Last updated: September 12, 2026
What an empathy statement does on a call
Empathy statements are the phrases an agent uses to acknowledge a caller's situation before moving to a solution. The ones that work name the specific problem in the caller's own words and skip the obvious script tell. Below are phrases grouped by call situation, with the scoring question that follows each one: whether the agent said it, and when.
The list everybody wants does not work on its own. A phrase delivered on autopilot lands worse than silence, because the caller hears the gap between the script and the person reading it. Hand a new agent the list and nothing else and you have built a slightly more irritating floor.
What the phrase does is small. It marks the moment the agent stops collecting information and shows the caller that the problem registered. An agent who jumps straight to the fix is usually right and still gets a caller saying the same thing three times.
The tell that ruins a good line is generic wording. "I understand how you feel" tells the caller a list exists. "You have been without heat since Friday" tells the caller somebody was listening. One detail separates them.
Phrases by call situation
Grouped by situation, since the line that works on a frightened caller is wrong for somebody driving with two minutes to spare.
Angry about a problem that keeps coming back
The repeat caller gets handled worst, because the agent arrives without the history and the caller has all of it.
| Phrase | What it acknowledges | Scoring note |
|---|---|---|
| "This is the third call about the same leak, and I have all three messages here." | The caller already did this twice | Check the agent named the prior contact first |
| "You should not have had to call back for this." | The failure belongs to the service | Counts only before intake restarts |
| "Let me read back what I am sending, so you can tell me if it is wrong this time." | The last message probably was wrong | Pairs with the read back criterion |
Distressed or frightened
| Phrase | What it acknowledges | Scoring note |
|---|---|---|
| "You reached the right number, this line goes to the on-call provider." | Fear of having wasted the call | Appeared before triage, or it did not |
| "I am not putting you on hold for this." | Hold is what a frightened caller dreads | Check it against the hold events |
| "Say it however it comes out, I am writing it down." | The caller cannot get the words straight | Check it came before the first probing question |
Confused about a charge or a process
| Phrase | What it acknowledges | Scoring note |
|---|---|---|
| "Let me say back what I think you are looking at, and you correct me." | The caller lacks the vocabulary for it | The restatement is already scored on most forms |
| "You are reading it right, there are two charges on that date." | The caller suspects they are the problem | Scorable where the agent could see the account |
| "Nothing you did caused this." | The embarrassment under the question | Scores false if the agent later blames the caller |
Grieving
Hospice and funeral home accounts drop these on an agent between two routine calls.
| Phrase | What it acknowledges | Scoring note |
|---|---|---|
| "I am sorry. Take whatever time you need, I am not going anywhere." | The caller cannot move at call pace | Score the silence after it |
| "I can call you back in an hour instead, if that is easier." | The caller may not manage it now | Score the offer and the scheduled callback |
| "You do not have to explain any of it to me." | The caller is braced to justify themselves | Appears before capture, or it does not count |
In a hurry
| Phrase | What it acknowledges | Scoring note |
|---|---|---|
| "Two questions and you are done." | The caller expects this to take too long | Compare against the questions actually asked |
| "I have the address and your number already, so we are most of the way there." | The caller expects to repeat everything | Scorable against whether the record was open |
| "Go ahead and drive, I will call this number back in ten minutes." | The caller is doing something else | Score whether the callback went out in the window |
Reporting something urgent
| Phrase | What it acknowledges | Scoring note |
|---|---|---|
| "I am treating this as an emergency, so I am paging the on-call tech while we talk." | Fear of being filed as routine | The urgency criterion scores the action, this the words |
| "Give me the address once more and I will read it back before I send it." | A wrong address costs the caller the night | Pairs with the read back criterion, digit by digit |
| "If this becomes a safety problem before they reach you, call 911 and then call me back." | The caller does not know where the line sits | Belongs to the account script, so score it per account |
Phrases that backfire
Replacements matter more than bans. An agent told to stop saying something reaches for the next generic line unless you hand over the specific one.
| Phrase to avoid | Why it lands badly | Say instead |
|---|---|---|
| "I understand how you feel." | Announces a list, and invites "no you do not" | "You have been without heat since Friday" |
| "Unfortunately, that is our policy." | Leaves the agent behind the wall with nothing to offer | "Here is what I can do from this seat" |
| "Calm down." | Makes the caller's reaction the subject | "I am with you. Tell me the part that matters most" |
| "I apologize for any inconvenience this may have caused." | Conditional enough that the caller hears no apology | "You should not have had to call back for this" |
| "As I said before..." | Scolds the caller for not listening | "Let me say that a different way" |
| "There is nothing I can do." | Ends the call emotionally while the caller is on it | "That is past what I can do, so I am getting it to someone who can" |
All twenty four rows download as a CSV, do-not-say lines included as a prohibited set.
Empathy as a scorecard criterion
On the scorecard where this criterion lives, empathy is one line worth 5 points out of 100: used an empathy statement where the caller expressed frustration. That wording names an observable behavior instead of a quality, and it attaches a condition, so the line does not apply when nobody was upset.
The condition is where most scorecards break. Apply it everywhere and reviewers grade a calm billing question against a standard built for a furious caller, and the average drifts up until the number means nothing. Mark it not applicable on calm calls and drop it from the denominator, so that call scores out of 95. How QA scores are built covers the arithmetic.
Five points sounds low for the thing trainers spend a whole session on, and we went back and forth before leaving it there. Weight it at 15 and the softest line on the form moves the score more than the callback number read back digit by digit.
Reviewers disagreeing on the same call usually start here, because one is scoring the words and the other the delivery. Arguing in a calibration session over whether a caller sounded acknowledged sucks, and it is still the fastest way to catch two reviewers measuring different things.
What a machine can check and what it cannot
Transcript matching confirms an acknowledgement phrase appeared and stamps the second it appeared at. That is the whole of it. Aircall and StackAI both describe automated scoring as reliable on rule based criteria and weaker on tone, which is the honest read. Machine scoring here is a presence check.
A second failure is worth knowing about. On mono recordings where both sides share one channel, speaker separation can hand a line to the wrong person, so a phrase the caller said gets credited to the agent. Listening to a few flagged calls finds it faster than any threshold change.
Loading the list into a vocabulary library
CallSniper ships a vocabulary library with an Empathy set in it, alongside Offer to Assist, Slang, and Filler Words. The CSV above drops into that structure unedited.
Once the set exists, every ingested call gets checked against it, and the output is presence plus a timestamp. What a supervisor opens on Monday is the list of calls where the set never appeared. Say the floor takes 9,000 calls a month and the set is missing from one in eight. That is roughly 1,100 calls, too many to coach, and a number that size points at the account script rather than the agents.
Vocabulary and keyword sets attach per account, so a service runs one empathy set for a hospice client and another for a plumbing contractor, same agents, same shift.
Where the list runs out
The coaching move after a failed empathy criterion is short. Play the moment, then ask the agent what the caller was worried about. Most of the time the answer is right, which means the wording was the problem rather than the read. Coaching from a scored moment has the longer version.
Eighteen phrases and a prohibited list gets you a floor, the least an agent can say and still have acknowledged the caller. Nobody gets to good on it. A phrase library also cannot fix a hold policy that leaves callers waiting four minutes, and on a floor like that the empathy scores look fine while the complaints keep arriving.
Scoring the criterion has the same ceiling. A green check on the empathy line means the agent said the words at 00:42, and whether they meant them is the part the check cannot carry. That gap is the weak half of automated QA in this category. A scored sample of your calls scores a batch of your own recordings against your own criterion wording and returns the calls where the empathy set never appeared, which is the shortest way to learn whether the problem is your agents or your phrases.
Frequently asked questions
What are good empathy statements for call center agents?
Three that hold up across most accounts: "You should not have had to call back for this", "You reached the right number, this line goes to the on-call provider", and "Say it however it comes out, I am writing it down". Each one carries a detail the caller gave.
How do you score empathy on a QA scorecard?
Write it as an observable behavior with a condition attached, such as used an empathy statement where the caller expressed frustration, and mark it not applicable when the caller was calm. A line reading "showed empathy" produces reviewer disagreement in the first calibration session.
What should you not say to an angry customer?
Drop "I understand how you feel", which announces a script, and use the specific thing the caller told you. Also drop "I apologize for any inconvenience this may have caused", conditional enough that the caller hears no apology at all, and say "You should not have had to call back for this" in its place.
Can AI detect empathy on a call?
It detects whether a phrase from your set appeared and at what second, which is useful and narrow. Sincerity is a human judgment made by listening, and no confidence threshold turns a presence check into one.
How many empathy criteria should a scorecard have?
One. Splitting it in two multiplies reviewer disagreement on the softest thing the form measures, and the extra line rarely changes what a supervisor coaches afterward.
Sources
- Trinsic feature sheet and call notes, 2026-08-27 through 2026-09-10. Vocabulary sets, per account scorecards, and transcript timestamps verified internally.
- Aircall and StackAI on rule based criteria scoring more reliably than tone. Third party, unverified.
- Speaker attribution on single channel recordings. A phrase can be credited to the wrong party when both sides share a channel, which is why the scoring note says to spot check attribution before trusting a phrase count.
- The 5 point weighting and the phrases themselves. Editorial judgment, with no outcome data behind it.
- The 9,000 call example. Hypothetical, used to size the output rather than report a result.