diff --git a/experimental/README.md b/experimental/README.md
new file mode 100644
index 000000000..c99209b24
--- /dev/null
+++ b/experimental/README.md
@@ -0,0 +1,31 @@
+# Experimental Workloads
+
+Experimental Speedometer workloads can be used for workloads that do not
+directly meet the criteria for official workloads:
+
+-   Incubation of new workloads
+-   Edge-cases that don't work well in the official setup
+-   Testing features that aren't fully shipped in all browsers
+
+The main purpose is to have a space for low-friction contributions to
+experiment with new ideas. There is no explicit requirement to promote
+experimental workloads to the official selection.
+
+## Review Process
+
+Contributions to experimental workloads can be merged with the approval from a
+single vendor.
+
+This is the opposite from official workloads that are scored, where we require
+approval from all vendors.
+
+## Requirements
+
+-   Stored in the `/experimental` folder
+-   Disabled by default and thus do not contribute to the official score
+-   Use the `experimental` tag in the Suites declaration
+
+## Hosting
+
+Experimental workloads will not be officially hosted on Speedometer releases on
+<https://browserbench.org>, unlike the official workloads.