diff --git a/nordabiz_chat.py b/nordabiz_chat.py
index 1d6df65..54cc4b7 100644
--- a/nordabiz_chat.py
+++ b/nordabiz_chat.py
@@ -1132,11 +1132,74 @@ BŁĘDNIE (NIE RÓB - resetuje numerację):
                 user_id=user_id,
                 temperature=0.7
             )
-            return response_text
+            # Post-process to ensure links are added even if AI didn't format them
+            return self._postprocess_links(response_text, context)
         else:
             # Legacy: direct API call (no centralized cost tracking)
             response = self.model.generate_content(full_prompt)
-            return response.text
+            # Post-process to ensure links are added even if AI didn't format them
+            return self._postprocess_links(response.text, context)
+
+    def _postprocess_links(self, text: str, context: Dict) -> str:
+        """
+        Post-process AI response to add markdown links for companies and people.
+        This ensures consistent linking regardless of AI behavior.
+
+        Args:
+            text: AI response text
+            context: Context dict with company_people data
+
+        Returns:
+            Text with names replaced by markdown links
+        """
+        import re
+
+        # Build lookup dict: name -> url
+        name_to_url = {}
+
+        # Extract companies and people from company_people context
+        company_people = context.get('company_people', {})
+        for company_name, data in company_people.items():
+            # Add company
+            if data.get('profile'):
+                name_to_url[company_name] = data['profile']
+            # Add people
+            for person in data.get('people', []):
+                if person.get('name') and person.get('profile'):
+                    name_to_url[person['name']] = person['profile']
+
+        # Also extract from companies list (context['companies'] has profile URLs)
+        # Companies format: list of dicts with 'name' and 'profile'
+        # This is populated by _company_to_compact_dict
+
+        # Sort by name length (longest first) to avoid partial replacements
+        sorted_names = sorted(name_to_url.keys(), key=len, reverse=True)
+
+        for name in sorted_names:
+            url = name_to_url[name]
+            if not name or not url:
+                continue
+
+            # Skip if already a markdown link
+            # Pattern: [Name](url) - already linked
+            already_linked = re.search(r'\[' + re.escape(name) + r'\]\([^)]+\)', text)
+            if already_linked:
+                continue
+
+            # Replace **Name** (bold) with [Name](url)
+            bold_pattern = r'\*\*' + re.escape(name) + r'\*\*'
+            if re.search(bold_pattern, text):
+                text = re.sub(bold_pattern, f'[{name}]({url})', text, count=1)
+                continue
+
+            # Replace plain "Name" at word boundaries (but not if already in link)
+            # Be careful not to replace inside existing markdown
+            plain_pattern = r'(?<!\[)(?<!\()' + re.escape(name) + r'(?!\])(?!\))'
+            if re.search(plain_pattern, text):
+                # Only replace first occurrence to avoid over-linking
+                text = re.sub(plain_pattern, f'[{name}]({url})', text, count=1)
+
+        return text
 
     def _calculate_cost(self, input_tokens: int, output_tokens: int) -> float:
         """