diff --git a/src/c++/perf_analyzer/docs/input_data.md b/src/c++/perf_analyzer/docs/input_data.md
index aa2448632..af2328fcd 100644
--- a/src/c++/perf_analyzer/docs/input_data.md
+++ b/src/c++/perf_analyzer/docs/input_data.md
@@ -37,9 +37,10 @@ of your model. You can select a different input data mode with the
   generates random data once per input and reuses that for all inferences
 - _zero_: Send zeros for each input.
 - directory path: A path to a directory containing a binary file for each input,
-  named the same as the input. Each binary file must contain the data required
-  for that input for a batch-1 request. Each file should contain the raw binary
-  representation of the input in row-major order.
+  named the same as the input (and optionally a binary file for each output for
+  validation, named the same as the output). Each binary file must contain the
+  data required for that input/output for a batch-1 request. Each file should
+  contain the raw binary representation of the input/output in row-major order.
 - file path: A path to a JSON file containing data to be used with every
   inference request. See the "Real Input Data" section for further details.
   [`--input-data`](cli.md#--input-datazerorandompath) can be provided multiple