Insurance Lead Qualification Scripts: What to Actually Say

Most AI voice agents sold to insurance agencies ship with the same generic script: a greeting, one open-ended question, and a hope that the caller keeps talking long enough to get transferred. It does not work, and when it does “work,” it fails a step later. The lead either hangs up because the first fifteen seconds sound like a robocall, or the call gets warm-transferred to a licensed agent carrying nothing but a name and a phone number, and the agent has to re-ask everything the caller already answered. A qualification script that actually holds up has four things in order: a compliant identification block, product-specific qualifying questions, a disclosure gate that fires at the right moment instead of a fixed timer, and a handoff that carries real information into the transfer. Here is what belongs in each part, why the rules force the order they do, and where a licensed human needs to take over from the machine.

What should an insurance lead qualification script say first?

It should say who is calling and why, before it asks the caller a single question, because two federal rules require exactly that order and neither one bends for an AI voice.

Under the current text of 47 CFR 64.1200(b)(1)-(2), a call placed using an automatic telephone dialing system or an artificial or prerecorded voice has to “state clearly the identity of the business, individual, or other entity that is responsible for initiating the call,” and either during or after that same message, state the callback telephone number of that business. Separately, under 16 CFR 310.4(d)(1)-(3), a telemarketer has to promptly disclose the identity of the seller, that the purpose of the call is to sell goods or services, and the nature of the goods or services being offered. Put plainly: before the script asks “are you still looking for coverage,” it has to say who is calling, that this is about insurance, and how to call back. That block runs ten to fifteen seconds at a natural pace, and it is not optional padding, it is the part of the call two different federal regulators specifically wrote rules to require.

Here are the terms that matter for the rest of this article, defined once. A qualifying question is a question whose answer changes what happens next in the call or who it routes to, as opposed to small talk. A disposition is the outcome tag applied to a call once it ends, such as “qualified, transferred” or “not interested.” A warm transfer is connecting a qualified, willing caller to a live licensed agent while they are still on the phone, as opposed to a callback promise. A TPMO, or third-party marketing organization, is any organization compensated to generate Medicare Advantage or Part D leads or enrollments, a category that includes most agencies running AI-assisted outbound. Prior express written consent is the specific written permission the TCPA requires before an automated or AI-placed marketing call or text can go to a cell phone, and it has to exist before the first dial, not get negotiated on the call itself.

Why a generic script fails before the qualifying questions even start

It fails because the vendors writing these scripts default to one flow for every product line, and insurance is not one product line, it is at least eight, each with a different first real question.

A caller who filled out a form about a Medicare Advantage plan is not the same conversation as a caller asking about final expense coverage, or a small business owner comparing a BOP quote, or someone whose homeowners policy just got non-renewed. A script that asks “what type of coverage are you interested in” and then runs the identical next four questions regardless of the answer treats all of them the same, and the caller can tell within one exchange that they are talking to a form, not a conversation. That is the mechanism behind AI voice agents sounding robotic: the problem is rarely the voice technology itself, it is a script written as one flat sequence instead of branching by what the caller actually said.

The fix is a branch point immediately after the identification block: one question that routes the rest of the call by product line, then a short, product-specific qualifying set. That single design decision, branch early, is most of what separates a script that gets used and works from one that gets muted three sentences in.

What it costs to get the script wrong

It costs two different things at once: a statutory exposure per bad call, and a lead who was reachable and is now gone because the first fifteen seconds sounded wrong.

Start with the exposure. Under 47 U.S.C. 227(b)(3), a person harmed by a TCPA violation can recover actual monetary loss or $500 in statutory damages per violation, whichever is greater, and a court can increase that award up to three times, to $1,500, for a willful or knowing violation. That figure is not an estimate from a law firm’s marketing page, it is the dollar amount written into the statute itself. A script that skips the identification block, mishandles an opt-out, or keeps dialing past a revoked consent is not a style problem, it is $500 to $1,500 of exposure per call, and a bad script run across a few thousand calls a month is not a rounding error.

Set that against the cost of writing and maintaining a script properly in the first place. The U.S. Bureau of Labor Statistics’ Occupational Employment and Wage Statistics program reports a median hourly wage of $16.58 for telemarketers as of May 2023, the most recent published OEWS detail for that occupation code, across roughly 81,580 people employed in the role. A managed AI caller running at $0.20 a minute, the published Single Account rate, costs roughly $12 an hour of call time at that per-minute rate, before the platform’s $200 monthly fee and $500 one-time setup are added in. The AI’s per-minute rate is cheaper than a telemarketer’s wage on a pure time basis, which is exactly why it is worth spending real effort getting the script right: a bad script at $12 an hour of AI call time still burns leads and still creates the same $500 to $1,500 per-violation exposure as a bad script read by a $16.58-an-hour human. Cheap minutes do not make a wrong disclosure cheap.

How to build the script: the four-part method, given away in full

Build it in this order and do not skip a part, because each one exists to satisfy a specific rule or to prevent a specific, predictable failure.

Part one: identification and purpose, every call, no exceptions. State the business name, that the call is about insurance, and a callback number, satisfying 47 CFR 64.1200(b)(1)-(2) and 16 CFR 310.4(d)(1)-(3) in the same breath. Keep it to one or two sentences. This is the part scripts skip most often because it feels like throat-clearing before the “real” conversation, and it is exactly the part both federal rules single out.

Part two: route, then qualify by product line. One routing question, then three to five product-specific qualifying questions from the table above. Stop at five. Every question past that point is marginal information collected at the cost of a caller who is starting to feel processed instead of heard. If the caller volunteers detail beyond what the script asks, let the script acknowledge it briefly and move on rather than ignoring it, because an AI that visibly ignores what was just said is the fastest way to confirm the caller’s suspicion that they are talking to a script.

Part three: the disclosure gate, timed to content, not to the clock. For anything touching Medicare Advantage or Part D, the TPMO disclaimer required under 42 CFR 422.2267 has to be delivered verbally before the conversation discusses any specific plan’s benefits, costs, or coverage, not on a fixed first-minute timer. That means the gate in the script is a conditional: if the next thing the flow is about to say is a benefit statement, the disclaimer fires immediately before it, regardless of how long the call has already run. Build this as a hard gate in the script logic, not a reminder to the AI to “mention it early,” because “early” is not what the current rule requires.

Part four: the warm transfer handoff, carrying everything collected. The moment a caller is qualified and willing to talk to a person, the transfer needs to arrive with the product line, the qualifying answers, the urgency signal, and a disposition code, spoken to the live agent or passed to their screen before they say a word, and synced to the CRM record in real time. A transfer that hands over a name and a callback number has thrown away everything parts one through three collected. This is also where the AI’s job ends and the licensed human’s job starts: coverage advice, plan comparison, and anything resembling underwriting detail belongs to the person taking the transfer, not the qualifying script.

Here is what that looks like laid out as an actual exchange, for a Medicare Advantage call, to make the four parts concrete rather than abstract. This is an illustrative example of the structure, not required legal wording; the exact phrasing is yours to write, but the order and the content of each part is not optional.

Notice that the disclosure gate sits between qualification and transfer, not at the very start. That placement is the direct, practical result of the current rule requiring the disclaimer before benefit discussion rather than on a fixed early timer: the script can ask county and enrollment status first, because neither question discusses a plan’s benefits, and only has to fire the disclaimer once the conversation is actually about to name what a specific plan covers.

The mechanics under the script: abandonment, ring time, and the answering-machine branch

The qualifying questions are only half the script. The other half is what the system does when nobody answers the way the script expects, and two federal rules specify that in exact numbers.

Under 16 CFR 310.4(b)(4), if a live sales representative, or an AI voice equivalent, is not connected within two seconds of the called person’s completed greeting, the system has to promptly play a recorded message stating the seller’s name and a callback number. Across an entire calling campaign, no more than 3 percent of calls answered by a live person can go unconnected this way, measured on a rolling basis. Separately, an unanswered call has to ring for at least 15 seconds or four rings before the system is allowed to disconnect and move to the next attempt. None of this is about the qualifying script’s wording; it is about what the dialing system is doing in the background, and it is worth confirming with whoever runs your calling platform, human-staffed or AI, exactly how these three numbers are enforced in practice, not just promised in a sales deck.

How we solve it

TheAffordableAI runs the branch-and-gate structure above as the default script logic, not an optional add-on, and syncs every qualifying answer to your HighLevel CRM the moment the call ends.

The AI caller identifies itself and the reason for the call before asking anything, routes by product line into the right qualifying set, fires the TPMO disclaimer gate before any Medicare benefit discussion, and warm-transfers a qualified caller to a live licensed agent with the full qualifying context already attached, or books the appointment against real calendar availability if nobody is free that second. Number warmup and spam defense run continuously in the background so the calls keep landing in the first place. A Single Account runs $200 a month plus a $500 one-time setup at $0.20 a minute, down to $0.15 at bulk volume; an Agency plan that routes a warm transfer to 20 or more agents at once runs $500 a month plus a $1,000 one-time setup at $0.18 a minute, down to $0.16 at bulk. Both are month to month, no contracts, cancel anytime. You can write and maintain this branching logic yourself on a dialer platform that gives you script control. Plenty of agencies do, and if you want to hear what the branch-and-gate flow above actually sounds like as a real call before you decide anything, there is a live demo on the homepage.

What you get from a script built this way

Where AI is the wrong tool for part of this call

It is worth saying plainly: the qualifying script should never try to be the whole conversation. The moment a caller needs coverage advice, a premium quote, a comparison between specific plans, or wants to disclose detailed health information, that is licensed work, and the script’s only job at that point is to recognize it and hand off cleanly, not to keep going because the AI technically could keep asking questions. A script that tries to extend its own role past qualification, to sound more like a closer than a router, is the version most likely to create both a bad caller experience and a compliance problem, because the deeper into substantive advice an automated flow goes, the closer it gets to activity that requires a license it does not have. Building the hand-off point deliberately, and early, is not a limitation of the technology. It is the correct design.

How to test the script before it goes live on real leads

Run it against a checklist before the first real call, because a script that reads correctly on paper can still fail once it is actually speaking to someone.

Call the AI yourself, on a real phone, and listen for four things in order. First, does the identification block land in the first fifteen seconds, clearly enough that a distracted person answering an unknown number would actually hear who is calling and why? Second, does the routing question correctly send you down the right product-line branch when you answer naturally, in your own words, rather than with the exact phrase the script expects? A branch that only recognizes “Medicare Advantage” and not “my Medicare plan” or “the Advantage thing” will misroute a meaningful share of real callers. Third, if you deliberately steer the conversation toward plan benefits, does the TPMO disclaimer actually fire before the AI says anything about coverage, copays, or premiums, every time, not most of the time? Fourth, when you ask to be transferred, does the live agent’s screen or opening sentence actually reflect what you just told the AI, or does the human start over from zero?

Log the results of this test pass with the same disposition and transcript fields a real call would produce, then have someone who was not involved in writing the script review the transcript cold. A script author reading their own transcript tends to hear the intended meaning; a fresh reviewer hears what an actual caller would hear, which is the version that matters.

The compliance part, said plainly

None of the script design above replaces reading and following the actual rules that apply to your calls, and “the AI followed the script” is not a defense if the script itself was built wrong.

The questions this raises next

Should the script be different for outbound versus inbound calls? The identification and disclosure requirements are written around outbound telemarketing calls specifically, so an inbound call the caller placed themselves does not carry the same identification-block obligation, though disclosing who they have reached is still good practice. The qualifying questions and the product-line branch structure, though, work the same way in both directions, since the goal, routing a caller efficiently without over-asking, does not change based on who dialed first.

What if the caller is calling about more than one product line? Let the routing question capture the primary reason for the call, qualify on that line first, and note the secondary interest in the disposition for a human follow-up, rather than trying to run two full qualifying branches in one call. A script that tries to fully qualify two product lines in a single pass runs long enough that it starts to feel like the interrogation this article warns against.

How often should the script itself be reviewed against current rules? At minimum, whenever CMS updates its Medicare Communications and Marketing Guidelines for a new plan year, and whenever the FCC or FTC amends 47 CFR 64.1200 or 16 CFR 310.4, both of which have changed multiple times in recent years. A script written against last year’s disclosure timing rule and never revisited is exactly how an agency ends up non-compliant on a rule it did follow correctly twelve months ago.

Does a shorter script always convert better? Not necessarily; a script that is short because it skipped the identification block or the disclosure gate is not better, it is exposed. The goal is not minimum length, it is no wasted or redundant questions past what routing and qualification actually require. A compliant four-part script run efficiently is usually still under a minute before the transfer point.

What should the AI do if the caller starts asking underwriting-style questions mid-qualification? Hand off immediately rather than attempting an answer, even a hedged one. A qualifying script’s value comes from staying inside its lane; the moment it starts answering coverage questions it was not designed or licensed to answer, it has taken on the exact risk the handoff point exists to avoid.

A script is not a formality sitting in front of the real conversation. It is the conversation’s first minute, and both federal telemarketing law and CMS’s Medicare marketing rules were written with specific requirements for exactly that minute. Build the identification block first, branch by product line into a short qualifying set, gate the Medicare disclosure to content instead of a clock, and hand the transfer off with everything the caller already said. Everything past that is the licensed human’s job, and the script’s last, best function is knowing exactly when to get out of the way.