AWS SES Soft Bounces: When Out-of-Office Replies Block Valid Email Addresses
AWS SES can report out-of-office replies as soft bounces. Learn how repeated-bounce rules block valid addresses and how to handle these events safely.
An email gets delivered. The recipient’s out-of-office responder replies. Amazon SES reports a soft bounce.
Your application handles it sensibly: a temporary problem should not permanently block an address, so you keep sending.
Then it happens again. And again.
A rule designed to protect your sender reputation sees repeated soft bounces and marks the address as undeliverable. Future emails stop.
The mailbox was receiving your messages the entire time.
We encountered this in SendOps. The individual decisions looked reasonable, but together they produced the wrong outcome. Here is what happened, how to investigate it in your own AWS SES integration, and what a better bounce policy needs to account for.
Why AWS SES reports out-of-office replies as soft bounces
AWS explicitly documents this behavior: an automatic response, such as an out-of-office message, can produce a bounce notification with:
bounceType: "Transient"bounceSubType: "General"
AWS also states that SES excludes automatic responses when calculating your account’s bounce rate. Receiving a bounce notification therefore does not necessarily mean SES considers it a reputation-damaging delivery failure. AWS bounce notification documentation
There is an important distinction here. An out-of-office reply describes the recipient’s availability. A conventional soft bounce describes a temporary delivery problem. SES can expose both through the soft-bounce notification path.
AWS’s monitoring documentation confirms that a soft bounce can occur even when a message reaches the inbox successfully, typically because of an out-of-office reply. AWS sending activity documentation
If your application treats every Transient notification as evidence of failed delivery, its interpretation is already broader than the underlying evidence.
How a reasonable repeated-soft-bounce rule went wrong
Our investigation began with a customer whose application had stopped emailing a working address.
SendOps had classified the address as repeated_transient. The customer’s application synchronized our undeliverable list into its own suppression table and blocked future sends.
The rule was straightforward: three transient bounces within 30 days made an address eligible for the undeliverable list.
That threshold was our application policy, not an AWS SES requirement.
The recipient generated three qualifying notifications on the same day. However, the corresponding messages also had delivery events and recorded opens.
The sequence looked like this:
| Message | Delivery recorded | Auto-reply generated | Running soft-bounce count |
|---|---|---|---|
| A | Yes | Yes | 1 |
| B | Yes | Yes | 2 |
| C | Yes | Yes | 3 — threshold reached |
The recipient’s responder replied to each message. Our counter converted those responses into accumulated evidence that the address could not receive mail.
Raising the threshold would only postpone the problem. With enough messages, a responder that replies every time will eventually cross any fixed threshold.
Why debouncing alone does not solve it
There are two different forms of repetition to handle:
- The same event arrives more than once. Your event processor needs deduplication.
- Different messages each generate an auto-reply. Those are distinct events, even though they reflect the same out-of-office state.
Deduplicating webhook deliveries addresses the first problem. It does nothing for the second.
An autoresponder might limit how frequently it replies to a sender, but your delivery policy cannot depend on that. RFC 3834 recommends limiting repeated automatic responses; implementations and configurations still vary. RFC 3834: Automatic Responses to Electronic Mail
Even a perfectly deduplicated counter can count the wrong thing.
Investigating a suspected SES auto-reply bounce
Start by reconstructing the message history for the affected recipient.
Do not inspect the bounce notification in isolation. Look for the original send, delivery, bounce, and any engagement events associated with that message.
A simplified, illustrative SES event might look like this:
{
"eventType": "Bounce",
"mail": {
"messageId": "example-ses-message-id"
},
"bounce": {
"bounceType": "Transient",
"bounceSubType": "General",
"feedbackId": "example-feedback-id",
"bouncedRecipients": [
{
"emailAddress": "person@example.com"
}
]
}
}
Notice the missing bounce.reportingMTA and per-recipient diagnosticCode. Those fields carry information from delivery status notifications and may be absent. Their absence is a useful investigative signal, but it is not proof of an auto-reply. AWS event field reference
In our investigation, we examined 120 days of production history and found 746 non-permanent bounce events with neither a reporting MTA nor a diagnostic code. Every one was associated with a message that also had a delivery event.
The genuine post-delivery failures we examined, including quota and transport-rule failures, carried diagnostics.
That gave us a useful distinction in our dataset. It does not establish a universal guarantee about every receiving mail system.
A delivery event is supporting evidence
Avoid replacing one overly broad rule with another:
If a message has a Delivery event, ignore every later bounce.
SES delivery means the recipient’s mail server accepted the message. A server can accept a message and subsequently generate a genuine failure notification. AWS explicitly documents that delivery and bounce notifications can both occur for the same recipient. SNS notification ordering is also not guaranteed. AWS notification behavior
Recorded opens can help an investigation, but they should not automatically override delivery failures either. An open event is not conclusive evidence that a person read the message.
The useful question is: what does the combined evidence say about this particular bounce?
Building a better soft-bounce policy on AWS SES
You can handle this directly in your own SES integration. The key is to separate the raw event from your interpretation and the action that follows.
1. Preserve the original evidence
Retain enough information to revisit a classification:
- SES message ID, AWS account, Region, and recipient.
- Bounce type, subtype, timestamp, and feedback ID.
- Reporting MTA and recipient-level diagnostics when present.
- Related delivery events.
- The rule and rule version that produced a suppression decision.
Process each entry in bouncedRecipients separately. A message sent to several recipients can have different outcomes for each one.
For event deduplication, use a stable identity appropriate to your event source. For a bounce, the feedback ID and recipient are useful inputs. Separately, ensure a repeated-failure policy does not accidentally count multiple notifications about one failed message as several failed messages.
2. Identify likely auto-replies before counting delivery failures
This JavaScript helper identifies the narrow shape worth investigating:
function isAutoReplyCandidate(bounce, recipient) {
const hasText = value =>
typeof value === "string" && value.trim().length > 0;
const generalSoftBounce =
bounce.bounceType === "Transient" &&
bounce.bounceSubType === "General";
const undeterminedBounce =
bounce.bounceType === "Undetermined";
return (
(generalSoftBounce || undeterminedBounce) &&
!hasText(bounce.reportingMTA) &&
!hasText(recipient.diagnosticCode)
);
}
This is a candidate detector, not a complete suppression handler.
For a conservative starting policy, correlate candidates with delivery evidence for the same message and recipient before excluding them from repeated-failure counts. Keep unresolved cases distinguishable from confirmed delivery failures, and validate the heuristic against your own traffic.
Do not apply this exception to permanent bounces, complaints, or unsubscribe decisions.
3. Count qualifying failures, not every bounce-shaped event
The original rule effectively asked:
Have we received three transient bounce events in 30 days?
The better question is:
Have three distinct messages had qualifying temporary
delivery failures for this recipient within the policy window?
That requires classification before aggregation.
A mailbox-full failure may warrant a temporary pause and later reassessment. A message rejected because of its size may call for changing the message. A likely out-of-office response should not accumulate evidence that the mailbox is unreachable.
For monitoring delivery attempts that SES is still retrying, use DeliveryDelay events. A Transient bounce notification generally represents a temporary failure for which SES has stopped retrying that message. AWS event monitoring guidance
4. Recompute when more evidence arrives
Your processor should handle both of these arrival orders:
Delivery → Bounce
Bounce → Delivery
Store events and reevaluate the affected message and recipient when new information arrives. An initial lack of delivery evidence should not become a permanent assumption.
The correction we designed for SendOps subtracts the identified auto-reply events from the relevant transient and undetermined counts before applying the classification rules.
Because those counts are evaluated at read time, the corrected interpretation can apply to historical events without rewriting the original event records.
Keeping the evidence made it possible to change the conclusion.
Check which suppression list actually blocked the email
“Suppressed” can refer to several different systems:
| Layer | What it represents |
|---|---|
| SES account-level suppression list | An AWS-managed delivery gate, subject to your configured suppression settings |
| SendOps undeliverable list | An advisory classification derived from event history and rules |
| Your application’s suppression table | A decision your application enforces before sending |
For bounce-related automatic additions, SES adds hard bounces to its account-level suppression list, not soft bounces. Applications can also explicitly add addresses through the API. AWS account-level suppression documentation
In our incident, the customer’s application enforced the block after importing the SendOps classification.
That creates a second recovery problem: correcting the source classification does not necessarily remove a copied suppression record.
If your synchronization only adds addresses, a temporary classification can become an indefinite block.
Record why each suppression exists and which system owns it. When a derived reason no longer applies, reconcile that reason without clearing independent blocks such as complaints, unsubscribes, or manual restrictions.
For the broader policy, see SES suppression-list limitations. For recovery procedures, see managing the SES suppression list and hard bounces safely.
Test the outcome your rules produce
A useful regression suite should include:
| Scenario | Expected outcome |
|---|---|
| Three delivered messages, each followed by a likely auto-reply | No repeated-delivery-failure block |
| Three distinct messages with qualifying temporary failures | Apply the configured temporary-failure policy |
| Duplicate delivery of one bounce event | Count once |
| Delivery followed by a genuine failure with diagnostics | Preserve and evaluate the failure |
| Bounce arrives before its matching delivery event | Reevaluate when delivery arrives |
| Derived block is withdrawn upstream | Reconcile that block downstream |
| Recipient also has an unsubscribe or complaint block | Preserve the independent block |
These cases test whether working addresses remain reachable while real delivery failures still receive the appropriate response.
What this means for running email on AWS SES
SES gives you delivery infrastructure and event data. Your application still needs to turn that data into decisions.
Here, that meant distinguishing auto-replies from delivery failures, correlating events, correcting historical classifications, and tracing a decision into another application’s suppression table.
We hit this edge case in our own product. Sharing it matters because a rule can look sensible in code review and still produce the wrong result in production.
You can build and maintain this logic directly on SES. The investment includes the history, diagnostics, policy changes, and recovery paths around the initial handler.
SendOps provides message history, reporting, and deliverability tooling alongside its email workflows and drip campaigns. It connects to your AWS account, and email sends through your own SES account. Contacts, consent, events, and campaign content are held by SendOps. That gives teams a place to investigate these decisions while retaining their sending infrastructure. SendOps architecture and API overview
A useful question to ask of your current setup is: when a working address gets blocked, can you explain exactly why—and reverse the decision wherever it was copied?