Last updated: September 12, 2026
This page describes statutes and is not legal advice. Take your final wording to your own counsel before it reaches the floor.
The line your agents read, and whether you need it
Put this at the front of the call: "This call is recorded for quality assurance. Is that all right with you?" Whether the question on the end is required depends on the consent rule in every state the call touches. One party states accept notice from a participant. All party states want everyone on the line informed.
The federal floor sits at 18 USC 2511(2)(d), which permits "a person not acting under color of law" to intercept a call they are a party to, or one where another party "has given prior consent," with an exception for interception aimed at a criminal or tortious act. Your agent is a party. Federal law by itself would let you record and say nothing.
State law is where the obligation actually comes from, and the states split. Some are satisfied when one participant consents, which your own agent does by pressing record. Others want every participant informed or agreeing. Which states sit on which side, plus the ones whose statutes fit neither bucket cleanly, belong to call recording laws by state rather than this page.
The scripts
Five versions cover almost every seat on a floor. Adjust the wording; keep the trigger points.
Inbound call, recorded:
"Thanks for calling [company]. This call is recorded for quality assurance. How can I help you?"
Outbound call, recorded:
"Hi, this is [name] with [company]. I want to let you know this call is recorded for quality assurance."
Live monitoring with nothing retained:
"This call may be monitored for quality assurance."
Transfer into a recorded queue:
"I am transferring you to [department] now, and that line is recorded for quality assurance."
All party consent, asking for agreement:
"This call is recorded for quality assurance. Is that all right with you?"
Watch the third one. Saying "monitored" while the audio is being written to storage describes something that did not happen, so an operator using it out of habit on a recorded account has given a notice that does not match the file.
All five download as a CSV with the placement column attached, ready to drop into an account script.
Where each one goes in the call
Placement is mostly operational practice. One statute in the set we read writes it down: Connecticut General Statutes 52-570d(a)(2) accepts verbal notification "which is recorded at the beginning and is part of the communication by the recording party." Worth borrowing as a house standard, since the start of the call is the only place a notice can precede what it gives notice about.
| Scenario | Where it goes | Why there |
|---|---|---|
| Inbound call, recorded | Directly after the company name, before the caller states their business | Gives notice while the caller can still decide to hang up |
| Outbound call, recorded | Inside the opening identification, before any question to the called party | The called party learns who is recording before answering anything |
| Live monitoring with nothing retained | Same slot as the inbound greeting, before the first substantive exchange | Notice arrives ahead of anything a supervisor would hear |
| Transfer into a recorded queue | Before the transfer connects, while the caller is still on your line | The receiving queue never played them a greeting |
| All party consent, asking for agreement | Before the first substantive exchange, with a pause held for the answer | Silence is not agreement, so the pause is the whole point |
Our first instinct was to put the notice above the greeting, since earlier is always safer. That turns out to be wrong on a real floor. A caller who hears "this call is recorded" before knowing which business picked up assumes they misdialed, and the operator eats the hangup. Company name, then notice, then the question.
The beep tone question
Somebody eventually tells you the law requires a periodic beep. In Connecticut that is close to true: 52-570d(a)(3) offers an automatic tone warning device "which automatically produces a distinct signal that is repeated at intervals of approximately fifteen seconds" as one of three ways to satisfy the statute.
Read the structure, though. The tone substitutes for the spoken notice, and it lives inside one state's statute instead of applying to recorded calls generally. A floor already reading a notice at the top of every call satisfies that statute by the second method and has no reason to add a signal every fifteen seconds.
What consent by continuing the call actually rests on
The common answer to all of this is that the caller heard the notice and kept talking, so they consented. Ordinary enough, and roughly how most operations behave. It is also thinner than it feels.
Continuing a call evidences consent from someone who heard the notice and understood it. The awkward cases are the ones where neither is obvious. Somebody joined mid conversation on a conference bridge and never heard the greeting. A caller on speakerphone put a second voice in your file that nobody ever addressed.
Neither situation makes the recording unlawful, and neither gets settled by a blog post. Both explain why the version that asks a question beats the version that announces: an answer sitting in the audio is better evidence than an absence of objection.
Calls that cross state lines
A single script survives right up until your dialing list crosses a border. Once the agent sits in one state and the caller in another, two consent rules are in play, and the conservative reading is that you satisfy both. Do that across twenty states and you are writing to the strictest one on the list, which lands you back on the version with the question. It costs three seconds a call and removes any need to know where the caller is sitting.
Medicare agencies carry a second obligation on top of this one and the two get conflated constantly. The recording notice is state law about recording. The TPMO disclaimer on a Medicare call is federal marketing regulation about plan representation, with its own wording and its own trigger, so a call can satisfy one and fail the other. Services taking calls for medical practices have a third layer in HIPAA and recorded calls.
Proving the notice was read
Having a script and following it are separate facts, and the second one is what a client asks about. A policy document shows the first. Recordings show the second.
A speaker separated transcript puts the disclosure at a timestamp and the caller's first substantive sentence at another, which turns the ordering question into arithmetic. Store each account's exact disclosure string in a vocabulary library and the criterion checks itself: those words, in that account's wording, ahead of the caller's first line. CallSniper scores it with searchable transcripts with timestamps behind every flag, and a missing notice usually belongs in the auto fail column, the way scoring a disclosure as a critical fail handles it on the general template.
Two limits worth stating. Scoring establishes that the words were said and nothing about whether the caller heard them, understood them, or was already talking over the greeting. Rule based criteria also hold up better under machine scoring than anything involving tone, which Aircall and StackAI both note about automated QA.
A perfect score proves nothing if the script itself went stale. An account you picked up last quarter may sit in a state whose rule your wording predates, and every agent passes a criterion written against the old line.
Frequently asked questions
What do you say when recording a call?
"This call is recorded for quality assurance" covers the notice, delivered right after your company name and before the caller gets into their reason for calling. Adding "Is that all right with you?" turns notice into an answer you can point at later. Use the longer version if any state you dial requires all parties to agree.
Is "this call may be recorded" enough?
It reads as notice rather than agreement, and those are different things. Where one participant's consent satisfies the statute, notice is more than you strictly owe, since your agent is already a party. All party states are the harder case, because you want an answer on the recording. Check yours in call recording laws by state.
Do you have to say the call is being monitored for quality?
A supervisor listening live without keeping audio is a different event from a recording, and "monitored for quality assurance" describes it accurately. Some statutes reach live interception as well as retention, so the safer practice announces both the same way. Saying "monitored" while the file is being written is the version to avoid.
Who has to be told a call is recorded?
Everyone whose voice ends up in the file, which is more people than the greeting reaches. A caller transferred into a recorded queue, a third party conferenced in halfway through, and a supervisor joining live all arrive after your opening line played. Each of those moments needs its own notice.
One script or a routing table
Two ways to run this. One national line written to the strictest state you dial, read on every call regardless of where the caller sits. Or per state routing, where the dialer picks the wording from the caller's number.
Routing is defensible for a floor with volume concentrated in a few states and a dialer that already knows where the call is headed. It adds a failure mode: a misrouted area code puts the wrong notice on a call with nobody watching. A single strict script fails one way, which is an agent who skipped it.
Small operations almost always take the single script, and we would tell you to do the same until somebody shows that three extra seconds a call costs real money. One string to score against beats a table of them.
Two of your accounts are the ones most likely to ask how their calls get handled. Start there, and find out whether the notice was actually read. A call review runs your approved wording against the recordings you already have, at a fixed pilot price.
Sources
- 18 USC 2511(2)(d), the federal consent exception for a party to the communication. Read via Cornell LII on 2026-09-12. Verified against a primary text mirror, to be confirmed on uscode.house.gov before publish.
- Connecticut General Statutes 52-570d(a), the three permitted methods including verbal notification at the beginning of the communication and the automatic tone warning device at approximately fifteen second intervals. Read via FindLaw on 2026-09-12. Confirm on the Connecticut General Assembly site before publish.
- Existence of all party consent states as a category. Established generally, with state level detail held on the state law page.
- Implied consent by continuing a call. General doctrine, unverified, framed as a reading rather than a rule.
- Aircall and StackAI on rule based scoring holding up better than tone scoring. Third party, unverified.
- Transcript timestamps and vocabulary library matching as evidence of a delivered notice. Trinsic feature sheet, verified internally.