Occupation manual 003 / Insurance production

Let AI prepare the client review. Keep insurance judgment licensed.

A conservative workflow for approved intake, source-linked policy facts, renewal difference flags, coverage-question preparation, escalation, and reviewed follow-up—without automating advice, underwriting, pricing, binding, or claims decisions.

Short answer: use AI like a preparation clerk. It may organize approved facts, extract exact fields with source references, flag differences without judging them, prepare questions, and draft a recap from reviewer-approved notes. Licensed and authorized humans retain coverage interpretation, recommendations, negotiation, underwriting, pricing, binding, claims, consumer treatment, and every customer-facing message.

State, license, and private-data boundaryThis is educational workflow design, not insurance, coverage, claims, legal, or compliance advice. Insurance rules vary by state, product, line of authority, carrier, appointment, and agency role. Never place customer nonpublic information into a general AI tool unless the authorized agency or carrier owner has approved the exact tool, account, purpose, data class, retention, and transmission method.

What the official sources make clear

AI may assist preparation

NAIC materials describe AI use across the insurance lifecycle but warn that large language models may produce confident-sounding wrong information. Human review, source control, governance, and risk-proportionate controls remain central.

Licensed responsibility does not transfer

The NAIC Producer Licensing Model Act defines negotiation to include direct advice on substantive benefits, terms, or conditions. Model unfair-practice and claims standards also make invented policy, rate, coverage, and claim statements unacceptable workflow outputs.

Jurisdiction warningNAIC model laws, bulletins, and maps are not automatically operative law in every state. Check the actual state authority, carrier and agency rules, product line, and role before production use.

Build the six-part Insurance Client Review Preparation Desk

01

Confirm authority and tools

Record jurisdiction, line of authority, licensed reviewer, system of record, approved AI tool, permitted data classes, retention rule, and authorized communication channel.

Output: authority and data gate
02

Capture customer-stated changes

Use the approved channel and label each fact by source and verification status. AI may identify blanks; it may not decide whether the customer qualifies or needs a product.

Output: intake and change-event card
03

Link every policy fact

Extract exact labels, dates, form numbers, shown limits, deductibles, premiums, and notices with the controlling document and page. Mark ambiguity “needs licensed review.”

Output: source-linked fact sheet
04

Flag renewal differences

Compare two human-supplied fact sheets. Report same, changed, or unclear—never better, worse, sufficient, covered, cheapest, approved, or recommended.

Output: neutral difference table
05

Prepare questions and escalation

Turn changes and conflicts into questions for the licensed reviewer. Route underwriting, pricing, claims, cancellations, complaints, sensitive traits, security incidents, and legal issues to qualified owners.

Output: review and escalation queue
06

Issue a human approval receipt

Draft only from reviewer-approved notes. Record sources, license/role, approver, authorized channel, system-of-record action, retention handling, and open escalations.

Output: reviewed follow-up receipt

Give AI bounded jobs, never insurance authority

JobApproved inputAI outputHuman gate
Intake clerkMinimal approved facts and blank fieldsStructured card and missing-item listLicensed/authorized staff validates source and purpose
Policy fact clerkAuthorized source documents in an approved systemExact fields with page referencesLicensed reviewer verifies against controlling language
Difference checkerTwo source-linked, human-supplied fact sheetsSame/changed/unclear flagsProducer interprets meaning and client impact
Question-prep clerkVerified differences and customer-stated changesNeutral questions and escalation labelsQualified owner investigates and decides
Follow-up drafterReviewer-approved notes and templateDraft with unresolved items clearly markedAuthorized human checks facts and sends

Do not delegate these decisions

AI may prepare

  • blank intake and review cards
  • exact source-linked fact tables
  • same/changed/unclear difference flags
  • missing and conflicting item lists
  • questions for a licensed reviewer
  • reviewer-approved recap drafts
  • public regulatory update summaries

Qualified humans must control

  • selling, soliciting, and negotiating
  • coverage interpretation and recommendations
  • eligibility, underwriting, pricing, and binding
  • cancellation, nonrenewal, replacement, and suitability
  • sensitive-trait or proxy-data treatment
  • claims coverage, fault, fraud, value, payment, or settlement
  • all customer communications and record changes

Difference finding is not coverage advice

A renewal comparison may show that a date, listed item, coverage label, shown limit, deductible, form number, premium, or carrier notice changed. That is a factual flag—not a conclusion that coverage is better, worse, sufficient, suitable, affordable, or recommended. The issued policy, endorsements, authorized carrier records, and licensed review control.

Required labelsKeep separate: customer stated; source document; AI extracted—unverified; licensed reviewer verified; carrier action; and customer decision.

Claims are a handoff, not an AI answer

The workflow may record that a customer reported an incident, provide the carrier's approved contact route, organize customer-supplied documents, and draft a factual handoff. It must not say whether coverage applies, predict payment, assign fault, characterize fraud, value a loss, recommend omission of facts, or imply that the agent controls the claim outcome.

Use this prompt only after the data gate

You are a preparation clerk, not an insurance producer, underwriter, adjuster, claims examiner, lawyer, or compliance officer.

Organize the supplied approved facts into a source-linked table, missing/conflicting-item list, questions for the licensed reviewer, escalation flags, and source ledger.

Do not recommend coverage, limits, deductibles, products, carriers, or customer actions. Do not determine eligibility, underwriting, pricing, bindability, claim coverage, fault, fraud, value, or payment. Do not infer missing facts or sensitive traits. Preserve exact source wording and mark every ambiguity “needs licensed review.”

Final review before anything leaves the desk

First-party source desk

Prepare the review before choosing another AI tool.

Download the worksheet, test only with synthetic or expressly approved inputs, and obtain qualified insurance compliance review for the intended states, lines, carriers, systems, and roles.