diff --git a/klog.go b/klog.go
index 907b44cf..3ee606c6 100644
--- a/klog.go
+++ b/klog.go
@@ -849,9 +849,17 @@ func kvListFormat(b *bytes.Buffer, keysAndValues ...interface{}) {
 		case error:
 			writeStringValue(b, true, v.Error())
 		case []byte:
-			// We cannot use the simpler strconv.Quote here
-			// because it does not escape unicode characters, which is
-			// expected by one test!?
+			// In https://github.com/kubernetes/klog/pull/237 it was decided
+			// to format byte slices with "%+q". The advantages of that are:
+			// - readable output if the bytes happen to be printable
+			// - non-printable bytes get represented as unicode escape
+			//   sequences (\uxxxx)
+			//
+			// The downsides are that we cannot use the faster
+			// strconv.Quote here and that multi-line output is not
+			// supported. If developers know that a byte array is
+			// printable and they want multi-line output, they can
+			// convert the value to string before logging it.
 			b.WriteByte('=')
 			b.WriteString(fmt.Sprintf("%+q", v))
 		default:
@@ -875,7 +883,21 @@ func writeStringValue(b *bytes.Buffer, quote bool, v string) {
 		return
 	}
 
-	// Complex multi-line string, show as-is with indention.
+	// Complex multi-line string, show as-is with indention like this:
+	// I... "hello world" key>>>
+	// <tab>line 1
+	// <tab>line 2
+	//  <<<
+	//
+	// Tabs indent the lines of the value while the end of string delimiter
+	// is indented with a space. That has two purposes:
+	// - visual difference between the two for a human reader because indention
+	//   will be different
+	// - no ambiguity when some value line starts with the end delimiter
+	//
+	// One downside is that the output cannot distinguish between strings that
+	// end with a line break and those that don't because the end delimiter
+	// will always be on the next line.
 	b.WriteString(">>>\n")
 	for index != -1 {
 		b.WriteByte('\t')