance review.
Run a free bank visibility snapshot →Assign each finding to an owner
Send product-term conflicts to the deposit product owner, web contradictions to the page owner, branch or market errors to operations, and consequential claims to the appropriate legal or compliance reviewer. Keep the original answer and verified source attached to the issue.
Correct the public record, not the score
Where a public source is stale, ambiguous, or inconsistent, update it only after the responsible team verifies the fact. Keep effective dates and qualifications visible. Do not publish unsupported parity claims, clone thin market pages, or remove important conditions merely to make a product easier for AI systems to summarize.
Repeat the unchanged matrix
After verified source changes have had a reasonable opportunity to become discoverable, rerun the same question-provider-market cells and compare like with like. Treat one changed answer as a signal to review; require repeated evidence before describing improvement as durable.
Plan evidence checkpoints →Continue the research
Related AI visibility guides
Primary sources
Primary documentation used for this guide
AI products, search behavior, and platform policies change. Check these maintained first-party sources before making technical decisions.
- BankFind Suite ↗
Federal Deposit Insurance Corporation
Official public tool for locating current and former FDIC-insured institutions by name, certificate number, website, or location, with institution and branch information. It verifies institution identity and status, not every product claim.
- Deposit Insurance FAQs ↗
Federal Deposit Insurance Corporation
FDIC consumer guidance on deposit insurance and using BankFind to determine whether a bank is FDIC-insured. Coverage depends on current rules and the specific institution, ownership category, and deposits.
- Questions and Answers Related to the FDIC’s Part 328 Final Rule ↗
Federal Deposit Insurance Corporation
Current FDIC implementation material on official signs, advertising statements, insured-status representations, digital channels, and non-deposit-product distinctions. Banks should review the maintained rule and qualified guidance for current obligations.
- 12 CFR Part 1030 — Truth in Savings (Regulation DD) ↗
Consumer Financial Protection Bureau
CFPB’s current navigable Regulation DD resource describing covered deposit-account types and disclosure topics such as APY, interest rates, minimum balances, opening disclosures, and fee schedules. It notes that official legal editions remain authoritative for legal research.
- § 1030.8 Advertising ↗
Consumer Financial Protection Bureau
Current Regulation DD advertising section covering misleading or inaccurate deposit advertising, rate terminology, triggering terms, fees, time-account features, and bonuses. The guide uses it to identify facts needing bank review, not to provide legal advice.
- Expanding the AI Evaluation Toolbox with Statistical Models ↗
National Institute of Standards and Technology
NIST AI 800-3 distinguishes performance on a fixed benchmark from generalized performance and emphasizes explicit assumptions and uncertainty. This guide applies that distinction conservatively to a bank’s declared answer sample.
- Evaluation best practices ↗
OpenAI Platform Documentation
First-party guidance on task-specific objectives, representative cases, explicit criteria, logging, continuous evaluation, and human calibration. It informs the workflow without validating any bank metric or predicting every public AI answer.
Measure your visibility
Turn the questions in this guide into an evidence-backed baseline.
Get AI Visibility ScoreTalk to us
Have a visibility question?
Tell us what your team is trying to measure or improve.
Contact AI Brand Lens →Keep learning
Explore every guide.
Browse practical answers about AI visibility, competitors, and measurement.
View all guides →