diff --git a/src/pcre2_compile.c b/src/pcre2_compile.c
index 38b76f697..79e98f83d 100644
--- a/src/pcre2_compile.c
+++ b/src/pcre2_compile.c
@@ -10895,8 +10895,19 @@ if ((re->overall_options & PCRE2_NO_START_OPTIMIZE) == 0)
   (these are not saved during the compile because they can cause conflicts with
   actual literals that follow). */
 
-  if (firstcuflags >= REQ_NONE)
-    firstcu = find_firstassertedcu(codestart, &firstcuflags, 0);
+  if (firstcuflags >= REQ_NONE) {
+    uint32_t assertedcuflags = 0;
+    uint32_t assertedcu = find_firstassertedcu(codestart, &assertedcuflags, 0);
+    /* It would be wrong to use the asserted first code unit as `firstcu` for
+     * regexes which are able to match a 1-character string (e.g. /(?=a)b?a/)
+     * For that example, if we set both firstcu and reqcu to 'a', it would mean
+     * the subject string needs to be at least 2 characters long, which is wrong.
+     * With more analysis, we would be able to set firstcu in more cases. */
+    if (assertedcuflags < REQ_NONE && assertedcu != reqcu) {
+      firstcu = assertedcu;
+      firstcuflags = assertedcuflags;
+    }
+  }
 
   /* Save the data for a first code unit. The existence of one means the
   minimum length must be at least 1. */
diff --git a/testdata/testinput1 b/testdata/testinput1
index 75c3b40aa..9c502f7cc 100644
--- a/testdata/testinput1
+++ b/testdata/testinput1
@@ -6677,4 +6677,10 @@ $/x
 /(?(?<=a(?:a.b|b))b).b/
     aaab
 
+/(?=a)b?a/
+    a
+
+/(?=a)b?a./
+    ab
+
 # End of testinput1 
diff --git a/testdata/testoutput1 b/testdata/testoutput1
index 9c2d78c4e..585258f6a 100644
--- a/testdata/testoutput1
+++ b/testdata/testoutput1
@@ -10536,4 +10536,12 @@ No match
     aaab
  0: ab
 
+/(?=a)b?a/
+    a
+ 0: a
+
+/(?=a)b?a./
+    ab
+ 0: ab
+
 # End of testinput1 
diff --git a/testdata/testoutput2 b/testdata/testoutput2
index e8bb3ed7f..b39664783 100644
--- a/testdata/testoutput2
+++ b/testdata/testoutput2
@@ -10747,9 +10747,9 @@ Subject length lower bound = 1
 /(?=a{3})[bcd]/Ii
 Capture group count = 0
 Options: caseless
-First code unit = 'a' (caseless)
+Starting code units: A a
 Last code unit = 'a' (caseless)
-Subject length lower bound = 2
+Subject length lower bound = 1
 
 /(abc)\1+/