diff --git a/CHANGES.rst b/CHANGES.rst
index 51d3df8a..6c3ce4b6 100644
--- a/CHANGES.rst
+++ b/CHANGES.rst
@@ -25,6 +25,10 @@ development version
   empty read. By @rhpvorderman.
 * :issue:`824`: Fix: discard_(un)trimmed values were always ``none`` in JSON
   report.
+* An index (that speeds up matching of many anchored adapters such as barcodes)
+  is now created for up to three allowed errors (instead of two). Since this can
+  be slow and memory intensive, particularly when indels are enabled,
+  a message is printed that suggests using ``--no-indels`` or ``--no-index``.
 * Dropped support for Python 3.8.
 * Added support for Python 3.13.
 
diff --git a/doc/guide.rst b/doc/guide.rst
index b64bb594..0fda2049 100644
--- a/doc/guide.rst
+++ b/doc/guide.rst
@@ -1916,32 +1916,37 @@ To demultiplex this type of data, the
 
 .. _speed-up-demultiplexing:
 
-Speeding up demultiplexing
---------------------------
+Speeding up demultiplexing/adapter indexing
+-------------------------------------------
 
 Finding many adapters/barcodes simultaneously (which is what demultiplexing in Cutadapt is about),
 can be sped up tremendously by using the right options since Cutadapt will then be able to create an
-index of the barcode sequences instead of checking for each barcode separately. Currently, the
-following conditions need to be met in order for index creation to be enabled:
+index of the barcode sequences instead of checking for each barcode separately.
+The following conditions need to be met for index creation to be enabled:
 
 * The barcodes/adapters must be anchored:
   For 5’ adapters, use ``-g ^ADAPTER`` or ``-g ^file:adapters.fasta``.
   For 3’ adapters, use ``-a ADAPTER$`` or ``-a file$:adapters.fasta``.
-* The maximum error rate (``-e``) must be set such that at most 2 errors are allowed:
-  Use ``-e 0``, ``-e 1`` or ``-e 2``.
+* The maximum error rate (``-e``) must be set such that at most 3 errors are allowed:
+  Use ``-e 0`` up to ``-e 3``.
 * No IUPAC wildcards must be used in the barcode/adapter. Also, you cannot use the option
   ``--match-read-wildcards``.
 
+Index creation is significantly faster and uses less memory if ``--no-indels`` is provided.
+
+Index creation can be disabled with ``--no-index``.
+
 An index will be built for all the adapters that fulfill these criteria if there are at least two
 of them. You can provide additional adapters/barcodes, and they will just not be included in the
-index. Whether an index is created or not should not affect the results, only how fast you get them.
+index. Instead, they will be searched for one by one.
+Whether an index is created or not should not affect the results, only how fast you get them.
 
 To see whether an index is created, look for a message like this in the first few lines of
 Cutadapt’s output::
 
     Building index of 23 adapters ...
 
-Hopefully some of the above restrictions will be lifted in the future.
+
 
 .. versionadded:: 1.15
    Demultiplexing of paired-end data.
@@ -1953,6 +1958,8 @@ Hopefully some of the above restrictions will be lifted in the future.
    An index can be built even when indels are allowed (that is, ``--no-indels``
    is no longer required).
 
+.. versionadded:: 5.0
+   An index is created up to three (instead of two) allowed errors.
 
 Demultiplexing paired-end reads in mixed orientation
 ----------------------------------------------------
diff --git a/src/cutadapt/adapters.py b/src/cutadapt/adapters.py
index b51aef84..f3d1415c 100644
--- a/src/cutadapt/adapters.py
+++ b/src/cutadapt/adapters.py
@@ -4,6 +4,7 @@
 The ...Adapter classes are responsible for finding adapters.
 The ...Match classes trim the reads.
 """
+
 import logging
 from enum import IntFlag
 from collections import defaultdict
@@ -1319,7 +1320,7 @@ def _accept(cls, adapter: SingleAdapter, prefix: bool):
         if adapter.adapter_wildcards:
             raise ValueError("Wildcards in the adapter not supported")
         k = int(len(adapter) * adapter.max_error_rate)
-        if k > 2:
+        if k > 3:
             raise ValueError("Error rate too high")
 
     @classmethod
@@ -1338,7 +1339,21 @@ def is_acceptable(cls, adapter: SingleAdapter, prefix: bool):
 
     def _make_index(self) -> Tuple[List[int], "AdapterIndexDict"]:
         start_time = time.time()
+        max_k = max(
+            (
+                int(adapter.max_error_rate * len(adapter.sequence))
+                for adapter in self._adapters
+                if adapter.indels
+            ),
+            default=0,
+        )
         logger.info("Building index of %s adapters ...", len(self._adapters))
+        if max_k == 3:
+            logger.info(
+                "Three errors and indels allowed for at least one of the adapter sequences: "
+                "Indexing could take long and use a lot of memory. "
+                "If this becomes a problem, try --no-indels and/or --no-index."
+            )
         index: Dict[str, Tuple[SingleAdapter, int, int]] = dict()
         lengths = set()
         has_warned = False
diff --git a/tests/test_adapters.py b/tests/test_adapters.py
index 3976bfa4..f6fb8dc7 100644
--- a/tests/test_adapters.py
+++ b/tests/test_adapters.py
@@ -516,8 +516,8 @@ def test_indexed_too_high_k():
     with pytest.raises(ValueError) as e:
         IndexedPrefixAdapters(
             [
-                PrefixAdapter("ACGTACGT", max_errors=3, indels=False),
-                PrefixAdapter("AAGGTTCC", max_errors=2, indels=False),
+                PrefixAdapter("ACGTACGT", max_errors=4, indels=False),
+                PrefixAdapter("AAGGTTCC", max_errors=3, indels=False),
             ]
         )
     assert "Error rate too high" in e.value.args[0]