diff --git a/web/README.md b/web/README.md
index 2a8ec62..0a51480 100644
--- a/web/README.md
+++ b/web/README.md
@@ -15,6 +15,9 @@ which runs benchmarks on a dedicated server with various GPU models.
   - GPU (Intel, release template)
   - GPU (NVIDIA, release template)
 -->
+- The produced results (as .json or .md) should be copied into the `src-data/benchmarks` folder.
+- The [./generate-content.py script](./generate-content.py) produces `.md` files
+  in the `content` folder, so that web pages can be generated for each graph and each benchmark.
 - [Hugo](https://gohugo.io/) is used to create a homepage listing recent
   benchmarked commits, plus a single page per engine commit. This allows linking
   to individual benchmarked commit in a future-proof way.
@@ -26,10 +29,13 @@ which runs benchmarks on a dedicated server with various GPU models.
 
 ## Development
 
-- Create JSON data or fetch existing JSON data from the live website.
-- Save this JSON data to the `content` folder with the following naming convention:
-  `YYYY-MM-DD_hash.json` where `hash` is a 9-character Git commit hash of the
-  Godot build used (truncated from a full commit hash).
+- Create JSON data or fetch existing JSON data from the live website, and copy the JSON files into 
+  the `src-data/benchmarks` folder. Files should be named using format `YYYY-MM-DD_hash.json` where 
+  `hash` is a 9-character Git commit hash of the Godot build used (truncated from a full commit hash).
+  The files can also have the `.md` extension (for backward compatibility), but they should still be 
+  JSON inside.
+- Run [./generate-content.py](./generate-content.py). This should create `.md` pages in both the 
+  `content/benchmark` and the `content/graph` folders.
 - Run `hugo server`.
 
 ## Production
diff --git a/web/data/.gitkeep b/web/data/.gitkeep
new file mode 100644
index 0000000..e69de29
diff --git a/web/generate-content.py b/web/generate-content.py
index df47723..603257e 100755
--- a/web/generate-content.py
+++ b/web/generate-content.py
@@ -14,7 +14,7 @@
 
 import json
 import sys
-from os import listdir
+import os
 from os.path import isdir, isfile, join
 
 # Source data paths.
@@ -40,7 +40,7 @@ data_output_json = {
 # Fetch the list of benchmark files
 benchmark_input_filename_test = lambda f: (f.endswith(".json") or f.endswith(".md"))
 benchmarks_files = [
-    f for f in listdir(benchmarks_path) if (isfile(join(benchmarks_path, f)) and benchmark_input_filename_test(f))
+    f for f in os.listdir(benchmarks_path) if (isfile(join(benchmarks_path, f)) and benchmark_input_filename_test(f))
 ]
 
 # Add the list of benchmarks.
@@ -48,7 +48,7 @@ for f in benchmarks_files:
     json_file = open(join(benchmarks_path, f))
 
     # Extract data from filename.
-    key = f.removesuffix(".json")
+    key = f.removesuffix(".json").removesuffix(".md")
     date = key.split("_")[0]
     commit = key.split("_")[1]
 
@@ -91,12 +91,16 @@ data_file.close()
 
 # Create a .md file for each benchmark.
 benchmarks_content_path = "./content/benchmark"
+if not os.path.exists(benchmarks_content_path):
+    os.mkdir(benchmarks_content_path)
 for benchmark in data_output_json["benchmarks"]:
     filename = benchmark["date"] + "_" + benchmark["commit"] + ".md"
     open(join(benchmarks_content_path, filename), "a").close()
 
 # Create a .md file for each graph.
 graphs_content_path = "./content/graph"
+if not os.path.exists(graphs_content_path):
+    os.mkdir(graphs_content_path)
 for graph in data_output_json["graphs"]:
     filename = graph["id"] + ".md"
     open(join(graphs_content_path, filename), "a").close()
diff --git a/web/src-data/benchmarks/.gitkeep b/web/src-data/benchmarks/.gitkeep
new file mode 100644
index 0000000..e69de29