diff --git a/src/server/features/ai-search/services/promptExplorer.test.ts b/src/server/features/ai-search/services/promptExplorer.test.ts
new file mode 100644
index 00000000..7d891226
--- /dev/null
+++ b/src/server/features/ai-search/services/promptExplorer.test.ts
@@ -0,0 +1,78 @@
+import { describe, expect, it, vi } from "vitest";
+import type { LlmResponseResult } from "@/server/lib/dataforseoLlmSchemas";
+
+vi.mock("cloudflare:workers", () => ({ waitUntil: vi.fn() }));
+
+const { extractCitations } = await import("./promptExplorer");
+
+// DataForSEO's LLM Responses payload nests references as untyped
+// `{ title, url }` objects under items[].sections[].annotations — mirroring the
+// SDK's AnnotationInfo, which has no citation-type discriminator.
+function response(
+  annotations: Array<{ title?: string; url?: string }>,
+): LlmResponseResult {
+  return {
+    model_name: "gpt-5",
+    web_search: true,
+    items: [
+      {
+        type: "reasoning",
+        sections: [{ type: "summary_text", text: "thinking" }],
+      },
+      {
+        type: "message",
+        sections: [{ type: "text", text: "answer", annotations }],
+      },
+    ],
+  };
+}
+
+describe("extractCitations", () => {
+  it("keeps untyped annotations (no citation-type discriminator exists)", () => {
+    const citations = extractCitations(
+      response([
+        { title: "Town & Country", url: "https://www.townandcountrymag.com/x" },
+        { title: "Stylevana", url: "https://www.stylevana.com/y" },
+      ]),
+    );
+    expect(citations.map((c) => c.url)).toEqual([
+      "https://www.townandcountrymag.com/x",
+      "https://www.stylevana.com/y",
+    ]);
+    expect(citations[0]?.domain).toBe("townandcountrymag.com");
+    expect(citations[0]?.title).toBe("Town & Country");
+  });
+
+  it("dedupes repeated URLs and drops unsafe schemes", () => {
+    const citations = extractCitations(
+      response([
+        { title: "A", url: "https://example.com/a" },
+        { title: "A dup", url: "https://example.com/a" },
+        { title: "evil", url: "javascript:alert(1)" },
+        { title: "no url" },
+      ]),
+    );
+    expect(citations).toHaveLength(1);
+    expect(citations[0]?.url).toBe("https://example.com/a");
+  });
+
+  it("ignores annotations outside message items and returns [] when absent", () => {
+    expect(extractCitations({ items: [] })).toEqual([]);
+    expect(
+      extractCitations({
+        items: [
+          {
+            type: "reasoning",
+            sections: [
+              {
+                type: "summary_text",
+                text: "t",
+                annotations: [{ title: "x", url: "https://x.test/1" }],
+              },
+            ],
+          },
+        ],
+      }),
+    ).toEqual([]);
+  });
+});
diff --git a/src/server/features/ai-search/services/promptExplorer.ts b/src/server/features/ai-search/services/promptExplorer.ts
index 6312f9e7..21516875 100644
--- a/src/server/features/ai-search/services/promptExplorer.ts
+++ b/src/server/features/ai-search/services/promptExplorer.ts
@@ -207,7 +207,7 @@ function extractText(response: LlmResponseResult): string {
   return textParts.join("\n\n").trim();
 }
 
-function extractCitations(
+export function extractCitations(
   response: LlmResponseResult,
 ): PromptExplorerCitation[] {
   const seen = new Set<string>();
@@ -217,10 +217,10 @@ function extractCitations(
     if (item.type !== "message") continue;
     for (const section of item.sections ?? []) {
       for (const annotation of section.annotations ?? []) {
-        if (annotation.type !== "citation") continue;
-        // Drop non-http(s) URLs — LLMs can be coaxed into emitting
-        // `javascript:` payloads as "citations" and we render these as
-        // <a href> in the UI.
+        // DataForSEO annotations are untyped `{ title, url }` reference
+        // objects (AnnotationInfo) — there is no citation-type discriminator
+        // to filter on. Guard on URL safety only: LLMs can be coaxed into
+        // emitting `javascript:` payloads, and we render these as <a href>.
         const safeUrl = safeHttpUrl(annotation.url);
         if (!safeUrl || seen.has(safeUrl)) continue;
         seen.add(safeUrl);