The short answer
Insurance teams should measure AI visibility at the shopper-scenario, policy-line, state, and distribution-role level—not as one carrier-wide rank. Separate unaided discovery from branded fact checks, resolve whether each named entity is a carrier, agency, producer, marketplace, or affiliate, and verify that the product can serve the declared state and need. Then grade recommendation role, fit, coverage claims, quote context, source support, and next-step handoff independently. Keep incomplete answers, ineligible candidates, and unresolved licensing or availability outside performance denominators rather than converting them into wins or losses.
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 guide proposes a measurement and review workflow for U.S. personal auto, homeowners, and renters discovery. It is not legal, insurance, actuarial, underwriting, licensing, coverage, or purchasing advice. The cited sources support the specific consumer and regulatory context described; they do not establish an AI visibility benchmark or certify an AI answer.
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 answer recommends a familiar brand but does not establish that the named carrier or product serves the shopper’s state.
- A comparison presents premiums as interchangeable even though the coverages, limits, deductibles, household facts, or quote dates differ.
- A lead-generation marketplace, local agency, licensed producer, underwriting carrier, and parent brand are counted as the same entity.
- Marketing reports a favorable recommendation while product, distribution, or compliance reviewers cannot inspect the prompt, full answer, sources, eligibility context, or handoff destination.
01
Define the insurance decision before the score
“Best insurance company” is not a stable unit of measurement. The useful unit is a bounded shopper decision with a policy line, state, need, risk context, and question family that reviewers can reproduce.
Bound the policy lines
Start with a declared scope such as U.S. personal auto, homeowners, or renters insurance. Keep health, life, annuity, commercial, specialty, and public-program questions in separate studies because their products, decision paths, authorities, eligibility, and material facts differ. A carrier-level average across unrelated lines hides whether the answer addressed the shopper’s actual need.
Describe a decision, not a persona slogan
Write the concrete job: identify carriers to research for a teen-driver household in Ohio, compare homeowners options for an older coastal property, or verify whether a renters product is offered in a state. Include only facts needed to define the candidate set. Never place personal, financial, medical, vehicle-identification, claim, or other sensitive data into a public AI prompt.
Separate discovery from verification
An unaided question asks which carriers or agencies a shopper should research. A fit question adds a material constraint. A branded verification question asks whether a named company offers a product or feature. A handoff question checks where the answer sends the shopper next. Report these families separately; naming the brand in the prompt is not evidence that it entered consideration unaided.
State what the sample can prove
Declare that the study observes completed provider answers to a fixed question × state × policy-line sample. It does not estimate policy sales, premiums, market share, customer preference, legal compliance, coverage, insurability, or claim outcomes. NIST distinguishes results on a fixed benchmark from broader generalized performance; use that boundary when writing conclusions.
Read NIST AI 800-3 ↗02
Resolve the carrier, agency, producer, and product
Insurance distribution creates identity traps. A consumer-facing brand may refer to a legal carrier, a group of underwriting companies, an agency, an individual producer, a marketplace, or a partner. Identity must be resolved before visibility can be scored.
Build an entity-and-role ledger
Give each observed entity a stable record: public brand, legal name when known, role, official domain, parent or affiliate relationship, states reviewed, policy lines reviewed, source URL, and checked date. Use explicit carrier, agency, producer, marketplace, affiliate, and unresolved labels. A shared logo, domain, or marketing name is not enough to merge roles.
Attach products to the serving entity
Record the policy line, public product or program name, state scope, sales channel, quote path, approved source, effective or checked date, and status. If the public evidence does not establish which legal entity would issue a policy, keep that field unresolved. Do not infer underwriting responsibility from the agency or marketplace that introduced the option.
Treat producer identity as state-scoped
NAIC explains that insurance producers sell, solicit, or negotiate insurance, that state insurance regulators license producers, and that state laws and requirements govern their activity. Use current state-regulator or approved licensing records when a review requires producer verification. A license in one context is not a universal endorsement, product-availability signal, or proof that an AI recommendation is appropriate.
Review NAIC producer-licensing context ↗Record mixed and wrong-entity answers
Use exact, probable, mixed, wrong-role, wrong-entity, former-name, and unresolved states. Preserve the answer span and source that support the label. If an answer combines one company’s brand with another company’s product, agency, phone number, or quote link, treat it as an identity failure even when the desired name appears.
Use the identity-confusion protocol →03
Build a shopper scenario × policy line × state matrix
A representative question set should cover genuinely different insurance decisions without turning every wording variation or hypothetical risk factor into a new market claim.
Stratify real decision cells
Cross a limited set of policy lines with states or regulatory markets, shopper tasks, meaningful constraints, and journey stages. Include markets and products with materially different availability or distribution paths. Do not let one national brand query stand in for state-level discovery, and do not let the largest line of business dominate a portfolio conclusion.
Use neutral unaided questions
Ask a question such as “Which auto insurers should I research for a two-car household in Arizona?” only after defining what makes a candidate eligible for that cell. Avoid naming the tracked brand, feeding the desired comparison criteria, or asking the provider to confirm a marketing claim. Save the complete candidate set, not only the tracked company.
Add one material constraint at a time
A separate cell may test a local-agent preference, a specific coverage need, a bundle consideration, or a digital-service requirement when that distinction changes the decision. Avoid mega-prompts that allow the answer to ignore half the conditions. Keep price-only questions separate from coverage-fit questions so a low-price rationale cannot silently substitute for a comparable policy.
Freeze a core and label exploration
Keep a stable core for period comparisons and place new states, emerging needs, seasonal risks, and experimental wording in an exploratory set. Promote a question only after reviewers agree it represents a recurring shopper decision. Retire or revise it with a reason and version so the denominator does not change invisibly.
Design a coverage-led question portfolio →04
Preserve one evidence record per attempted answer
Insurance reviewers need the answer and its decision context, not a percentage detached from the state, policy line, sources, and collection conditions that produced it.
Use a reproducible observation key
Store question ID and version × policy line × state × shopper scenario × provider surface × conversation state × repeat × collection time. Save the exact prompt, full answer, visible sources and links, completion state, retry relationship, evaluator, rubric version, and entity-ledger version. Keep incomplete answers and failures visible rather than silently replacing them.
Classify the recommendation role
Distinguish absent, passing mention, considered option, qualified recommendation, primary recommendation, caution, and rejection. Save the language that supports the label. An unordered list is not a ranking, and a company named only as an example or unavailable alternative is not an eligible recommendation.
Use the recommendation-role codebook →Grade independent layers
Record entity and distribution role, state and line eligibility, recommendation role, scenario fit, material-claim accuracy, source support, quote-context sufficiency, and handoff destination separately. A brand can be visible but ineligible, eligible but described inaccurately, accurate but unsupported, or correctly described while routed to the wrong intermediary.
Calibrate with inspectable examples
OpenAI evaluation guidance recommends task-specific objectives, representative cases, explicit criteria, logging, continuous evaluation, and calibration with human feedback. Adapt those principles to observable insurance answers. The documentation does not make a public answer deterministic or certify an insurance-specific label, so adjudicate consequential and ambiguous cases.
Review evaluation best practices ↗05
Verify coverage and quote claims like for like
A premium, deductible, limit, exclusion, or discount is meaningful only with the policy and shopper context that qualifies it. Extract claims atomically and preserve unknowns instead of completing the policy from inference.
Create a dated fact ledger
For each reviewed product and state, record only approved public facts: availability, coverage or option name, limit or deductible when stated, material qualifier, exclusion or separate-policy note when stated, discount condition, quote disclaimer, channel, authoritative URL, owner, and checked date. Version changed facts rather than overwriting the evidence used for an earlier answer.
Compare equivalent fields
NAIC consumer guidance says auto quotes should use the same coverages and limits and the same shopper information. Use that as an evidence rule: an AI answer cannot support a cheaper, better, or best-value conclusion when the compared coverage, limits, deductibles, inputs, or dates are missing or different. Mark the comparison insufficient rather than repairing it with assumptions.
Review NAIC auto-insurance shopping guidance ↗Keep policy language scoped
NAIC’s homeowners shopping tool compares dwelling limits, valuation basis, liability protection, deductibles, policy forms, optional coverage, premiums, and separate policies, and cautions that companies may define terms differently in a policy. Use the maintained policy and approved product materials for verification. A shared label does not prove identical coverage.
Open the NAIC homeowners shopping tool ↗Treat premiums as dated quote context
Record whether a number is an example, advertised figure, estimate, or actual quote; the state; policy line; coverage and limits; deductible; shopper inputs; discount assumptions; term; channel; currency; and collection time. Never place sensitive data into the monitoring set. A public example is not a quote for a person, and a quote is not a durable brand attribute.
06
Check eligibility, authority, sources, and handoff
An answer may sound useful while leaving the shopper with an unavailable carrier, an unresolved intermediary, weak evidence, or a destination that cannot complete the represented next step.
Verify state and line availability
Use the responsible carrier, agency, product, and regulatory records for the declared state and policy line. Record available, unavailable, conditional, unclear, and not reviewed. Do not generalize one state’s result to another or treat a national website as proof that every displayed product can be issued everywhere.
Use regulatory records for their stated purpose
NAIC’s Consumer Insurance Search materials let consumers research carrier information by state, company, and insurance type, including complaints and financial information. Those records can help resolve entity and regulatory context; they are not a single-factor quality ranking and do not validate every coverage or recommendation claim in an AI answer.
Review NAIC carrier-research guidance ↗Audit sources claim by claim
Map every material statement to the accessible source shown beside it. A regulator page may establish a license or complaint record but not a current discount; a carrier homepage may establish brand identity but not state-specific coverage; an agency page may describe service but not identify the underwriting company. Record fully supported, partly supported, unsupported, contradicted, inaccessible, and no-source states separately.
Use the claim-to-source audit →Classify the next-step destination
Record whether the answer points to an official carrier quote path, authorized agency, named producer, marketplace or lead form, informational page, unrelated entity, broken destination, or unresolved route. Preserve disclosures and role language visible at collection. A functioning link is not proof of eligibility, licensure, coverage, price, or suitability.
07
Report findings with honest denominators
The report should show where eligible products enter consideration, whether answers can be trusted, and where the shopper journey breaks. One blended visibility score cannot answer all three questions.
Publish collection and eligibility first
Show planned attempts, completed usable answers, incomplete answers, failures, unresolved entities, and unresolved state-product eligibility. Then publish the eligible denominator for each metric. Discovery presence, recommendation role, claim accuracy, source support, and handoff quality can legitimately use different denominators.
Compare within decision strata
Report by policy line, state, shopper task, question family, carrier or agency role, provider, and repeat before any rollup. Show counts beside rates. A strong renters result in one state should not hide an auto-eligibility failure elsewhere, and a branded fact-check result should not raise the unaided-discovery rate.
Run a bounded fictional pilot
A clearly labeled fictional pilot might use 12 questions: 4 unaided discovery cells, 3 constraint-fit cells, 3 branded verification cells, and 2 handoff cells across auto, homeowners, and renters in 3 states. With 2 providers × 2 fresh-context repeats, that creates 48 planned answers. If 43 are usable and 4 candidate-state pairs remain unresolved, report those evidence losses and exclusions—not a corrected national score.
Separate visibility from business outcomes
An eligible recommendation, accurate coverage description, supported source, or official handoff does not prove a quote, bound policy, retained customer, premium, loss ratio, or revenue effect. Connect web and business systems only through a declared attribution method, and keep sensitive customer and underwriting data out of the answer-review record.
Use the AI search attribution model →08
Turn findings into owned corrections and retests
Route each problem to the team that can verify the public record, product, distribution relationship, or collection method. The right action may be a correction, a state-specific clarification, an agency handoff fix, a new test, or no change.
Prioritize by shopper consequence
Escalate wrong carrier or agency identity, false availability, material coverage or exclusion errors, misleading quote comparisons, licensing ambiguity, and broken or deceptive handoffs before cosmetic language. Preserve the original prompt, answer, sources, and fact check so product, distribution, legal, compliance, or service owners can reproduce the finding.
Match the intervention to the evidence gap
Correct contradictory owned records at the responsible source. Clarify brand relationships, state availability, policy names, coverage qualifiers, agent roles, and quote context only when approved and useful to shoppers. Do not create thin state pages, unsupported superlatives, or repetitive copy merely to chase a provider answer.
Freeze the retest contract
Record the finding, verified truth, owner, approved change, release date, discovery check when available, target observation keys, review date, and success rule. Retest the same question version, state, policy line, provider surface, conversation condition, and rubric. Allow improved, unchanged, mixed, regressed, collection-failure, and method-break outcomes.
Start with one non-sensitive shopper question
Run one representative insurance-discovery question through the free AI Visibility Snapshot. Inspect the complete ChatGPT and Gemini answers, resolve every carrier and intermediary, and verify state, policy-line, material coverage, source, and handoff claims. Treat the result as a two-provider starting observation—not a coverage recommendation, quote, compliance review, national benchmark, or reason to buy a policy.
Run a free AI visibility snapshot →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.
- Consumer ↗
National Association of Insurance Commissioners
NAIC consumer resources explain that U.S. states, the District of Columbia, and U.S. territories have insurance departments; organize guidance by insurance type; and provide tools for company research, complaints, terminology, and regulator contact. The guide uses this as sector context, not as a product recommendation.
- Producer Licensing ↗
National Association of Insurance Commissioners
NAIC regulatory background defining producers and explaining that state insurance regulators license them and oversee sales and marketing activity. The guide uses it to require state-scoped role and licensing review, not to make a legal determination.
- A Consumer’s Guide to Auto Insurance ↗
National Association of Insurance Commissioners
Consumer guidance on choosing coverage and limits, comparing quotes with the same inputs, checking state requirements and licenses, and recognizing that premiums can change with the information supplied. It supports like-for-like evidence review, not an AIBL scoring formula.
- A Shopping Tool for Homeowners Insurance ↗
National Association of Insurance Commissioners
A consumer worksheet covering policy forms, coverage limits, valuation basis, deductibles, optional coverage, premiums, exclusions, and differences in policy definitions. The article uses those fields as examples of material claim context.
- How to File a Complaint and Research Complaints Against Insurance Carriers ↗
National Association of Insurance Commissioners
NAIC guidance on researching carrier information by state, company, and insurance type, while cautioning against relying on a complaint index as the only decision factor. It supports entity and regulatory-context checks, not a quality endorsement.
- Evaluation best practices ↗
OpenAI Platform Documentation
First-party guidance on task-specific evaluation objectives, representative cases, explicit criteria, logging, continuous evaluation, and human calibration. This guide adapts those principles to observable public insurance answers without implying provider control or insurance certification.
- Expanding the AI Evaluation Toolbox with Statistical Models ↗
National Institute of Standards and Technology
NIST AI 800-3 distinguishes evaluation on a fixed benchmark from broader generalized-performance claims and emphasizes evaluation design choices and uncertainty. It does not define an insurance visibility score or this workflow.
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 →