Every business messaging platform reports the same success metric: delivered. The message left, the network accepted it, nothing bounced. Green tick.
iOS 26 broke the assumption underneath that metric.
Two walls, not one
Until recently there was one thing to worry about: the sending account getting throttled or suspended. That failure is loud. Sends start erroring, and you know.
iOS 26 added a second, quieter layer — an on-device classifier that sorts incoming messages. A message it considers suspect goes into a separate area with no notification, no badge, no sound.
The critical detail: the message is still delivered. It reached the device. Every metric upstream of the recipient's screen says it worked.
| | Account suspension | On-device filtering | |---|---|---| | Where it happens | Apple's backend | Recipient's phone | | What you see | Sends fail | Sends succeed | | Detectable? | Immediately | Barely |
The second one is more dangerous for a business precisely because it looks like success.
What actually got added
Two related but separate features, which vendor write-ups routinely conflate:
Spam detection. On-device machine learning files suspected junk into a Spam folder. This is automatic.
Screen Unknown Senders. Messages from numbers the user has never interacted with go into a separate Unknown Senders area rather than the main inbox.
A correction, because several vendor blogs state the opposite: Screen Unknown Senders is off by default in iOS 26. It is on only for users who had "Filter Unknown Senders" enabled before upgrading. Apple's own documentation is explicit about this.
So the situation is less catastrophic than some coverage suggests. But the classifier runs regardless of that setting, and the direction of travel is clear: Apple is steadily reducing how much unsolicited messaging reaches a person's attention.
Notably, Apple's own framing acknowledges the collateral damage — a delivery driver, a new doctor's office, a repair service are all legitimate senders the user has never interacted with. Time-sensitive messages such as verification codes get notification treatment even from unknown senders. The system is trying to be discriminating rather than blunt.
What this means for your metrics
If you track delivery only, you cannot tell a healthy channel from a filtered one. Both look identical.
The one proxy available is the delivery-receipt rate: of the messages you sent, how many ever came back with a delivery or read confirmation.
That ratio behaves usefully. Healthy conversations produce receipts. Filtered messages tend not to — the recipient never opens the folder, so nothing comes back. When receipts collapse while sends keep succeeding, something between your API call and a human's attention is absorbing your messages.
It is not a perfect signal. It is the only one you have.
We alert on it internally: if the delivered-to-sent ratio on a line drops below 50% over a meaningful sample in 24 hours, that line gets flagged even though every send "succeeded".
What reduces the odds of being filtered
The classifier is trying to answer one question: does this look like something the recipient wants?
Be someone they have talked to. The strongest signal available. A message in an existing thread is not an unknown sender by any definition. This is the same reason we only send iMessage to people who have messaged first — it happens to solve the filtering problem and the suspension problem with one rule.
Get into their contacts. A saved contact is not an unknown sender. Ask customers to save your number when they first text you; it costs nothing and it is durable.
Do not write like a scam. The classifier was trained on real spam. Urgency, prize language, shortened links to unfamiliar domains, all-caps — these are features of the training set. Write like a person.
Use a link domain that is yours. A generic shortener carries whatever reputation every other user of that shortener has earned. Your own domain carries only yours.
Keep the first message plausible. The first message from a new number is judged with the least context. Making it short and specific beats making it promotional.
The strategic read
Apple is not filtering business messages. It is filtering unwanted messages, and it keeps getting better at telling the difference.
Every measure that works against filtering is the same measure that works against suspension: be wanted, be expected, be in an existing conversation. There is no clever exception, and the platforms selling one are selling access to a channel that is closing on exactly the traffic they help you send.
The businesses that will still be reaching iPhone users in two years are the ones whose customers chose to be in the conversation.
FAQ
Questions
Is Screen Unknown Senders on by default in iOS 26?
No. It is off by default, and on only for users who had 'Filter Unknown Senders' enabled before upgrading. The separate spam classifier does run regardless, so plan for filtering without assuming every recipient has the setting enabled.
How do I know if my messages are being filtered?
You largely cannot see it directly. Watch the ratio of delivery receipts to sends — filtered messages tend not to generate them, because the recipient never opens the folder. A collapse in that ratio while sends keep succeeding is the signature.
Does being in someone's contacts help?
Yes, materially. A saved contact is not an unknown sender. Asking customers to save your number when they first message you is one of the cheapest durable improvements available.
Does filtering affect iMessage and SMS equally?
The on-device classifier sees both, since it runs on the recipient's device rather than in the network. The protective factor is not the channel — it is whether an existing conversation is there.