diff --git a/README.md b/README.md
index 9450c0bc51..fda89c2a5e 100644
--- a/README.md
+++ b/README.md
@@ -44,6 +44,35 @@ response = client.responses.create(
 print(response.output_text)
 ```
 
+### Conversation state
+
+For multi-turn conversations with the Responses API, use `previous_response_id`
+to have the API retain context between turns.
+
+```python
+from openai import OpenAI
+
+client = OpenAI()
+
+response = client.responses.create(
+    model="gpt-5.2",
+    input="Write a haiku about recursion in programming.",
+)
+print(response.output_text)
+
+response = client.responses.create(
+    model="gpt-5.2",
+    input="Now explain it in plain English.",
+    previous_response_id=response.id,
+)
+print(response.output_text)
+```
+
+If you manually manage conversation history instead, preserve all items from
+`response.output` in their original order. Reasoning models may return reasoning
+items together with assistant messages, and filtering those items down to only
+messages can break subsequent requests.
+
 The previous standard (supported indefinitely) for generating text is the [Chat Completions API](https://platform.openai.com/docs/api-reference/chat). You can use that API to generate text from the model with the code below.
 
 ```python
diff --git a/examples/responses/conversation_state.py b/examples/responses/conversation_state.py
new file mode 100644
index 0000000000..d7e7a0cc35
--- /dev/null
+++ b/examples/responses/conversation_state.py
@@ -0,0 +1,22 @@
+from openai import OpenAI
+
+
+client = OpenAI()
+
+response = client.responses.create(
+    model="gpt-5.2",
+    input="Write a haiku about recursion in programming.",
+)
+print(response.output_text)
+
+response = client.responses.create(
+    model="gpt-5.2",
+    input="Now explain it in plain English.",
+    previous_response_id=response.id,
+)
+print(response.output_text)
+
+# If you manually manage conversation history instead of using
+# previous_response_id, append response.output items in order. Reasoning models
+# may return reasoning items together with assistant messages, and filtering
+# those items down to only messages can break the next request.