diff --git a/floatchat/router.py b/floatchat/router.py
index 6dcf8db..226ffa6 100644
--- a/floatchat/router.py
+++ b/floatchat/router.py
@@ -64,6 +64,9 @@ def _prompt(question: str, history: list[dict]) -> str:
         '{"path": "analytical" | "semantic", "intent": <intent or null>, "filters": {...}, "question": <standalone>}\n\n'
         "path=analytical for statistics (averages, comparisons, counts, trends) answered with SQL; "
         "path=semantic for finding or describing particular profiles or floats.\n"
+        "A request to see, find or list profiles is semantic even when it names a place or a date "
+        "(e.g. 'Profiles off Sri Lanka in June 2026'). Use analytical only when the question asks for a number, "
+        "a comparison, a trend or a summary table.\n"
         f"Intents (analytical only):\n{intents}\n"
         f"Filter keys: {', '.join(FILTER_KEYS)}. region is one of: {', '.join(sorted(REGION_NAMES))}. "
         "month_num is 1-12; has_* filters are true when the question is about that sensor.\n"
diff --git a/tests/unit/test_router.py b/tests/unit/test_router.py
index d664a43..96053a1 100644
--- a/tests/unit/test_router.py
+++ b/tests/unit/test_router.py
@@ -78,3 +78,12 @@ def test_llm_prompt_includes_history():
     chat = fake_llm({"path": "analytical", "intent": "avg_by_depth", "filters": {}, "question": "q"})
     llm_route("now only the Arabian Sea", [PREVIOUS_SQL], chat)
     assert "What is the average temperature at different depths?" in chat.calls[0][0]
+
+
+def test_llm_prompt_sends_requests_for_profiles_to_search():
+    # A bare "Profiles in <place> in <month>" asks to see profiles, not for a statistic
+    chat = fake_llm({"path": "semantic", "intent": None, "filters": {}, "question": "q"})
+    llm_route("Profiles near Indonesia in January 2026", [], chat)
+    prompt = chat.calls[0][0]
+    assert "even when it names a place or a date" in prompt
+    assert "Profiles off Sri Lanka in June 2026" in prompt