diff --git a/telegram/bot.py b/telegram/bot.py
index 2edd3cb..de6fd8f 100644
--- a/telegram/bot.py
+++ b/telegram/bot.py
@@ -306,6 +306,56 @@ def _format_conversation_history(chat_id: int, user_id: int) -> str:
 RESEARCH_PATTERN = re.compile(r'/research(?:@\w+)?\s+(.+)', re.IGNORECASE)
 
 
+async def _research_and_followup(msg, query: str, user):
+    """Run X search and send a follow-up message with findings.
+
+    Used when Opus triggers RESEARCH: tag — the user expects results back,
+    not silent archival.
+    """
+    from x_client import search_tweets as _search
+    logger.info("Research follow-up: searching X for '%s'", query)
+    tweets = await _search(query, max_results=10, min_engagement=0)
+    if not tweets:
+        await msg.reply_text(f"Searched X for '{query}' — nothing recent found.")
+        return
+
+    # Build concise summary of findings
+    lines = [f"Found {len(tweets)} recent posts about '{query}':\n"]
+    for t in tweets[:5]:
+        author = t.get("author", "?")
+        text = t.get("text", "")[:200]
+        url = t.get("url", "")
+        lines.append(f"@{author}: {text}")
+        if url:
+            lines.append(f"  {url}")
+        lines.append("")
+
+    followup = "\n".join(lines)
+    # Split if needed
+    if len(followup) <= 4096:
+        await msg.reply_text(followup)
+    else:
+        chunks = []
+        remaining = followup
+        while remaining:
+            if len(remaining) <= 4096:
+                chunks.append(remaining)
+                break
+            split_at = remaining.rfind("\n\n", 0, 4000)
+            if split_at == -1:
+                split_at = remaining.rfind("\n", 0, 4096)
+            if split_at == -1:
+                split_at = 4096
+            chunks.append(remaining[:split_at])
+            remaining = remaining[split_at:].lstrip("\n")
+        for chunk in chunks:
+            if chunk.strip():
+                await msg.reply_text(chunk)
+
+    # Also archive for pipeline
+    await handle_research(msg, query, user, silent=True)
+
+
 async def handle_research(msg, query: str, user, silent: bool = False):
     """Handle a research request — search X and archive results as sources.
 
@@ -716,8 +766,10 @@ IMPORTANT: Two special tags you can append at the end of your response (after yo
         display_response = re.sub(r'\nRESEARCH:\s+.+$', '', display_response, flags=re.MULTILINE).rstrip()
         if not research_context:  # Only fire if Haiku didn't already search
             for query in research_lines:
-                asyncio.get_event_loop().create_task(handle_research(msg, query.strip(), user, silent=True))
-                logger.info("Auto-research triggered: %s", query[:80])
+                # Send follow-up with findings (not silent — user expects results)
+                asyncio.get_event_loop().create_task(
+                    _research_and_followup(msg, query.strip(), user))
+                logger.info("Auto-research triggered (will follow up): %s", query[:80])
 
     # Post response (without LEARNING lines)
     # Telegram has a 4096 char limit — split long messages