diff --git a/src/plugins/intel_gpu/src/graph/graph_optimizer/add_required_reorders.cpp b/src/plugins/intel_gpu/src/graph/graph_optimizer/add_required_reorders.cpp
index 5f39d55342b..a11964c3878 100644
--- a/src/plugins/intel_gpu/src/graph/graph_optimizer/add_required_reorders.cpp
+++ b/src/plugins/intel_gpu/src/graph/graph_optimizer/add_required_reorders.cpp
@@ -27,11 +27,14 @@ If not than required reorder is added to the network.
 /*
 Add a reorder in between node and usr
 */
-void add_required_reorders::add_reorder(program& p, program_node* node, program_node* usr) {
+void add_required_reorders::add_reorder(program& p, program_node* node, program_node* usr, bool keep_original_dt) {
     layout reorder_layout = node->get_output_layout();
     reorder_layout.format = usr->get_output_layout().format;
     reorder_layout.data_type = usr->get_output_layout().data_type;
 
+    if (keep_original_dt)
+        reorder_layout.data_type = node->get_output_layout().data_type;
+
     auto new_reorder = std::make_shared<reorder>(node->id() + "_reorder_" + usr->id(), node->id(), reorder_layout);
     auto& new_reorder_node = p.get_or_create(new_reorder);
     new_reorder_node.set_output_layout(reorder_layout, false);
@@ -376,8 +379,16 @@ void add_required_reorders::run(program& p) {
                         continue;
                 }
 
-                if (usr->get_output_layout() != node.first->get_output_layout())
-                    add_reorder(p, node.first, usr);
+                if (usr->get_output_layout() != node.first->get_output_layout()) {
+                    // Preserve original data type to prevent Convolution input data type from changing
+                    // in the following sequence: Node(U8, unsupported format) -> Conv(FP16, bfyx).
+                    // Without this condition, inserted reorder will change Conv's input to FP16, instead of
+                    // expected U8 format.
+                    bool keep_original_dt = false;
+                    if (usr->is_type<convolution>())
+                        keep_original_dt = true;
+                    add_reorder(p, node.first, usr, keep_original_dt);
+                }
             }
         }
     }
diff --git a/src/plugins/intel_gpu/src/graph/include/pass_manager.h b/src/plugins/intel_gpu/src/graph/include/pass_manager.h
index 2ea466764d5..de9ec170096 100644
--- a/src/plugins/intel_gpu/src/graph/include/pass_manager.h
+++ b/src/plugins/intel_gpu/src/graph/include/pass_manager.h
@@ -57,7 +57,7 @@ public:
 
 private:
     void run(program& p) override;
-    void add_reorder(program& p, program_node* node, program_node* usr);
+    void add_reorder(program& p, program_node* node, program_node* usr, bool keep_original_dt = false);
 };
 
 class add_reshape_to_primitives : public base_pass {
diff --git a/src/plugins/intel_gpu/tests/unit/passes/add_required_reorders_test.cpp b/src/plugins/intel_gpu/tests/unit/passes/add_required_reorders_test.cpp
index 75a955155ed..cb65717800d 100644
--- a/src/plugins/intel_gpu/tests/unit/passes/add_required_reorders_test.cpp
+++ b/src/plugins/intel_gpu/tests/unit/passes/add_required_reorders_test.cpp
@@ -116,6 +116,37 @@ TEST(add_required_reorders, eltwise_input_reorder) {
     ASSERT_EQ(eltwise_node.get_output_layout().format, format::bfzyx);
 }
 
+TEST(add_required_reorders, prevent_input_dt_changing_for_convs) {
+    auto& engine = get_test_engine();
+
+    int input_b = 1, input_f = 16, input_y = 3, input_x = 3;
+    int output_b = input_b, output_f = 16, output_y = 6, output_x = 6;
+
+    auto input_mem = engine.allocate_memory({ {input_b, input_f, input_y, input_x}, data_types::u8, format::bs_fs_yx_bsv16_fsv32 });
+    auto input2_mem = engine.allocate_memory({ {input_b, input_f, input_y, input_x}, data_types::u8, format::bs_fs_yx_bsv16_fsv32 });
+    auto weights_mem = engine.allocate_memory({ {16, 16, 1, 1}, data_types::i8, format::bfyx });
+
+    auto input = input_layout("input", input_mem->get_layout());
+    auto input_const = data("input_const", input2_mem);
+    auto weights = data("weights", weights_mem);
+    auto eltwise1 = eltwise("eltwise1", input_info("input"), input_info("input_const"), eltwise_mode::sum);
+    auto conv1 = convolution("conv1", input_info("eltwise1"), "weights", "", 1, { 1, 1 }, { 1, 1 }, { 1, 1 }, { 2, 2 }, false);
+    auto output_reorder = reorder("reorder", input_info("conv1"), { data_types::f32, format::bfyx, { output_b, output_f, output_y, output_x } });
+
+    topology topology_test(input, input_const, eltwise1, weights, conv1, output_reorder);
+
+    ExecutionConfig config_test = get_test_default_config(engine);
+    ov::intel_gpu::ImplementationDesc conv1_impl_test = { format::bfyx, "", impl_types::ocl };
+    config_test.set_property(ov::intel_gpu::force_implementations(ov::intel_gpu::ImplForcingMap{ { "conv1", conv1_impl_test } }));
+
+    auto prog = program::build_program(engine, topology_test, config_test, false, true);
+    program_wrapper::apply_opt_pass<add_required_reorders>(*prog);
+
+    ASSERT_NE(prog, nullptr);
+    ASSERT_TRUE(prog->has_node("conv1"));
+    ASSERT_EQ(prog->get_node("conv1").get_input_layout(0).data_type, data_types::u8);
+}
+
 TEST(add_required_reorders, skip_adding_reorder_batch_axis_padding) {
     auto& engine = get_test_engine();