diff --git a/CHANGELOG.md b/CHANGELOG.md
index a71113f..9b2e1bb 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -5,6 +5,9 @@
   - Better subclassing support: Determine classes dynamically,
     so that methods like str() are aware when our types are subclassed.
   - Removed no-op check and extra list allocation during chop()
+  - Fixed merge_neighbors() TypeError on non-numeric endpoints
+    (e.g. datetime): the default distance is now inferred from
+    the endpoint type instead of being the int 1
 - Maintainers:
   - Load issue4 test data via JSON to save on compilation and collection time.
 
diff --git a/intervaltree/intervaltree.py b/intervaltree/intervaltree.py
index e273127..fdf00a4 100644
--- a/intervaltree/intervaltree.py
+++ b/intervaltree/intervaltree.py
@@ -777,7 +777,7 @@ def merge_neighbors(
         self,
         data_reducer=None,
         data_initializer=None,
-        distance=1,
+        distance=None,
         strict=True,
     ):
         """
@@ -806,12 +806,23 @@ def merge_neighbors(
         will not be merged. If strict is False, both neighbors and overlapping
         intervals are merged.
 
+        If distance is None (the default), it is inferred from the endpoints:
+        1 for numeric endpoints, or the zero of the endpoint type (e.g.
+        timedelta(0) for datetime endpoints) so that only touching or
+        overlapping intervals are merged.
+
         Completes in O(n*logn) time.
         """
         if not self:
             return
 
         sorted_intervals = sorted(self.all_intervals)  # get sorted intervals
+        zero = sorted_intervals[0].begin - sorted_intervals[0].begin
+        if distance is None:
+            # type-aware default: numeric endpoints keep the historical
+            # default of 1, other endpoint types (e.g. datetime) default
+            # to zero so only touching/overlapping intervals are merged
+            distance = 1 if isinstance(zero, (int, float)) else zero
         merged = []
         # use mutable object to allow new_series() to modify it
         current_reduced = [None]
@@ -832,7 +843,7 @@ def new_series():
                 lower = merged[-1]
                 margin = higher.begin - lower.end
                 if margin <= distance:  # should merge
-                    if strict and margin < 0:
+                    if strict and margin < zero:
                         new_series()
                         continue
                     else:
diff --git a/test/intervaltree_methods/restructure_test.py b/test/intervaltree_methods/restructure_test.py
index 9d178b3..5ce2351 100644
--- a/test/intervaltree_methods/restructure_test.py
+++ b/test/intervaltree_methods/restructure_test.py
@@ -19,6 +19,7 @@
 limitations under the License.
 """
 from __future__ import absolute_import
+from datetime import datetime, timedelta
 from intervaltree import Interval, IntervalTree
 import pytest
 from test import data
@@ -459,9 +460,62 @@ def reducer(old, new):
     ]
 
 
-# -----------------------------------------------------------------------------
+def test_merge_neighbors_non_numeric_endpoints():
+    # regression: merge_neighbors crashed with TypeError on non-numeric
+    # endpoints (e.g. datetime) because the default distance was the
+    # int 1 and the strict check compared margin against the int 0
+    def reducer(old, new):
+        return "%s, %s" % (old, new)
+
+    # touching datetime intervals merge by default
+    t = IntervalTree.from_tuples([
+        (datetime(2026, 1, 1), datetime(2026, 1, 2), '[1/1,1/2)'),
+        (datetime(2026, 1, 2), datetime(2026, 1, 3), '[1/2,1/3)'),
+    ])
+    t.merge_neighbors(data_reducer=reducer)
+    t.verify()
+    assert len(t) == 1
+    for begin, end, _data in t.items():
+        assert begin == datetime(2026, 1, 1)
+        assert end == datetime(2026, 1, 3)
+        assert _data == '[1/1,1/2), [1/2,1/3)'
+
+    # a gap wider than the zero default stays unmerged
+    t = IntervalTree.from_tuples([
+        (datetime(2026, 1, 1), datetime(2026, 1, 2), '[1/1,1/2)'),
+        (datetime(2026, 1, 3), datetime(2026, 1, 4), '[1/3,1/4)'),
+    ])
+    t.merge_neighbors()
+    t.verify()
+    assert len(t) == 2
+
+    # an explicit distance of the endpoint type still merges gaps
+    t = IntervalTree.from_tuples([
+        (datetime(2026, 1, 1), datetime(2026, 1, 2), '[1/1,1/2)'),
+        (datetime(2026, 1, 2, 2), datetime(2026, 1, 3), '[1/2,1/3)'),
+    ])
+    t.merge_neighbors(data_reducer=reducer, distance=timedelta(hours=3))
+    t.verify()
+    assert len(t) == 1
+    for begin, end, _data in t.items():
+        assert begin == datetime(2026, 1, 1)
+        assert end == datetime(2026, 1, 3)
+        assert _data == '[1/1,1/2), [1/2,1/3)'
+
+    # overlapping datetime intervals under strict mode keep the split
+    # behavior instead of raising
+    t = IntervalTree.from_tuples([
+        (datetime(2026, 1, 1), datetime(2026, 1, 4), 'a'),
+        (datetime(2026, 1, 3), datetime(2026, 1, 5), 'b'),
+    ])
+    t.merge_neighbors(strict=True)
+    t.verify()
+    assert len(t) == 2
+
+
+# ------------------------------------------------------------------------------
 # CHOP
-# -----------------------------------------------------------------------------
+# ------------------------------------------------------------------------------
 def test_chop():
     t = IntervalTree([Interval(0, 10)])
     t.chop(3, 7)