Last updated: September 12, 2026
How to handle dead air while it is happening
Say what you are doing, attach a number to it, and get back on the line before that number runs out. "I'm pulling your account up now, give me about twenty seconds" turns silence into a wait the caller agreed to. Dead air is any silence long enough that the caller starts wondering whether the call dropped.
The number is the part agents skip. "One moment" has no end in it, so the caller starts counting on their own, and their count runs faster than yours. When twenty seconds turns into forty, come back before the first number expires: "still with you, this is slower than I said, about another thirty." Where the work is genuinely long, stop narrating and ask for a hold, which at least gives them music and a decision to make.
| Situation | Something to say |
|---|---|
| Waiting on a record to load | "Account's coming up now, about fifteen seconds." |
| Reading a policy before answering | "I want to read you the current wording rather than guess, give me a minute." |
| You do not know the answer yet | "I don't have that in front of me. Let me find who does, and I'll stay on while I look." |
Telling a caller you would rather check than guess buys more patience than any filler phrase, and it happens to be true.
What counts as dead air
Dead air is silence on a live connection where neither person is talking and nothing tells the caller a wait is underway. A hold is different, because music or a recorded notice says the line is open, and a pause of a second or two in normal conversation is just conversation.
Where the line sits is a choice your program makes. Common practice puts it somewhere between three and ten seconds, and nothing published fixes that number for you, so write your threshold into the rubric and say why you picked it. Count the longest single gap rather than total silence, because the long gap is what the caller lived through.
Whatever number you pick will read as strict to the floor, because an agent in a gap is busy searching a record and occupied time is short time. Eighteen seconds of work feels like five. The caller has none of that, and on the phone there is no way to see somebody typing, so the guesses that fill the silence run unflattering: the call dropped, or the agent stopped caring. By the time somebody asks whether you are still there, the call has moved somewhere you have to recover from.
Where dead air comes from
Two different problems produce the same silence, and telling them apart before you coach anybody is the difference between fixing it and annoying the floor.
| Cause | What it looks like on the recording | Where the fix lives |
|---|---|---|
| Slow lookup | Every agent goes quiet at the same point after the account number | The system, or a workflow forcing a lookup mid call |
| Systems that do not share a record | Long gaps while the agent retypes what the caller already gave | Integration, or a script that stops asking twice |
| Not knowing where the answer lives | Gaps scattered anywhere, concentrated in agents under ninety days | Training and the knowledge base |
| Reading silently | Steady pauses of five to fifteen seconds, same agent, every call | Coaching, and it is a fast fix |
| Avoiding "I don't know" | A gap right after a question the agent cannot answer | Coaching, and it is a slow one |
Silence spread across every agent at the same moment of the call is a tooling problem wearing a coaching problem's clothes, and a supervisor who takes that into a one on one loses the room. Four agents out of forty is where call center coaching does something.
Holds and transfers, where most of it lives
Pull a batch of flagged calls and much of the dead air sits at the edges of a hold or a transfer. An agent burns eight seconds deciding how to phrase the hold request, or comes back with the record open and starts reading before they start talking. Worst of the set is the transfer, where the button gets pressed and the caller sits in silence with no idea anything is happening.
Script the bookends and most of it goes away. Ask, wait for a real answer, give a length, then come back with a status even when the status is that you are still working. On a transfer, say who is getting the call and what they will already know, then stay on until somebody picks up. A cold transfer during a rush is a decision about queue time, which is worth naming in calibration rather than scoring as carelessness.
Setting a dead air line on your scorecard
Dead air belongs in the handling section of a rubric, worth a few points, and it is rarely a critical fail by itself. One long gap on an otherwise clean call is a coaching note. A call running 40 percent silence is a different conversation, usually about the tool.
| Where you draw the line | What it catches | What it costs you |
|---|---|---|
| 3 seconds | Nearly every lookup pause on the floor | So many flags that the criterion stops carrying information |
| 5 seconds | Gaps long enough that a caller starts filling them | A share of flagged calls will be an agent reading carefully |
| 10 seconds | Gaps where the caller has usually spoken up already | Misses the low grade pausing that irritates people most |
| 30 seconds | Calls where something actually broke | Too rare to coach against, useful as an alert |
Which brings up the part most rubrics get wrong. Some silence is an agent doing careful work, and a criterion that deducts for every gap teaches the floor to fill air with noise. Somebody who reads the current policy wording before answering leaves twelve seconds of hole in the recording and gives the caller a correct answer.
Write the criterion against unnarrated silence instead. An agent who says "I want to read you the exact wording" and then goes quiet for twelve seconds passes. Twelve quiet seconds with nothing said does not. Harder to define, fairer to defend, and it survives a disputed score. Put the threshold and that wording on the call center QA scorecard with the other handling criteria.
What a transcript can tell you about silence, and what it cannot
Silence duration is one of the few things on a QA form that a timestamped transcript measures exactly instead of approximately. The last word of one turn carries a timestamp, the next turn's first word carries another, and the gap is arithmetic. Two reviewers can argue all afternoon about whether an agent sounded empathetic. Neither will argue about whether a gap ran 14 seconds.
That exactness changes the question you can ask. Sampling stops being the constraint, so instead of asking whether this call was quiet you ask where on the floor the silence lives and when. Say 30 agents handle 42,000 calls a month. Two scored calls per agent per week comes to 240 calls, a little over half a percent, and a lookup that stalls on one account will not reliably surface in a sample that thin.
What the measurement does not do is explain itself. The transcript says the gap was 14 seconds and nothing about whether the agent was reading a policy or lost in a menu, or whether the caller minded. Speaker separation has to be reliable for any of it to hold, and a recording where both sides share one channel gives the system less to work with. Machine scoring is dependable on a rule based criterion like this one and shakier on judgment ones like tone, which Aircall and StackAI both note about automated QA. So the number gets you to the second it started, and reviewing the scored call is where you find out what was happening inside it.
CallSniper ingests recordings after the call ends, transcribes them with the speakers separated, and scores against the rubric your team writes, so the criterion runs on the whole month rather than the two calls somebody had time for. Manual upload covers a floor with no integration work, and a paid pilot processes six months of your own recordings for a fixed pilot price. Definitions for the rest of the terms on your form sit in the call center glossary.
Frequently asked questions
What is dead air in a call center?
Dead air is silence on a connected call where neither person is speaking and the caller has no signal that a wait is underway. It usually happens while an agent searches a system or waits for a colleague, and from the caller side it sounds the same as a dropped line.
How many seconds of silence counts as dead air?
Your program decides. Most rubrics put the threshold between three and ten seconds, and no published standard fixes it, so pick a number and hold it steady so scores stay comparable month to month.
What can an agent say to avoid dead air?
Narrate the task and give it a length, along the lines of "your account is loading, about fifteen seconds," then come back with a status before that length runs out. Longer work than a sentence can cover calls for a hold or a callback offer.
Is dead air the same as putting a caller on hold?
No. A hold gives the caller music or a recorded notice, so they know the line is open. Dead air leaves them nothing to interpret, which is why an unannounced ten second gap can land harder than a two minute hold.
Sources
- Trinsic feature sheet and call notes, 2026-08-27 and 2026-09-03. Verified internally.
- Aircall and StackAI on rule based criteria scoring more reliably than tone. Third party, unverified.
- Dead air threshold ranges of three to ten seconds, and the point weighting guidance. Editorial judgment for this article, not a published standard.