Skip to content

Operations Guide

Insurance Verification with an AI Dental Receptionist: What Should Route to Staff

AI dental receptionists that quote insurance coverage create more problems than they solve. The right boundary is the one Velyn enforces by design: the AI recognizes insurance and pricing questions, declines to answer them specifically, and routes the call to staff with a clear callback task. This guide walks through why that boundary matters and how to set the routing rules that make it work.

Problem framing

Insurance answers that are wrong cost the practice more than insurance answers that are slow. A patient who is quoted $50 for a procedure and billed $400 has a story they tell their friends. A patient who is told 'I'll have the billing team call you back this afternoon' usually does not.
The hardest part of insurance verification is not the lookup — it is the coordination between the plan's stated benefits, the practice's contracted rates, the patient's deductible status, and the specific procedure code being asked about. AI systems that try to handle this fail in subtle ways that the patient does not see until they get the bill.
Front-desk staff have the institutional knowledge to navigate the gray areas. They know which plans the practice routinely fights for prior authorization, which patients have unusual coverage histories, and which questions are easier to verify than they sound. That knowledge cannot be replaced by an AI; it can only be supported by an AI that knows when to route to it.

Implementation checklist

Confirm the AI receptionist routes ALL specific insurance coverage questions to staff — no exceptions, no 'easy' answers.

Define the exact phrasing the AI uses to decline an insurance question gracefully ('I can't speak to your specific coverage — I'll have the billing team confirm. Want me to set up that callback?').

Configure the callback task format to include the specific question asked, the procedure code if mentioned, and the patient's plan if known.

Never let the AI quote copays, deductibles, or out-of-pocket estimates — even for plans you know well.

Decide what new-patient insurance intake the AI can capture (plan name, member ID, group number) vs what staff must verify (eligibility, benefits, in-network status).

Set the urgency flag for insurance callbacks based on whether the patient is calling about a scheduled appointment vs general inquiry.

Route coordination of benefits questions ALWAYS to staff — the rules are too plan-specific for safe AI handling.

Confirm the AI does not attempt to explain dental insurance concepts (annual maximums, frequency limits, downgrades) to patients.

Track callback completion time for insurance routes — slow callbacks are where the patient-trust risk shows up.

Review insurance-routed calls monthly to catch any AI attempt to answer a question it should have routed.

Why AI receptionists should never quote insurance coverage

Dental insurance is plan-specific, patient-specific, and procedure-specific in ways that no AI can verify in real time. A given plan's stated benefits depend on the patient's deductible status (which changes throughout the year), prior procedures (which affect frequency limits), the specific procedure code (which the AI rarely has from a phone call), and the practice's contracted rate (which varies by plan version).

Even with all that information available, the answer the patient cares about — 'how much will I owe?' — requires the practice's billing system to run an estimate against actual current data. Quoting from a stale lookup or a generic plan summary creates a number the patient will hold the practice to, and that number will often be wrong.

The patient does not distinguish between 'the AI gave me a wrong number' and 'the practice gave me a wrong number.' Either way, the practice owns the bill and the trust hit. The safest pattern is the one Velyn uses by design: route every specific coverage question to staff. The AI's job is to recognize the question, not answer it.

What patients actually ask about insurance (vs what they think they ask)

Most patient insurance questions sound like 'do you take my insurance?' but actually mean one of several different things: 'are you in my plan's network?' (eligibility), 'will this specific procedure be covered?' (benefits), 'how much will I owe?' (estimate), or 'can you handle the paperwork for me?' (billing process). Each requires a different answer, and only the last is suitable for a generic response.

The AI receptionist's job is to recognize the question pattern and route accordingly. 'Do you take Delta Dental?' is an in-network question that staff can answer quickly from a maintained list. 'Will my crown be covered?' is a benefits question that requires looking up the patient's plan and procedure code. 'How much will the cleaning cost?' is an estimate question that requires running the practice's pricing against the patient's specific plan.

All three should route to staff. The AI's contribution is making sure the callback task includes enough context that the billing team can answer in one call back — not in two calls or three.

The right routing rules for insurance-related calls

Insurance routing rules should be broad and conservative. Anything that touches coverage, payment, copay, deductible, or 'will my insurance pay for X' should route to staff. Even questions that feel borderline — 'I just want to know if you take Delta Dental' — should route to staff if the practice's network status varies by plan version, which is common.

The cost of a false positive (routing a question the AI could have handled) is small: the patient gets a callback, which is the experience they would have gotten from voicemail anyway. The cost of a false negative (the AI answering a question it should have routed) is large: a wrong answer becomes a billing dispute. Route conservatively.

The exception is general practice information: 'are you accepting new patients,' 'what are your hours,' 'where are you located,' 'do you do pediatric cleanings.' Those are operational, not insurance, and the AI should handle them directly.

How insurance questions become callback tasks

When the AI routes an insurance call, the callback task should arrive at the front desk with enough context that the billing team can verify and call back in a single attempt. A vague task — 'Patient called about insurance' — forces the team to call the patient just to find out what was asked. That is two calls for the patient and double the front-desk effort.

The right task format includes: patient name and best callback number, the specific question asked (as close to the patient's exact words as possible), the plan name and member ID if the patient mentioned them, the procedure or service in question if applicable, and the urgency flag based on whether the patient has a scheduled appointment soon.

A well-formatted insurance callback task lets the billing team prepare before they call: look up the patient's coverage, run the estimate, confirm in-network status. The callback then takes three minutes instead of fifteen.

New-patient insurance intake: what AI can capture vs what staff must verify

New-patient insurance intake has two parts: capture and verify. Capture is recording the patient's plan name, member ID, group number, and policyholder relationship. Verify is confirming the captured information against the plan, checking eligibility, looking up benefits, and confirming in-network status.

AI receptionists can handle capture safely. They can ask the patient to spell out the plan name, repeat the member ID for confirmation, and note the relationship to the policyholder. They cannot handle verify — that requires hitting the payer's eligibility system, which has plan-specific access patterns the AI is not equipped to navigate.

The right pattern is for the AI to capture during the call (the patient has the information in front of them) and then queue a verify task for the billing team. The team verifies in their normal workflow, usually within the same business day, and reaches out to the patient only if there is a problem.

Coordination of benefits: always staff, no exceptions

Coordination of benefits — when a patient has more than one insurance plan and the practice has to figure out which one pays first — is the hardest question in dental billing. The rules vary by plan, by state, by the patient's relationship to the policyholder, and by whether the secondary plan has specific COB language.

No AI receptionist should attempt to answer COB questions, even at a high level. A patient who asks 'I have my dad's plan and my own plan, which one pays first?' should be routed to staff every time, without exception. The right AI response: 'That depends on the specific plans — let me have the billing team work through it with you. They will call you back to walk through it.'

The downside of routing COB calls is a delayed answer for the patient. The upside is the patient gets a correct answer. The cost-benefit favors routing every time.

Common AI failure modes around insurance

Most AI receptionist failures around insurance come from systems that were trained to handle 'simple' insurance questions and could not distinguish them from complex ones. The failure pattern is consistent and worth naming directly so practices know what to watch for during evaluation.

Quoting a generic copay from a plan summary that does not account for the patient's deductible status.
Answering 'do you take Delta Dental' yes when the practice only takes specific Delta Dental plans.
Estimating procedure cost from a publicly visible fee schedule that does not match the practice's contracted rates.
Explaining dental insurance concepts (annual maximums, frequency limits, downgrades) in ways that the patient interprets as a commitment.
Attempting coordination of benefits for patients with multiple plans.
Confirming in-network status from cached data that has not been updated for the current plan year.
Quoting estimated patient responsibility without including the carrier fee that staff would normally explain.

FAQ

Why should the AI receptionist route ALL insurance questions to staff? Some seem easy.

Because the patient cannot tell which insurance questions are 'easy' from their side, and the AI cannot reliably tell which questions are easy from its side. 'Do you take Delta Dental?' sounds simple but depends on which specific Delta Dental plan, which the patient often does not know. 'What's my copay for a cleaning?' depends on deductible status, plan version, and the practice's contract. The cost of a wrong answer is a billing dispute and lost trust. The cost of a routed answer is a callback the patient would have gotten from voicemail anyway. Route conservatively.

What exact phrasing should the AI use to decline an insurance question?

Something like: 'I can't speak to your specific coverage — I'll have the billing team confirm. Want me to set up that callback?' The pattern names the limit ('I can't speak to your specific coverage'), names what happens next ('I'll have the billing team confirm'), and offers the patient agency ('Want me to set up that callback?'). Avoid apology language and avoid making the AI sound deficient — the limit is intentional, not a failure.

Can the AI receptionist capture new-patient insurance information during the call?

Yes. Capture is different from verify. The AI can ask the patient to spell out the plan name, repeat the member ID for confirmation, and note the relationship to the policyholder. That captured information then queues a verify task for the billing team, who runs eligibility and benefits in their normal workflow. The split — AI captures, staff verifies — is the right boundary because the patient has the information in front of them during the call, and the billing team has the systems to verify it correctly afterward.

What should the AI do if a patient insists on getting an insurance answer during the call?

Hold the boundary calmly. 'I really can't speak to your specific coverage without checking — the billing team has access to the systems that would give you the right answer. They usually call back within a few hours during business days. Would you like me to flag this as urgent or is later today fine?' The patient is asking because they want to know — giving them a wrong answer to satisfy the urgency creates the bill dispute later. The right answer is the same boundary plus a clear path to the correct answer.

How fast should an insurance callback happen?

Same business day, ideally within four hours of the original call. Insurance questions where the answer takes too long start to feel like the practice is hiding something, even when the delay is just operational. Set internal targets: insurance callbacks before the patient's scheduled appointment if there is one, otherwise within four hours during business hours. Track the callback completion time monthly and tighten the target if it slips.

Can Velyn answer questions about the practice's general fees (e.g., 'how much is a cleaning')?

Only the practice's posted general-public price (not the patient-specific price). For example: 'Our standard new-patient exam is $X — but your specific cost depends on your insurance and the billing team can confirm.' This is the right pattern because the posted price is public and accurate, but it explicitly names that the patient-specific cost requires staff verification. Most patients understand the distinction immediately.

What about questions like 'do you accept my insurance' in general?

Route to staff unless the practice maintains a current, plan-specific in-network list that the AI can reference safely. 'Do you take Aetna?' is rarely answerable as yes-or-no — most practices are in-network for some Aetna plans and not others, and the answer depends on which specific Aetna plan the patient has. The safer pattern is route every 'do you take X' question to staff and let the team confirm against current plan-specific data. The exception is small practices with a clean in-network list that the AI can reference reliably — but those practices are rare in 2026.