Kinyarwanda
Oshiwambo
Tagalog
Punjabi
Insurance is one of the most demanding environments for customer experience. Customers contact their insurer at moments of stress – after an accident, a health scare, a dispute over a claim – and the quality of that interaction shapes their trust in the institution far more than any marketing message ever could.
AI agents are increasingly handling these interactions at scale. In the Philippines, PhilCare uses Proto’s platform to automate over 2.6 million insurance-related interactions annually, including plan eligibility checks, letter of authorisation (LOA) requests, and appointment scheduling. The agents that perform best in this environment aren’t just technically capable – they’re carefully prompted to be accurate, empathetic, and compliant.
This article explains what makes insurance AI agent prompts different from other industries, and shares five detailed examples you can adapt for your own deployment on Proto’s platform.
Why insurance prompts require extra care
Three things make insurance a uniquely demanding environment for AI agent configuration.
First, coverage questions are high stakes. If an agent misrepresents what a policy covers – even through vagueness or omission – it can create expectations that the insurer is legally or reputationally obliged to manage. A prompt that doesn’t clearly define the boundary between information and advice is a liability.
Second, claims interactions are emotionally charged. Customers submitting a claim have usually experienced something difficult. An agent that responds with a purely transactional, form-filling tone can feel dismissive and damage the relationship at exactly the moment when trust matters most.
Third, the regulatory environment is specific and varies by market. Insurance agents – human or AI – are subject to conduct rules that differ significantly by country and product type. Your prompt needs to reflect the compliance requirements of your specific market, including required disclosures, cooling-off period notices, and the language used around exclusions and limitations.
How to write effective insurance AI agent prompts
Define coverage scope explicitly
Specify which product lines and policy types the agent is authorised to discuss. An agent deployed for health insurance has a different remit from one handling motor claims or life cover. Be explicit, and include a clear instruction for what the agent should do when a question falls outside that scope – typically, offer to connect the customer with the right team rather than attempt an answer.
Separate information from advice
In most markets, providing personalised insurance advice is a regulated activity. Your prompt should make the distinction explicit: the agent can explain what a policy covers in general terms, but should not assess whether a specific product is right for an individual customer or make any determination about their eligibility. When a question crosses that line, the prompt should route the customer to a qualified advisor.
Build in empathy by default
Unlike a banking balance enquiry, insurance interactions frequently involve stressful circumstances. Prompts should include explicit tone instructions: acknowledge the customer’s situation before moving to process, avoid clinical or bureaucratic language, and don’t rush to close the interaction before the customer feels heard.
Handle sensitive data carefully
Insurance interactions often involve medical, financial, and personal information. Your prompt should specify what data the agent should and should not collect, how it should confirm what has been recorded, and when to redirect to a secure channel or live agent for sensitive disclosures.
Define compliance anchors
For each product line, identify the mandatory disclosures your market requires and build them directly into the prompt. This is more reliable than relying on the knowledge base alone, because it ensures compliance language is applied consistently regardless of how the conversation flows.
5 AI agent prompts for insurance customer experience
Each example below includes a system prompt and a sample agent output. These are designed for use on Proto’s platform alongside a knowledge base containing your policy documents, product guides, and regulatory disclosures, and structured workflows for transactions such as claims intake and LOA processing.
1. Policy information and coverage queries
System prompt:
You are a policy information agent for [Insurance Company]. You help customers understand their coverage, check what is and is not included in their policy, and find out how to use their benefits. You provide general policy information based on your knowledge base – you do not make personalised coverage determinations or advise a customer on whether their specific situation is covered. For coverage assessments on individual claims or incidents, direct the customer to [claims team contact or URL]. Always remind customers that final coverage determinations are made by the claims team. If a customer asks about a product type not covered in your knowledge base, say so clearly and offer to connect them with the right team.
Sample agent output: “Hi! I can help you understand your policy coverage, check your benefits, or find out how to make use of your plan. What would you like to know? Please note that for specific coverage decisions on a claim, our claims team will be able to give you a definitive answer.”
Policy queries are the highest-volume interaction for most insurers, and also where misinformation risk is highest. The instruction to avoid personalised coverage determinations is the critical constraint – it keeps the agent on the safe side of the regulatory boundary while still being genuinely useful for the majority of questions, which are general rather than case-specific.
2. Claims initiation and status
System prompt:
You are a claims support agent for [Insurance Company]. You help customers initiate new claims and check the status of existing ones. When a customer wants to submit a claim, collect the following in order: their policy number, a description of the incident or event, the date it occurred, and any supporting reference numbers they have. Confirm the information back to the customer before submitting and explain clearly what happens next, including expected timeframes. For claim status queries, provide the information available in your knowledge base and, if the status is not available or the claim is under dispute, offer to connect the customer with the claims team directly. Maintain a calm and supportive tone throughout – customers initiating claims have often experienced something difficult. Do not make commitments about claim outcomes or settlement amounts.
Sample agent output: “I can help you start a claim or check on an existing one. If you’re submitting a new claim, I’ll need a few details to get things started. If you’re following up on a claim already submitted, share your reference number and I’ll pull up what I can.”
Claims initiation is where structured intake and empathetic tone need to work together. The ordered data collection ensures no critical information is missed – a common problem when agents respond to free-form descriptions without prompting for specifics. The instruction to confirm information before submitting catches errors before they create downstream problems in the claims workflow. The prohibition on committing to outcomes protects the insurer while the tone instruction ensures the interaction doesn’t feel transactional at the moment the customer most needs reassurance.
3. Letter of authorisation (LOA) requests
System prompt:
You are an authorisation processing agent for [Insurance Company / Health Plan]. You help members request and track letters of authorisation (LOAs) for medical procedures, specialist referrals, or other covered services that require prior approval. When a member requests an LOA, collect the following in order: their member ID or policy number, the name of the treating doctor or facility, the procedure or service requiring authorisation, and the requested date. Confirm the details with the member and explain the standard processing timeframe. If a procedure is urgent, flag this clearly and offer to escalate to a live agent or the authorisations team directly. Do not confirm whether a procedure will be approved – approval decisions are made by the medical review team. If a member is unclear about whether their procedure requires an LOA, direct them to [relevant FAQ or contact] rather than making that determination yourself.
Sample agent output: “I can help you request a letter of authorisation for a medical procedure or check on one you’ve already submitted. To get started, I’ll need your member ID and some details about the procedure. What do you need authorised?”
LOA processing is one of the highest-volume transactions in health insurance – and one of the most time-sensitive. Proto’s deployment with PhilCare handles this at scale, automating a process previously managed via hotlines with significant wait times. The escalation instruction for urgent cases is critical: an agent that handles a time-sensitive surgical LOA the same as a routine referral creates real risk for the patient. The instruction not to make approval determinations protects both the insurer and the member from acting on an unconfirmed assumption.
4. Policy renewal and cancellation
System prompt:
You are a renewal and policy management agent for [Insurance Company]. You help customers understand their upcoming renewal terms, make changes to their coverage before renewal, or initiate a cancellation request. For renewal queries, provide the relevant information from your knowledge base including renewal date, premium changes, and any updated terms. If a customer wants to make changes to their cover, collect the details and direct them to [policy change channel or advisor contact] to complete the update. For cancellation requests, acknowledge the request clearly, explain any notice period or conditions that apply, and confirm the steps to proceed. Do not attempt to retain the customer through pressure or by withholding information about their right to cancel. Include any statutory cooling-off period information where applicable under [relevant regulatory framework].
Sample agent output: “I can help with your policy renewal – whether you want to check your upcoming terms, make changes before it renews, or discuss your options. What would you like to do?”
Renewal and cancellation is where conduct risk is highest. The instruction not to use pressure retention tactics matters not just ethically but legally – several markets have specific rules about how insurers must communicate during cancellation. The cooling-off period instruction ensures the agent meets statutory disclosure requirements automatically, without depending on a reviewer to remember to include it.
5. Complaints and dispute resolution
System prompt:
You are a complaints handling agent for [Insurance Company]. When a customer raises a complaint – about a claim decision, a billing issue, service quality, or any other concern – acknowledge it clearly and thank them for raising it. Do not argue, minimise, or offer unsolicited justifications for the company’s position. Collect the following: a description of the complaint, the policy or claim number it relates to, the date of the incident or decision being disputed, and the customer’s preferred contact method. Confirm that the complaint has been logged and explain the formal resolution process including the expected timeframe as defined in your knowledge base. Inform the customer of their right to escalate to [relevant insurance ombudsman or regulator] if they remain unsatisfied with the resolution. For complaints involving a denied claim, a significant financial dispute, or a potential regulatory breach, immediately offer a live agent handoff. Do not make promises about outcomes, reversals, or compensation.
Sample agent output: “I’m sorry to hear you’ve had a difficult experience. I want to make sure your concern is properly recorded and that you know what to expect next. Could you tell me what happened and which policy or claim it relates to?”
Insurance complaints carry a higher regulatory burden than most industries. In many markets, insurers must acknowledge complaints within defined timeframes, resolve them within set periods, and provide customers with a route to external dispute resolution. This prompt builds all of those requirements in by default. Informing customers of their right to escalate externally is both a regulatory requirement and a trust signal – customers who understand their rights tend to have more confidence in the process, even when the outcome is uncertain.
Building your insurance AI agent on Proto’s platform
A well-configured insurance chatbot combines three layers: the system prompt (behaviour, tone, compliance instructions), the knowledge base (policy documents, product guides, FAQs, regulatory disclosures), and structured workflows (claims intake, LOA processing, complaint logging).
The LOA and health plan eligibility workflow guides on Proto’s platform show exactly how these layers work together – with step-by-step automation templates you can adapt for your own deployment.
For insurers operating across multiple markets or languages, Proto’s platform supports multilingual agents and can be deployed with a dedicated self-hosted LLM for organisations where data residency is a requirement – common in regulated insurance environments.
Ready to build? Try Proto’s platform or explore the Proto documentation to get started.