diff --git a/NEWS.md b/NEWS.md
index 0fd950dcf6..019a4c2e3f 100644
--- a/NEWS.md
+++ b/NEWS.md
@@ -2,6 +2,8 @@
 
 ### Bug fixes
 
+* Fixed regression where `NULL`-aesthetics contributed to plot labels too 
+  insistently. Now they contribute only as fallback labels (@teunbrand, #6616)
 * Fixed regression where `draw_key_rect()` stopped using `fill` colours 
   (@mitchelloharawild, #6609).
 * Fixed regression where `scale_{x,y}_*()` threw an error when an expression
diff --git a/R/aes-evaluation.R b/R/aes-evaluation.R
index e29d0c5d25..2a7a69d382 100644
--- a/R/aes-evaluation.R
+++ b/R/aes-evaluation.R
@@ -350,7 +350,10 @@ strip_stage <- function(expr) {
 make_labels <- function(mapping) {
   default_label <- function(aesthetic, mapping) {
     # e.g., geom_smooth(aes(colour = "loess")) or aes(y = NULL)
-    if (is.null(mapping) || is.atomic(mapping)) {
+    if (is.null(mapping)) {
+      return(structure(aesthetic, fallback = TRUE))
+    }
+    if (is.atomic(mapping)) {
       return(aesthetic)
     }
     mapping <- strip_stage(mapping)
diff --git a/tests/testthat/test-aes-calculated.R b/tests/testthat/test-aes-calculated.R
index 2d389106cf..297ac0b171 100644
--- a/tests/testthat/test-aes-calculated.R
+++ b/tests/testthat/test-aes-calculated.R
@@ -47,7 +47,11 @@ test_that("make_labels() deparses mappings properly", {
   expect_match(x_lab, "...$")
   # if the mapping is a literal or NULL, the aesthetics is used
   expect_identical(make_labels(aes(x = 1)), list(x = "x"))
-  expect_identical(make_labels(aes(x = NULL)), list(x = "x"))
+  # NULL labels should only be used as fallback labels
+  expect_identical(
+    make_labels(aes(x = NULL)),
+    list(x = structure("x", fallback = TRUE))
+  )
 })
 
 test_that("staged aesthetics warn appropriately for duplicated names", {
diff --git a/tests/testthat/test-aes.R b/tests/testthat/test-aes.R
index b0922383cc..c3786e3975 100644
--- a/tests/testthat/test-aes.R
+++ b/tests/testthat/test-aes.R
@@ -103,7 +103,8 @@ test_that("quosures are squashed when creating default label for a mapping", {
 test_that("labelling doesn't cause error if aesthetic is NULL", {
   p <- ggplot(mtcars) + aes(x = NULL)
   labels <- ggplot_build(p)@plot@labels
-  expect_identical(labels$x, "x")
+  # NULL labels should only be used as fallback labels
+  expect_identical(labels$x, structure("x", fallback = TRUE))
 })
 
 test_that("aes standardises aesthetic names", {