diff --git a/README.md b/README.md
index 5eee90c..e6f9720 100644
--- a/README.md
+++ b/README.md
@@ -18,10 +18,13 @@ email = client.emails.send({
     "to": ["customer@example.org"],
     "subject": "Your receipt",
     "html": "<p>Thanks for your order.</p>",
+    "category": "transactional",
 })
 print(email["id"])
 ```
 
+Email is marketing unless you set `category` to `transactional`. Marketing email gets a one-click unsubscribe header and is not sent to people who unsubscribed. Set `transactional` on receipts, password resets and account alerts so a marketing opt-out does not block them.
+
 Requests and responses are plain dicts, typed with `TypedDict`s from `smtpfast.types`, so your editor and type checker know every field. A new field or value from the API never breaks your code, because the SDK does not validate responses at runtime.
 
 Every API operation is a method, named after its path: `client.domains.verify(id)`, `client.contacts.segments.add(contact_id, segment_id)`, `client.inboxes.threads.emails.reply(inbox_id, thread_id, email_id, params)`. Path parameters come first, then the request body, and query parameters are keyword arguments: `client.emails.list(limit=20, status="bounced")`.