diff --git a/app/services/word_watcher.rb b/app/services/word_watcher.rb
index 119c8e71da2..c92be317236 100644
--- a/app/services/word_watcher.rb
+++ b/app/services/word_watcher.rb
@@ -61,7 +61,6 @@ class WordWatcher
 
   # This regexp is run in miniracer, and the client JS app
   # Make sure it is compatible with major browsers when changing
-  # hint: non-chrome browsers do not support 'lookbehind'
   def self.compiled_regexps_for_action(action, engine: :ruby, raise_errors: false)
     words = cached_words_for_action(action)
     return [] if words.blank?
@@ -283,13 +282,28 @@ class WordWatcher
 
   private_class_method :censor_text_with_regexp
 
-  # Returns a regexp that transforms a regular expression into a regular
-  # expression that matches a whole word.
+  SPACELESS_SCRIPTS = {
+    "Han" => "\\u4E00-\\u9FFF\\u3400-\\u4DBF",
+    "Hiragana" => "\\u3040-\\u309F",
+    "Katakana" => "\\u30A0-\\u30FF",
+    "Hangul" => "\\uAC00-\\uD7AF",
+    "Thai" => "\\u0E00-\\u0E7F",
+    "Lao" => "\\u0E80-\\u0EFF",
+    "Myanmar" => "\\u1000-\\u109F",
+    "Khmer" => "\\u1780-\\u17FF",
+    "Tibetan" => "\\u0F00-\\u0FFF",
+  }.values.join
+
   def self.match_word_regexp(regexp, engine: :ruby)
+    s = SPACELESS_SCRIPTS
     if engine == :js
-      "(?:\\P{L}|^)(#{regexp})(?=\\P{L}|$)"
+      leading = "(?:[\\P{L}#{s}]|^|(?=[#{s}]))"
+      trailing = "(?:(?=[\\P{L}#{s}]|$)|(?<=[#{s}]))"
+      "#{leading}(#{regexp})#{trailing}"
     elsif engine == :ruby
-      "(?:[^[:word:]]|^)(#{regexp})(?=[^[:word:]]|$)"
+      leading = "(?:(?<![[:word:]&&[^#{s}]])|(?=[#{s}]))"
+      trailing = "(?:(?![[:word:]&&[^#{s}]])|(?<=[#{s}]))"
+      "#{leading}(#{regexp})#{trailing}"
     else
       raise "unknown regexp engine: #{engine}"
     end
diff --git a/frontend/discourse/admin/components/modal/watched-word-testing.gjs b/frontend/discourse/admin/components/modal/watched-word-testing.gjs
index 2d7f3190f15..13777adee47 100644
--- a/frontend/discourse/admin/components/modal/watched-word-testing.gjs
+++ b/frontend/discourse/admin/components/modal/watched-word-testing.gjs
@@ -106,7 +106,10 @@ export default class WatchedWordTesting extends Component {
           options.case_sensitive ? "gu" : "gui"
         );
 
-        matches.push(...(this.value.match(wordRegexp) || []));
+        let match;
+        while ((match = wordRegexp.exec(this.value)) !== null) {
+          matches.push(match[1] || match[0]);
+        }
       } catch {
         hasCompiledExpressionError = true;
       }
diff --git a/spec/services/word_watcher_spec.rb b/spec/services/word_watcher_spec.rb
index 0867d2d2b4f..7ee982e8be2 100644
--- a/spec/services/word_watcher_spec.rb
+++ b/spec/services/word_watcher_spec.rb
@@ -80,10 +80,14 @@ RSpec.describe WordWatcher do
         SiteSetting.watched_words_regular_expressions = false
         regexps = described_class.compiled_regexps_for_action(:block)
 
+        s = WordWatcher::SPACELESS_SCRIPTS
+        leading = "(?:(?<![[:word:]&&[^#{s}]])|(?=[#{s}]))"
+        trailing = "(?:(?![[:word:]&&[^#{s}]])|(?<=[#{s}]))"
+
         expect(regexps).to be_an(Array)
         expect(regexps.map(&:inspect)).to contain_exactly(
-          "/(?:[^[:word:]]|^)(#{word1}|#{word2})(?=[^[:word:]]|$)/i",
-          "/(?:[^[:word:]]|^)(#{word3}|#{word4})(?=[^[:word:]]|$)/",
+          "/#{leading}(#{word1}|#{word2})#{trailing}/i",
+          "/#{leading}(#{word3}|#{word4})#{trailing}/",
         )
       end
 
@@ -126,7 +130,11 @@ RSpec.describe WordWatcher do
       it "works correctly when regular expressions are disabled" do
         regexps = described_class.compiled_regexps_for_action(:block)
         expect(regexps).to be_an(Array)
-        expect(regexps).to contain_exactly(/(?:[^[:word:]]|^)(\S*abc)(?=[^[:word:]]|$)/i)
+        expect(regexps.size).to eq(1)
+        expect(regexps.first).to match("xyzabc")
+        expect(regexps.first).to match(" abc")
+        expect(regexps.first).to match("testabc")
+        expect(regexps.first).not_to match("abcdef")
       end
 
       it "skips invalid watched words when regular expression are enabled" do
@@ -257,6 +265,56 @@ RSpec.describe WordWatcher do
         ).to eq("love")
       end
 
+      it "handles CJK characters as word boundaries" do
+        Fabricate(:watched_word, word: "测试", action: WatchedWord.actions[:require_approval])
+
+        expect(described_class.new("测试").word_matches_for_action?(:require_approval)[1]).to eq("测试")
+        expect(
+          described_class.new("这是一个测试文本").word_matches_for_action?(:require_approval)[1],
+        ).to eq("测试")
+        expect(
+          described_class.new("hello 测试 world").word_matches_for_action?(:require_approval)[1],
+        ).to eq("测试")
+        expect(described_class.new("API测试结果").word_matches_for_action?(:require_approval)[1]).to eq(
+          "测试",
+        )
+      end
+
+      it "handles Latin watched words adjacent to CJK text" do
+        Fabricate(:watched_word, word: "Test", action: WatchedWord.actions[:require_approval])
+
+        expect(
+          described_class.new("我的Test很好").word_matches_for_action?(:require_approval)[1],
+        ).to eq("Test")
+        expect(
+          described_class.new("Testing").word_matches_for_action?(:require_approval),
+        ).to be_falsey
+      end
+
+      it "handles CJK boundaries with the JS engine" do
+        Fabricate(:watched_word, word: "测试", action: WatchedWord.actions[:require_approval])
+
+        regexps = described_class.compiled_regexps_for_action(:require_approval, engine: :js)
+        expect(regexps.size).to eq(1)
+
+        re = regexps.first
+        expect(re).to match("测试")
+        expect(re).to match("这是一个测试文本")
+        expect(re).to match("hello 测试 world")
+        expect(re).to match("API测试结果")
+      end
+
+      it "handles Latin words adjacent to CJK text with the JS engine" do
+        Fabricate(:watched_word, word: "Test", action: WatchedWord.actions[:require_approval])
+
+        regexps = described_class.compiled_regexps_for_action(:require_approval, engine: :js)
+        expect(regexps.size).to eq(1)
+
+        re = regexps.first
+        expect(re).to match("我的Test很好")
+        expect(re).not_to match("Testing")
+      end
+
       context "when there are multiple matches" do
         context "with non regexp words" do
           it "lists all matching words" do