First, find out what arrived, from which system and when
Ask for the message itself, or find it in the send log, before changing anything. Four facts locate most leaks. What it was: a newsletter, a flow message, a sales follow-up or a receipt. Who sent it: the From address, and the sending system named in the message headers or the log, because a business with an email platform, a CRMThe system that stores an organisation's contacts, companies, deals and interactions, and the processes built around it., a shop and a booking tool has several systems able to send as its name. When the person opted out, and how: through the link in an email, a one-click request from their mail app, a preference centreA page where a subscriber chooses which messages, topics and channels they receive and updates their own details, instead of an all-or-nothing unsubscribe., a reply, a phone call or a message on social media. When the message was sent, and when that send or flow was built.
Then rule out the case that is not a fault. An unsubscribe from marketing does not stop service messages such as order confirmations, delivery updates and password resets; the suppression list entry says so plainly, and the promotional versus transactional comparison explains where the line sits. If the message was a genuine receipt, the opt-out worked. If it was a receipt carrying offers, or a newsletter, the opt-out failed, and the rest of this page applies.
With the message in hand, ask why it was sent, and keep asking. The five whys method fits this problem well, because each answer names a system: the newsletter went out because the platform listed the person as subscribed; the platform listed them as subscribed because the overnight sync wrote that value; the sync wrote it because the CRM never received the unsubscribe; and so on until the answer is a rule someone can change. Stop at the first answer that names a rule rather than a person.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| The person used the link or their mail app's unsubscribe, and the platform shows no opt-out at all | A test unsubscribe through the link and through the one-click headers on a recent send | The opt-out was never recorded: a one-click request or unsubscribe link that failed | Repair the link and the one-click endpoint so either records the opt-out at once |
| The person says they asked by reply, by phone or in a message | The opt-out log, and the person's status in each tool | An opt-out made by another channel that never reached the email tools | Log it with its scope and apply it in every tool that holds the address |
| The platform shows the opt-out, then later shows the person as subscribed again | The field history around the sync times, and the person's status in the CRM | A sync or another sending tool still treats the person as reachable | Make the most restrictive consent value win in both directions |
| The person opted out partway through a sequence or after a send was scheduled | Whether the flow or send checks suppression before each message or only at entry | A flow or scheduled send that checked suppression only when the person entered | Check the suppression list before every send, not only at entry |
| The opt-out is on one record, and the message went to another record or a newly imported one | Every record holding the address, and the source and date of the one that was mailed | The person held twice, or back through an import after being deleted | Match imports against the suppression list; merge records so the opt-out wins |
| Leaks began just after a move to a new email platform | Whether the old suppression files were loaded, and whether the old platform still processes old links | Suppressions or old unsubscribe links lost in the move | Load the old suppressions now and capture every opt-out the old platform received |
| The message came from the order or account system and carried offers | The content and the sending stream of the message | Marketing content sent through the service stream | Take the promotion out of service mail; send marketing only from the marketing stream |
Fix, verify and prevent each cause
An unsubscribe that was never recorded
An unsubscribe has two layers, and either can fail. The visible layer is the link in the body, which should work in one or two clicks without a login and confirm the result. The header layer lets a mail app offer its own unsubscribe button: the one-click unsubscribe standard has the app send a single request that the sender must process with no further step from the person, and Google's email sender guidelines document one-click unsubscribe and a visible link for bulk senders, with requests processed within the window stated on that page. The list building and consent guide sets out both layers and the three checks that matter. A link pointing at an old page, a landing page that asks the person to log in, or an endpoint that returns a page but records nothing all leave the person subscribed while they believe otherwise.
Fix: repair the link and the one-click endpoint so either one removes the person from marketing at once.
Verify: unsubscribe a test address through each route and confirm its status changes before the next send.
Prevent: repeat that test after any change to templates, sending domains or the platform's settings.
An opt-out that came by reply, phone or message
People do not always use the link. They reply "please stop", tell a member of staff on the phone or write to the business on social media, and those requests land in an inbox no system reads. The marketing opt-out request procedure handles exactly this: log the request with its date, channel and scope, find the person in every tool the consent map lists, record the opt-out in the system of record and on the suppression list, and check that it reached each sending tool. When the person's words are broad, the procedure records the broad scope; "stop emailing me" is not a request to leave one newsletter.
Fix: run the procedure for each request found, including old ones in the shared inbox.
Verify: each person shows as opted out in every sending tool, and the log has an entry for each.
Prevent: give the reply address, the phone team and the social inbox one route into that log, and read the reply inbox as part of every send.
A sync or another tool that still says yes
Most businesses send from more than one tool, and an opt-out recorded in one does not reach the others by itself. The guide to syncing the CRM and the marketing platform sets the rule that settles this: for consent fields, the most restrictive value wins, whichever system recorded it and whenever. A sync that lets an older "subscribed" value overwrite a newer opt-out turns every unsubscribe into a temporary one, and a sales tool or second platform that never received the opt-out keeps mailing as before. The consent map lists every tool that holds the address and how a withdrawal travels to it.
Fix: set the consent fields to most-restrictive-wins in both directions, and connect or update every tool on the map that can send.
Verify: an opt-out made in each tool appears in all the others after the next sync, and a reachable value from the CRM no longer overwrites it.
Prevent: keep a seeded contact that has opted out in both systems and check it after every change to the sync.
A send that checked only at the start
A flow decides who enters when the trigger fires; a campaign builds its audience when it is scheduled. If either checks the suppression listA list of addresses or contacts excluded from every send, such as unsubscribes, hard bounces and spam complaints. only at that moment, a person who opts out afterwards still receives the rest of the sequence or the scheduled send. The automation governance guide asks for one global suppression list that every workflow checks at entry and before every send, because status changes mid-sequence.
Fix: add a suppression check before each message in every flow, and rebuild scheduled sends' audiences at send time rather than when they were drafted.
Verify: a test contact that opts out halfway through a flow receives nothing further.
Prevent: make the before-each-send check part of every new flow's launch review; when a flow stops sending altogether instead, the flow diagnosis covers it.
A second record, or a contact that came back
The opt-out can be correct on one record while the message goes to another. A person who exists twice, under the same address in two places or as a customer record and a newsletter record, can be suppressed on one and reachable on the other; the CRM fundamentals guide explains how duplicates arise and why merges keep the most recent consent value. A person deleted rather than suppressed has no trace left, so the next event list, spreadsheet or integration brings them back as new. The tutorial on importing contact lists matches every file against the suppression list before anything is loaded.
Fix: suppress rather than delete, merge duplicate records so the opt-out survives, and match each import against the suppression list.
Verify: the address now appears once, as suppressed, and a test import containing it leaves it suppressed.
Prevent: keep the email list hygiene procedure's rule that opted-out contacts stay on the suppression list indefinitely.
Suppressions left behind in a platform move
A move to a new email platform is when suppression lists most often disappear: the active contacts are exported, the people who asked to leave sit in a separate file, and the new platform sees them as ordinary subscribers. The tutorial on migrating to a new email platform loads the suppression files before a single contact, and keeps the old platform processing unsubscribe links from emails sent before the cutover until the final export, because those links still point at it.
Fix: export the old platform's suppressions and every opt-out it received after the cutover, and load them as a global suppression in the new one.
Verify: a sample of known opt-outs shows as suppressed in the new platform.
Prevent: follow the sending domain setup checklist's line on carrying suppressions across for every future move, and keep a single list across markets, as the guide to email across countries recommends.
Marketing sent through the service stream
Service messages are exempt from a marketing opt-out because of what they contain. Add a promotion to an order confirmation, or send a newsletter through the system that sends receipts, and the message becomes marketing to a person who said no. The list building guide puts it directly: an unsubscribed person should never receive marketing dressed as a receipt, and the comparison of promotional and transactional email explains why promotional content changes how a message is treated. A transactional emailAn email triggered by a person's own action or account, such as a receipt, password reset or shipping notice, rather than a marketing send. can carry help with the order, not offers.
Fix: remove promotional blocks from service templates and send marketing only from the marketing stream.
Verify: the service templates contain only the information the person needs about what they did.
Prevent: decide the stream for every new message type before it is built.
Prevention: a seeded opt-out that every change has to pass
Every cause above is found the same way: by a person who opted out and still received mail. The cheapest prevention is to be that person first. Keep test contacts that have unsubscribed through the link, through the one-click headers, by reply and by phone, held in every tool that can send, and check them on a schedule and after every change that touches lists, syncs, flows or sending. For a change to a sync or connector, the CRM sync change checklist makes reading these contacts in every tool part of the gate before the change goes live.
| When | Check | What it catches |
|---|---|---|
| After every send | Unsubscribes recorded against the send, and replies asking to stop | Opt-outs that the link or the reply inbox failed to record |
| After any sync, integration or import change | The seeded opted-out contacts in every tool | A reachable value overwriting an opt-out; deleted people returning |
| Before a flow goes live or is edited | A test contact that opts out partway through | Suppression checked only at entry |
| Monthly | Opt-out counts in each tool, side by side | Tools whose suppression has drifted from the rest |
| Before and after a platform move | Old suppressions loaded; old links still processed | Opt-outs left behind in the move |
Read the monthly comparison with care. The counts in each tool will rarely match exactly, but the gap should be stable; a gap that grows means one tool has stopped learning about opt-outs. Rising complaintsShare of delivered emails that recipients marked as spam. while unsubscribesShare of delivered emails that led the recipient to opt out. hold steady, which the spam complaint rate entry reads as people failing to find the unsubscribe, can equally mean an unsubscribe that does not work. Keep the consent recordThe stored evidence of a marketing permission: the channel, the wording the person saw, the date and the source of each opt-in, kept with the contact record. for each opt-out, with its source, date and scope, so the next person to investigate can see what was asked and when.
When a leak has already reached people, handle them as the tutorial on fixing a mistake in a sent email describes: stop the send, re-apply the suppression, record what happened and do not email them again, not even to apologise. The email troubleshooting section holds the neighbouring diagnoses, including the unsubscribe spike page for the opposite problem, many people leaving at once. None of this is legal advice; whoever handles data protection should hear about every leak, and the regulators' own guidance, such as the FTC's CAN-SPAM guide and the UK regulator's direct marketing guidance in this page's sources, is the reference for what the rules require.
A stationery shop finds the copy that forgot (illustrative)
Haskivelt Stationery
Haskivelt Stationery, an invented online stationery shop, sends a monthly newsletter to 9,400 subscribers. In March a customer replies angrily that she unsubscribed in January, and a second says she asked by reply in February to stop. Tracing the first message shows the email platform held the first customer as subscribed again: she had unsubscribed through the link, but the shop's CRM never received the change, and the nightly sync copied its older value back over the opt-out three days later. A search of the platform's history finds 37 other people whose opt-outs were reversed the same way over 10 weeks, and the shared reply inbox holds 12 unlogged requests to stop. The shop sets the consent field to most-restrictive-wins in both directions, applies all 49 opt-outs in the CRM and the platform, logs reply and phone requests through its opt-out procedure, adds four seeded opted-out addresses that it checks after every sync change, and does not email the affected customers again. The April newsletter's send log shows none of the 49 addresses, and complaints fall from 9 on the March send to 2.
Common mistakes
- Treating every received email as a failure. A receipt or password reset is not marketing; check what the message contained before changing anything.
- Fixing the one person who complained. The same rule that mailed them mailed others; trace the cause and find everyone it affected.
- Deleting people to make them go away. A deleted contact returns in the next import; suppress instead.
- Trusting one tool's status. The email platform can show an opt-out while the CRM or a sales tool still sends; check every tool that can send.
- Checking suppression only at entry. People opt out halfway through sequences and after sends are scheduled.
- Apologising by email. Another message to someone who asked for none adds to the problem; record it and stop.
