diff --git a/CHANGELOG.md b/CHANGELOG.md
index eeb13b6..1d3a1a8 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,10 @@
 # Changelog
 
+## 2026-06-04
+
+### Changes
+- [Tester] When a click or locator keeps failing on a button or link that is plainly visible on the page, the Tester now always tries clicking it by its visual appearance at least once before re-researching, switching targets, or deciding the element is unreachable. Previously it could exhaust broken locators and wrongly conclude a visible control did not exist.
+
 ## 2026-06-03
 
 ### Changes
diff --git a/src/ai/tester.ts b/src/ai/tester.ts
index 65961e4..bb634b7 100644
--- a/src/ai/tester.ts
+++ b/src/ai/tester.ts
@@ -750,7 +750,7 @@ export class Tester extends TaskAgent implements Agent {
     - When filling complex form with lot of actions performed, use see() to look which fields were filled and which are not
     - When verify() fails, use see() to visually confirm the result — visual confirmation is equally valid evidence
     - For visual state verification (active tabs, selected items, counts, colors), prefer see() over DOM-based verify()
-    - When click() fails and element is visually present, use visualClick() as fallback
+    - When click() fails on an element you can see — or believe is there but may look different — you MUST try visualClick() before giving up; don't repeat visualClick in a row
     - If you land on a "Not Found", 404, or error page that is NOT part of the scenario, call reset() immediately to return to the initial page and try again
     - If you see a server error page (500, 503, etc.), record it with record({ notes: ["Server error on /path"], status: "fail" }) and call reset() to continue testing
     </rules>