Yes, by design. The system prompt is configured to stay on the educational side of CMS marketing rules. The bot explains Medicare structure (A, B, C, D, supplemental, Advantage), enrollment periods, and general framing, but it does not recommend specific plans, run plan-comparison tools, or take enrollment information on the public site. Personalised plan questions route to a licensed agent with active CMS appointments in the prospect's state, which is where CMS expects plan-specific conversations to happen.

It explains subsidies conceptually, income-based premium tax credits, cost-sharing reductions for silver plans below certain income thresholds, the 8.5% income cap on premiums, without trying to compute the prospect's specific subsidy amount. Specific subsidy figures route to the agent or to healthcare.gov directly, both because the calculation depends on data the public bot shouldn't store and because the agent is the right person to walk the application with the prospect.

Not on the public site. The bot is configured to ask only what's needed for routing (ZIP, age band, market) and to defer detailed medical history, medications, and provider lists to a licensed agent in a more appropriate channel (phone, secure portal, agent-facing intake form). PHI handling has real requirements and the public chatbot isn't the right place for it. The bot captures enough context to route well, no more.

Yes, and most brokerages run them separately. Multibot plus display conditions lets you run a Medicare-focused bot on Medicare pages (with the CMS-aware framing) and an under-65 individual or family bot on marketplace pages, plus an employer group bot on small-business pages. Each bot has its own system prompt and routing rules, which keeps the senior market and the individual market cleanly separated.

Yes. Encode your carrier appointments and state license map as a page or ACF fields, and the bot uses that to route to the right agent. Cases where the brokerage isn't licensed or appointed get a clean referral or a polite decline rather than wasting agent time, which is a remarkably common failure mode for generic chatbots on health broker sites.

GPT-4o and Claude Sonnet both handle health insurance terminology well and follow the educational-not-plan-specific framing reliably. Stronger models are better at the Medicare timing logic (IEP, ICEP, SEP triggers) and at keeping CMS-compliant language consistent. Smaller or older models drift on enrollment periods and on plan-type distinctions under pressure. Bring your own key for whichever provider your compliance team has approved.

Every conversation is logged in WordPress with the model name, token usage, and page URL. Brokerages typically export transcripts on a recurring schedule via webhook into the same archive that captures the agency's other regulated communications. Retention is under your control, and the logs are searchable from the WordPress admin so compliance can locate a specific conversation later if needed for review or audit.

Yes, at an educational level. The bot describes HDHP-plus-HSA, FSA, HRA, and ICHRA at a general framing level, with the contribution limits and use rules, but it does not give specific tax advice or recommend a particular savings vehicle for an individual. Specific advice routes to a CPA partner or to the licensed agent if the question is about how an HDHP fits with the prospect's other coverage. That keeps the bot useful without crossing into tax advice.