From 5c640d02f7c52001d015f005c7319ed75e2d9008 Mon Sep 17 00:00:00 2001
From: Oliver Rasche <o.rasche@evomedien.de>
Date: Tue, 10 Nov 2026 11:36:48 +0100
Subject: [PATCH] News: redaktionell eingefuegtes FAQPage-JSON als echtes
 Schema ausgeben

17 SEO-Artikel tragen ein rohes FAQPage-JSON-Objekt in einem
abschliessenden Absatz ihres Bodytexts - das Frontend renderte es als
sichtbaren Text, Google sah es nie als strukturierte Daten.

Neue Klasse Seo\EmbeddedJsonLd trennt solche schema.org-Bloecke aus dem
RTE-HTML (fail-safe: was nicht sauber dekodiert, bleibt stehen):
- NewsJsonRenderer: Detail-bodytext und Listen-Exzerpt ohne die Bloecke
- PageJsonLdRenderer: die dekodierten Nodes wandern in den @graph der
  Seite (FAQPage landet damit als <script type=application/ld+json> im
  Head, den das Frontend bereits injiziert)
- VitecMetaHandler: Description-Fallback aus dem Bodytext ohne JSON-Muell

Verifiziert an 3 Artikeln: @graph mit NewsArticle + FAQPage (8-9 Fragen),
kein FAQPage-Rest im Textkoerper, kein sichtbarer JSON-Absatz im HTML.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 packages/vitec/Classes/Seo/EmbeddedJsonLd.php | 56 +++++++++++++++++++
 .../vitec/Classes/Seo/VitecMetaHandler.php    |  4 +-
 .../Classes/UserFunc/NewsJsonRenderer.php     |  6 +-
 .../Classes/UserFunc/PageJsonLdRenderer.php   |  7 +++
 4 files changed, 71 insertions(+), 2 deletions(-)
 create mode 100644 packages/vitec/Classes/Seo/EmbeddedJsonLd.php

diff --git a/packages/vitec/Classes/Seo/EmbeddedJsonLd.php b/packages/vitec/Classes/Seo/EmbeddedJsonLd.php
new file mode 100644
index 0000000..17597cd
--- /dev/null
+++ b/packages/vitec/Classes/Seo/EmbeddedJsonLd.php
@@ -0,0 +1,56 @@
+<?php
+
+declare(strict_types=1);
+
+namespace Evomedien\Vitec\Seo;
+
+/**
+ * Splits editor-pasted schema.org JSON blocks out of RTE HTML.
+ *
+ * The SEO-maintained news articles carry a raw FAQPage JSON object in a
+ * trailing <p>…</p> of their bodytext (found 2026-11-10: 17 records). Left
+ * in place, the frontend renders the JSON as visible text while Google
+ * never sees it as structured data. This helper removes such paragraphs
+ * from the HTML and returns the decoded nodes, ready to join the page's
+ * JSON-LD @graph (their own @context is dropped - the graph carries one).
+ *
+ * Fail-safe: a paragraph that merely looks like JSON but does not decode
+ * to a schema.org node stays in the HTML untouched.
+ */
+final class EmbeddedJsonLd
+{
+    /**
+     * @return array{html: string, nodes: list<array<string,mixed>>}
+     */
+    public static function extract(string $html): array
+    {
+        if ($html === '' || !str_contains($html, 'schema.org')) {
+            return ['html' => $html, 'nodes' => []];
+        }
+
+        $nodes = [];
+        $clean = preg_replace_callback(
+            '#<p[^>]*>\s*(\{.+?\})\s*</p>#s',
+            static function (array $match) use (&$nodes): string {
+                $candidate = $match[1];
+                if (!str_contains($candidate, 'schema.org')) {
+                    return $match[0];
+                }
+                $decoded = json_decode($candidate, true);
+                if (!is_array($decoded)) {
+                    // RTE processing may have entity-encoded the quotes.
+                    $decoded = json_decode(html_entity_decode($candidate, ENT_QUOTES | ENT_HTML5), true);
+                }
+                if (!is_array($decoded) || empty($decoded['@type'])) {
+                    return $match[0];
+                }
+                unset($decoded['@context']);
+                $nodes[] = $decoded;
+                return '';
+            },
+            $html
+        ) ?? $html;
+
+        return ['html' => trim($clean), 'nodes' => $nodes];
+    }
+}
diff --git a/packages/vitec/Classes/Seo/VitecMetaHandler.php b/packages/vitec/Classes/Seo/VitecMetaHandler.php
index 1d6b9ca..beb1413 100644
--- a/packages/vitec/Classes/Seo/VitecMetaHandler.php
+++ b/packages/vitec/Classes/Seo/VitecMetaHandler.php
@@ -85,7 +85,9 @@ class VitecMetaHandler extends MetaHandler
                 }
                 $description = $this->plainText((string)($news['teaser'] ?? ''));
                 if ($description === '') {
-                    $description = $this->plainText((string)($news['bodytext'] ?? ''));
+                    $description = $this->plainText(
+                        EmbeddedJsonLd::extract((string)($news['bodytext'] ?? ''))['html']
+                    );
                 }
                 $image = $this->firstImage('tx_news_domain_model_news', 'fal_media', $newsUid);
             }
diff --git a/packages/vitec/Classes/UserFunc/NewsJsonRenderer.php b/packages/vitec/Classes/UserFunc/NewsJsonRenderer.php
index 5f3dbe6..4c41edf 100644
--- a/packages/vitec/Classes/UserFunc/NewsJsonRenderer.php
+++ b/packages/vitec/Classes/UserFunc/NewsJsonRenderer.php
@@ -375,9 +375,12 @@ final class NewsJsonRenderer
             // SSR HTML to 2.8 MB). Cards render from `teaser`; only records
             // WITHOUT a teaser get a plain-text excerpt in `bodytext` so no
             // card goes blank. The detail mode is unchanged.
+            // Editor-pasted schema.org blocks (FAQPage JSON in a trailing
+            // paragraph) are split out of the body - they reach Google via
+            // the page-level jsonLd @graph instead (PageJsonLdRenderer).
             'bodytext' => $mode === 'list'
                 ? (trim((string)($news['teaser'] ?? '')) !== '' ? '' : $this->listExcerpt((string)($news['bodytext'] ?? '')))
-                : RteResolver::html($news['bodytext'] ?? ''),
+                : \Evomedien\Vitec\Seo\EmbeddedJsonLd::extract(RteResolver::html($news['bodytext'] ?? ''))['html'],
             'datetime' => (int)($news['datetime'] ?? 0),
             'archive' => (int)($news['archive'] ?? 0),
             'istopnews' => (bool)($news['istopnews'] ?? false),
@@ -579,6 +582,7 @@ final class NewsJsonRenderer
      */
     private function listExcerpt(string $html): string
     {
+        $html = \Evomedien\Vitec\Seo\EmbeddedJsonLd::extract($html)['html'];
         $text = trim((string)preg_replace('/\s+/u', ' ', strip_tags($html)));
         if (mb_strlen($text) <= 300) {
             return $text;
diff --git a/packages/vitec/Classes/UserFunc/PageJsonLdRenderer.php b/packages/vitec/Classes/UserFunc/PageJsonLdRenderer.php
index e9dfec5..433ec79 100755
--- a/packages/vitec/Classes/UserFunc/PageJsonLdRenderer.php
+++ b/packages/vitec/Classes/UserFunc/PageJsonLdRenderer.php
@@ -156,6 +156,13 @@ final class PageJsonLdRenderer
                         'author' => (string)($newsRow['author'] ?? ''),
                         'author_email' => (string)($newsRow['author_email'] ?? ''),
                     ], $base, $this->recordImageUrls('tx_news_domain_model_news', 'fal_media', $newsUid, $base, $service), $hasOrganization);
+
+                    // Editor-pasted schema blocks (e.g. FAQPage JSON in a
+                    // trailing bodytext paragraph) join the @graph here;
+                    // NewsJsonRenderer strips them out of the rendered body.
+                    foreach (\Evomedien\Vitec\Seo\EmbeddedJsonLd::extract((string)($newsRow['bodytext'] ?? ''))['nodes'] as $embeddedNode) {
+                        $nodes[] = $embeddedNode;
+                    }
                 }
             }