diff --git a/tools/protobench/README.md b/tools/protobench/README.md
index 693e335a..7a94bea9 100644
--- a/tools/protobench/README.md
+++ b/tools/protobench/README.md
@@ -1,3 +1,5 @@
+## Introduction
+
 The idea is to record the wire messages of the profiling agent and see how well they compress using different
 compressors and what the CPU impact is.
 
@@ -7,25 +9,49 @@ This will write the wire messages into the given directory. The directory will b
 You can then use the `protobench` tool to compress the wire messages and see how well they compress and how much
 CPU time it takes to compress them.
 
-To run the profiling agent, first have a receiving endpoint, e.g. `devfiler` listening on localhost:11000.
-Then run the profiling agent with the `-reporter-save-outputs-to` flag:
+### Recording wire messages
+
+Make sure you have a receiving endpoint, e.g. `devfiler` listening on localhost:11000.
+Now run the profiling agent with the `-reporter-save-outputs-to` flag:
 ```shell
 sudo ./opentelemetry-ebpf-profiler -reporter-save-outputs-to=/tmp/protobuf -collection-agent=127.0.0.1:11000 -disable-tls
 ```
 The wire messages are written to `protobuf/`, one file per message.
 
-To compress the wire messages and generate a bar chart, run the `protobench` tool:
+### Benchmark compression of wire messages
+
+In reality, the previously recorded wire messages are sent compressed over the network.
+Compression efficiency and CPU usage are important factors to consider when choosing a compression algorithm.
+
+The `protobench` tool helps to compare compression algorithms and their performance:
+- for realistic results, wire messages are compressed one-by-one
+- different compressors with different compression levels are used
+- the compression ratio and CPU usage are measured
+- the results are written as a bar chart or a CSV file
+
+To compress the wire messages and generate a bar chart, run the `protobench` tool with an output file ending in `.png`:
 ```shell
 cd tools/protobench
 go run ./... -bench-proto-dir=/tmp/protobuf -output-file=results.png
 ```
 If you don't see any errors, the tool will generate a PNG file with a bar chart showing the compression ratio and
-compression time for each compressor.
+CPU usage for each compressor/level.
 The extension `.csv` can be used to generate a CSV file with the raw data instead of a PNG file.
 No `-output-file` flag will display the results in the terminal.
 
 Of course, you can also use the `protobench` tool to compare compression of any other files.
 
+### Reproducible reporter outputs
+
+The profiling agent supports recording and replaying reporter inputs with the `-reporter-record-inputs-to` and
+`-reporter-replay-inputs-from` flags.
+
+Replaying in combination with `-reporter-save-outputs-to` generates a (nearly) reproducible set of wire messages,
+which be easily compared with the `protobench` tool. This can be useful for comparing different implementations of
+the reporter or the wire protocol.
+
+Generated inputs can be shared for CI, benchmarking, development, testing or debugging purposes.
+
 ### Example PNG output
 
 ![Example output](example.png)