diff --git a/src/lib/server/db/migrations/010_thumbnail_cache_rerun_negatives.sql b/src/lib/server/db/migrations/010_thumbnail_cache_rerun_negatives.sql
new file mode 100644
index 0000000..5e5fac0
--- /dev/null
+++ b/src/lib/server/db/migrations/010_thumbnail_cache_rerun_negatives.sql
@@ -0,0 +1,6 @@
+-- Der Recipe-Detektor prüft ab jetzt zusätzlich zu JSON-LD auch Microdata
+-- (itemtype=schema.org/Recipe). Der Cache kann has_recipe=0-Einträge
+-- enthalten, die mit dem alten Check falsch-negativ waren (z.B. rezeptwelt.de,
+-- das Microdata statt JSON-LD nutzt). Einmalig wegräumen, damit die Seiten
+-- beim nächsten Search neu klassifiziert werden. Reiner Cache-Flush.
+DELETE FROM thumbnail_cache WHERE has_recipe = 0;
diff --git a/src/lib/server/parsers/json-ld-recipe.ts b/src/lib/server/parsers/json-ld-recipe.ts
index 1827bb0..908ae8a 100644
--- a/src/lib/server/parsers/json-ld-recipe.ts
+++ b/src/lib/server/parsers/json-ld-recipe.ts
@@ -106,7 +106,13 @@ function findRecipeNode(html: string): JsonLdNode | null {
   return null;
 }
 
-export function hasRecipeJsonLd(html: string): boolean {
+// Microdata-Alternative zum JSON-LD: viele SSR-Sites (inkl. rezeptwelt.de)
+// nutzen <div itemtype="https://schema.org/Recipe"> statt application/ld+json.
+// Ein einfacher Regex reicht — wir brauchen nur das Flag, nicht die Daten.
+const MICRODATA_RECIPE = /itemtype\s*=\s*["']https?:\/\/schema\.org\/Recipe["']/i;
+
+export function hasRecipeMarkup(html: string): boolean {
+  if (MICRODATA_RECIPE.test(html)) return true;
   try {
     return findRecipeNode(html) !== null;
   } catch {
@@ -114,6 +120,11 @@ export function hasRecipeJsonLd(html: string): boolean {
   }
 }
 
+// @deprecated use hasRecipeMarkup
+export function hasRecipeJsonLd(html: string): boolean {
+  return hasRecipeMarkup(html);
+}
+
 export function extractRecipeFromHtml(html: string): Recipe | null {
   const node = findRecipeNode(html);
   if (!node) return null;
diff --git a/src/lib/server/search/searxng.ts b/src/lib/server/search/searxng.ts
index e6dc587..dbdcf23 100644
--- a/src/lib/server/search/searxng.ts
+++ b/src/lib/server/search/searxng.ts
@@ -2,7 +2,7 @@ import type Database from 'better-sqlite3';
 import { parseHTML } from 'linkedom';
 import { listDomains, normalizeDomain } from '../domains/repository';
 import { fetchText } from '../http';
-import { hasRecipeJsonLd } from '../parsers/json-ld-recipe';
+import { hasRecipeMarkup } from '../parsers/json-ld-recipe';
 
 export type WebHit = {
   url: string;
@@ -235,7 +235,7 @@ async function enrichPageMeta(
     });
     meta = {
       image: extractPageImage(html, url),
-      hasRecipe: hasRecipeJsonLd(html) ? 1 : 0
+      hasRecipe: hasRecipeMarkup(html) ? 1 : 0
     };
   } catch {
     // Fetch failed — leave hasRecipe null (unknown) so we don't permanently
@@ -363,9 +363,19 @@ export async function searchWeb(
   );
   if (opts.enrichThumbnails !== false) {
     const enriched = await enrichAndFilterHits(db, hits);
+    const droppedUrls = hits
+      .filter((h) => !enriched.find((e) => e.url === h.url))
+      .map((h) => h.url);
     console.log(
-      `[searxng] q=${JSON.stringify(trimmed)} pageno=${pageno} enrich=${hits.length} dropped_non_recipe=${hits.length - enriched.length} final=${enriched.length}`
+      `[searxng] q=${JSON.stringify(trimmed)} pageno=${pageno} enrich=${hits.length} dropped_non_recipe=${droppedUrls.length} final=${enriched.length}`
     );
+    // Nur die ersten 3 URLs mitloggen, damit das Log nicht explodiert. Genug
+    // um eine Seite manuell zu analysieren („warum wurde die abgelehnt?").
+    if (droppedUrls.length > 0) {
+      console.log(
+        `[searxng] dropped samples: ${droppedUrls.slice(0, 3).join(' | ')}`
+      );
+    }
     return enriched;
   }
   return hits;
diff --git a/tests/unit/json-ld-recipe.test.ts b/tests/unit/json-ld-recipe.test.ts
index fb62dd0..f6f8240 100644
--- a/tests/unit/json-ld-recipe.test.ts
+++ b/tests/unit/json-ld-recipe.test.ts
@@ -2,7 +2,10 @@ import { describe, it, expect } from 'vitest';
 import { readFileSync } from 'node:fs';
 import { dirname, join } from 'node:path';
 import { fileURLToPath } from 'node:url';
-import { extractRecipeFromHtml } from '../../src/lib/server/parsers/json-ld-recipe';
+import {
+  extractRecipeFromHtml,
+  hasRecipeMarkup
+} from '../../src/lib/server/parsers/json-ld-recipe';
 
 const here = dirname(fileURLToPath(import.meta.url));
 
@@ -42,3 +45,31 @@ describe('extractRecipeFromHtml', () => {
     expect(extractRecipeFromHtml(html)).toBeNull();
   });
 });
+
+describe('hasRecipeMarkup', () => {
+  it('detects JSON-LD Recipe', () => {
+    const html = `<html><head>
+      <script type="application/ld+json">{"@type":"Recipe","name":"x"}</script>
+      </head></html>`;
+    expect(hasRecipeMarkup(html)).toBe(true);
+  });
+
+  it('detects schema.org/Recipe microdata', () => {
+    const html = `<html><body>
+      <div itemscope itemtype="https://schema.org/Recipe">
+        <span itemprop="name">Königsberger Klopse</span>
+      </div>
+      </body></html>`;
+    expect(hasRecipeMarkup(html)).toBe(true);
+  });
+
+  it('also detects http:// variant of schema.org/Recipe', () => {
+    const html = `<div itemscope itemtype="http://schema.org/Recipe"></div>`;
+    expect(hasRecipeMarkup(html)).toBe(true);
+  });
+
+  it('returns false for pages without any recipe markup', () => {
+    const html = '<html><body><p>nothing here</p></body></html>';
+    expect(hasRecipeMarkup(html)).toBe(false);
+  });
+});