diff --git a/README.md b/README.md
index f3fa6b7..7613ed5 100644
--- a/README.md
+++ b/README.md
@@ -247,6 +247,22 @@ The model can output predictions in two formats:
 }]
 ```
 
+Use `min_confidence` with span output to drop predictions below a threshold:
+
+```python
+detector.predict(
+    context=contexts,
+    question=question,
+    answer=answer,
+    output_format="spans",
+    min_confidence=0.8,
+)
+```
+
+For transformer detectors, emitted hallucination spans already have
+confidence values of at least `0.5`, so thresholds below `0.5` are no-ops;
+higher thresholds raise that floor.
+
 ### Token Format
 ```python
 [{
diff --git a/lettucedetect/detectors/rag_fact_checker.py b/lettucedetect/detectors/rag_fact_checker.py
index 18d9e0c..c5b87d0 100644
--- a/lettucedetect/detectors/rag_fact_checker.py
+++ b/lettucedetect/detectors/rag_fact_checker.py
@@ -54,6 +54,8 @@ def predict(
         :param question: Question (optional)
         :param output_format: "tokens", "spans", or "detailed"
         :param min_confidence: Drop ``"spans"`` below this confidence threshold (``[0, 1]``).
+            Does not filter ``output_format="detailed"``; detailed output returns
+            the unfiltered spans, triplets, and fact-check results.
         :param kwargs: Additional arguments
 
         :return: List of predictions in lettuceDetect format, or dict for detailed format