diff --git a/services/comms/scripts/atlasbot/bot.py b/services/comms/scripts/atlasbot/bot.py
index eca5fef..7f22ad5 100644
--- a/services/comms/scripts/atlasbot/bot.py
+++ b/services/comms/scripts/atlasbot/bot.py
@@ -163,6 +163,8 @@ CLUSTER_HINT_WORDS = {
     "documentation",
     "docs",
     "playbook",
+    "utilization",
+    "usage",
     "grafana",
     "victoria",
     "prometheus",
@@ -561,8 +563,10 @@ def _has_any(text: str, phrases: tuple[str, ...]) -> bool:
 def _detect_operation(q: str) -> str | None:
     if _has_any(q, OPERATION_HINTS["top"]):
         return "top"
+    if _has_any(q, OPERATION_HINTS["bottom"]):
+        return "bottom"
     for op, phrases in OPERATION_HINTS.items():
-        if op == "top":
+        if op in ("top", "bottom"):
             continue
         if _has_any(q, phrases):
             return op
@@ -1353,6 +1357,11 @@ def snapshot_metric_answer(
         failed = metrics.get("pods_failed")
         succeeded = metrics.get("pods_succeeded")
         status_terms = ("running", "pending", "failed", "succeeded", "completed")
+        if ("most pods" in q or ("most" in q and "pod" in q and "node" in q)) and not nodes_in_query:
+            return _format_confidence(
+                "I don't have per-node pod counts in the snapshot.",
+                "medium",
+            )
         if "total" in q or "sum" in q:
             values = [v for v in (running, pending, failed, succeeded) if isinstance(v, (int, float))]
             if values:
@@ -1363,13 +1372,13 @@ def snapshot_metric_answer(
                 return _format_confidence(f"Pods not running: {sum(parts):.0f}.", "high")
         if sum(1 for term in status_terms if term in q) > 1:
             parts = []
-            if running is not None:
+            if "running" in q and running is not None:
                 parts.append(f"running {running:.0f}")
-            if pending is not None:
+            if "pending" in q and pending is not None:
                 parts.append(f"pending {pending:.0f}")
-            if failed is not None:
+            if "failed" in q and failed is not None:
                 parts.append(f"failed {failed:.0f}")
-            if succeeded is not None:
+            if ("succeeded" in q or "completed" in q) and succeeded is not None:
                 parts.append(f"succeeded {succeeded:.0f}")
             if parts:
                 return _format_confidence(f"Pods: {', '.join(parts)}.", "high")
@@ -1461,7 +1470,12 @@ def structured_answer(
         if hw_line:
             return _format_confidence(hw_line, "high")
 
-    if entity == "node" and op == "status" and metric is None:
+    if (
+        entity == "node"
+        and op == "status"
+        and metric is None
+        and not (include_hw or exclude_hw or nodes_in_query or only_workers or role_filters)
+    ):
         summary = _nodes_summary_line(inventory, snapshot)
         if summary:
             return _format_confidence(summary, "high")