diff --git a/docs/why.md b/docs/why.md
new file mode 100644
index 000000000..a01857c03
--- /dev/null
+++ b/docs/why.md
@@ -0,0 +1,38 @@
+# Why?
+
+You may think that pandas and Polars are quite similar. But are they really?
+
+For example, do the following produce the same output?
+
+```python
+import pandas as pd
+import polars as pd
+
+print(3 in pd.Series([1, 2, 3]))
+print(3 in pl.Series([1, 2, 3]))
+```
+
+Try it out and see ;)
+
+How about
+```python
+df_left = pd.DataFrame({'a': [1, 2, 3], 'b': [4, 5, 6]})
+df_right = pd.DataFrame({'a': [1, 2, 3], 'c': [4, 5, 6]})
+df_left.merge(df_right, left_on='b', right_on='c', how='left')
+```
+versus
+
+```python
+df_left = pl.DataFrame({'a': [1, 2, 3], 'b': [4, 5, 6]})
+df_right = pl.DataFrame({'a': [1, 2, 3], 'c': [4, 5, 6]})
+df_left.join(df_right, left_on='b', right_on='c', how='left')
+```
+
+?
+
+There are several such subtle difference between the libraries. By having a unified,
+simple, and predictable API, you can focus on behaviour rather than on slight implementation differences.
+
+Furthermore, both pandas and Polars frequently deprecate behaviour. By having a simple, predictable, and
+mostly stable API, which is tested nightly against the most recent commits from both pandas and Polars,
+you can develop with a lot more confidence.