fix(seo): canoniques cross-locale et URLs noindex au sitemap

public/llms-full.txt

@@ -1,6 +1,6 @@
 # ankora.be — Full content export for LLMs
 
-Last generated: 2026-07-25
+Last generated: 2026-07-26
 Canonical URL: https://ankora.be
 License: content available for citation with attribution. Code is proprietary.
 

src/app/[locale]/(public)/faq/page.tsx

@@ -4,6 +4,7 @@ import { getTranslations } from 'next-intl/server';
 import { Header } from '@/components/layout/Header';
 import { Footer } from '@/components/layout/Footer';
 import { JsonLd } from '@/components/seo/JsonLd';
+import { buildCanonicalUrl } from '@/lib/glossary';
 
 const QUESTION_KEYS = [
   'bankConnection',
@@ -16,12 +17,25 @@ const QUESTION_KEYS = [
   'sharing',
 ] as const;
 
-export async function generateMetadata(): Promise<Metadata> {
+export async function generateMetadata({
+  params,
+}: {
+  params: Promise<{ locale: string }>;
+}): Promise<Metadata> {
+  // `params`, not `getLocale()`: the latter works only because the locale
+  // layout happens to call `cookies()` for the theme, which forces dynamic
+  // rendering. Reading the segment directly removes that hidden coupling and
+  // matches how the glossary pages already do it.
+  const { locale } = await params;
   const t = await getTranslations('faq');
   return {
     title: t('metaTitle'),
     description: t('metaDescription'),
-    alternates: { canonical: '/faq' },
+    // Locale-aware: when a segment declares `alternates`, Next REPLACES the
+    // parent's value instead of recomputing it per locale. Hardcoding '/faq'
+    // made /en/faq canonicalise to the French page, contradicting the hreflang
+    // Link header emitted for that same URL.
+    alternates: { canonical: buildCanonicalUrl('/faq', locale) },
   };
 }
 

src/app/[locale]/(public)/glossaire/page.tsx

@@ -23,6 +23,21 @@ export async function generateMetadata({ params }: LocaleParams) {
   return {
     title: t('metaTitle'),
     description: t('metaDescription'),
+    // Without an explicit `alternates`, Next keeps the PARENT layout's canonical
+    // — so /glossaire, /en/glossaire and /nl-BE/glossaire each canonicalised to
+    // their own homepage instead of to themselves, and none of them could rank.
+    // `glossaire/[slug]` already did this correctly; the index did not.
+    alternates: {
+      canonical: buildCanonicalUrl('/glossaire', locale),
+      // Same `languages` block as `[slug]/page.tsx`. Omitting it here would
+      // repeat, on the fix itself, the very defect this PR closes: a canonical
+      // with no matching hreflang, leaving the head to say nothing about the
+      // other locales. Scoped to GLOSSARY_LOCALES because only those three have
+      // translated terms.
+      languages: Object.fromEntries(
+        GLOSSARY_LOCALES.map((l) => [l, buildCanonicalUrl('/glossaire', l)]),
+      ),
+    },
   };
 }
 

src/app/[locale]/(public)/legal/cgu/page.tsx

@@ -4,17 +4,27 @@ import { getTranslations } from 'next-intl/server';
 import { Header } from '@/components/layout/Header';
 import { Footer } from '@/components/layout/Footer';
 import { Prose, ProseMeta } from '@/components/layout/Prose';
+import { buildCanonicalUrl } from '@/lib/glossary';
 
 const LAST_UPDATED = '16 avril 2026';
 const VERSION = '1.0.0';
 
-export async function generateMetadata(): Promise<Metadata> {
+export async function generateMetadata({
+  params,
+}: {
+  params: Promise<{ locale: string }>;
+}): Promise<Metadata> {
+  // `params`, not `getLocale()`: the latter works only because the locale
+  // layout happens to call `cookies()` for the theme, which forces dynamic
+  // rendering. Reading the segment directly removes that hidden coupling and
+  // matches how the glossary pages already do it.
+  const { locale } = await params;
   const t = await getTranslations('legal.cgu');
   return {
     title: t('metaTitle'),
     description: t('metaDescription'),
     robots: { index: false, follow: true },
-    alternates: { canonical: '/legal/cgu' },
+    alternates: { canonical: buildCanonicalUrl('/legal/cgu', locale) },
   };
 }
 

src/app/[locale]/(public)/legal/cookies/page.tsx

@@ -4,16 +4,26 @@ import { getTranslations } from 'next-intl/server';
 import { Header } from '@/components/layout/Header';
 import { Footer } from '@/components/layout/Footer';
 import { Prose, ProseMeta } from '@/components/layout/Prose';
+import { buildCanonicalUrl } from '@/lib/glossary';
 
 const LAST_UPDATED = '16 avril 2026';
 
-export async function generateMetadata(): Promise<Metadata> {
+export async function generateMetadata({
+  params,
+}: {
+  params: Promise<{ locale: string }>;
+}): Promise<Metadata> {
+  // `params`, not `getLocale()`: the latter works only because the locale
+  // layout happens to call `cookies()` for the theme, which forces dynamic
+  // rendering. Reading the segment directly removes that hidden coupling and
+  // matches how the glossary pages already do it.
+  const { locale } = await params;
   const t = await getTranslations('legal.cookies');
   return {
     title: t('metaTitle'),
     description: t('metaDescription'),
     robots: { index: false, follow: true },
-    alternates: { canonical: '/legal/cookies' },
+    alternates: { canonical: buildCanonicalUrl('/legal/cookies', locale) },
   };
 }
 

src/app/[locale]/(public)/legal/privacy/page.tsx

@@ -6,17 +6,27 @@ import { brand } from '@/lib/brand';
 import { Header } from '@/components/layout/Header';
 import { Footer } from '@/components/layout/Footer';
 import { Prose, ProseMeta } from '@/components/layout/Prose';
+import { buildCanonicalUrl } from '@/lib/glossary';
 
 const LAST_UPDATED = '16 avril 2026';
 const VERSION = '1.0.0';
 
-export async function generateMetadata(): Promise<Metadata> {
+export async function generateMetadata({
+  params,
+}: {
+  params: Promise<{ locale: string }>;
+}): Promise<Metadata> {
+  // `params`, not `getLocale()`: the latter works only because the locale
+  // layout happens to call `cookies()` for the theme, which forces dynamic
+  // rendering. Reading the segment directly removes that hidden coupling and
+  // matches how the glossary pages already do it.
+  const { locale } = await params;
   const t = await getTranslations('legal.privacy');
   return {
     title: t('metaTitle'),
     description: t('metaDescription'),
     robots: { index: false, follow: true },
-    alternates: { canonical: '/legal/privacy' },
+    alternates: { canonical: buildCanonicalUrl('/legal/privacy', locale) },
   };
 }
 

src/app/sitemap.ts

@@ -1,17 +1,37 @@
 import type { MetadataRoute } from 'next';
 import { SITE } from '@/lib/site';
-import { routing } from '@/i18n/routing';
+import { LOCALES_VISIBLE, routing } from '@/i18n/routing';
 import { GLOSSARY_LOCALES, getGlossaryTerms, GLOSSARY_LOCALE_PREFIXES } from '@/lib/glossary';
 
-const PUBLIC_ROUTES = ['', '/faq', '/legal/cgu', '/legal/privacy', '/legal/cookies'] as const;
+/**
+ * Routes worth submitting for indexing.
+ *
+ * `/legal/*` is deliberately absent: those pages set
+ * `robots: { index: false, follow: true }`, so submitting them made Search
+ * Console report "Submitted URL marked noindex" on 15 URLs (3 pages × 5
+ * locales) — a self-inflicted error report, not a ranking opportunity.
+ */
+const PUBLIC_ROUTES = ['', '/faq'] as const;
+
+/**
+ * Locales the sitemap advertises.
+ *
+ * `LOCALES_VISIBLE` (FR + EN), not `routing.locales`. `nl-BE`, `de-DE` and
+ * `es-ES` resolve — deep links and QA bookmarks keep working — but their
+ * `landing.*` copy is still French verbatim, so submitting them asks Google to
+ * index untranslated pages under a Dutch/German/Spanish URL. The glossary
+ * already scopes itself this way through `GLOSSARY_LOCALES`; this aligns the
+ * rest. Add a locale back here once its translation is reviewed.
+ */
+const INDEXABLE_LOCALES = LOCALES_VISIBLE;
 
 export default function sitemap(): MetadataRoute.Sitemap {
   const now = new Date();
   const base = SITE.url;
 
   const entries: MetadataRoute.Sitemap = [];
 
-  for (const locale of routing.locales) {
+  for (const locale of INDEXABLE_LOCALES) {
     const prefix = locale === routing.defaultLocale ? '' : `/${locale}`;
     for (const route of PUBLIC_ROUTES) {
       entries.push({
@@ -21,7 +41,7 @@ export default function sitemap(): MetadataRoute.Sitemap {
         priority: route === '' ? 1 : 0.6,
         alternates: {
           languages: Object.fromEntries(
-            routing.locales.map((l) => [
+            INDEXABLE_LOCALES.map((l) => [
               l,
               `${base}${l === routing.defaultLocale ? '' : `/${l}`}${route}`,
             ]),

tests/seo/sitemap.test.ts

@@ -0,0 +1,90 @@
+import { describe, it, expect } from 'vitest';
+
+import sitemap from '@/app/sitemap';
+import { GLOSSARY_LOCALES } from '@/lib/glossary';
+import { LOCALES_VISIBLE, routing } from '@/i18n/routing';
+
+/**
+ * Sitemap invariants.
+ *
+ * Two defects motivated these, both found by `seo-geo-auditor` and both silent:
+ * nothing failed, Search Console simply reported errors weeks later.
+ *
+ *   1. The three `/legal/*` pages set `robots: { index: false }` yet were
+ *      submitted for all five locales — 15 URLs guaranteed to come back as
+ *      "Submitted URL marked noindex".
+ *   2. `nl-BE`, `de-DE` and `es-ES` were submitted for the marketing routes
+ *      while their `landing.*` copy is still French verbatim, asking Google to
+ *      index untranslated pages under a localised URL.
+ *
+ * The glossary is the deliberate exception: `GLOSSARY_LOCALES` scopes it to the
+ * three locales whose terms are actually translated, `nl-BE` included. That
+ * asymmetry is intended — do not "align" it without checking the content first.
+ */
+
+const urls = () => sitemap().map((entry) => entry.url);
+const pathOf = (url: string) => url.replace(/^https?:\/\/[^/]+/, '') || '/';
+
+describe('sitemap — never submits what robots forbid', () => {
+  it('excludes the noindex legal pages entirely', () => {
+    const legal = urls().filter((url) => pathOf(url).includes('/legal/'));
+    expect(
+      legal,
+      'these pages set robots.index=false — submitting them only produces Search Console errors',
+    ).toEqual([]);
+  });
+});
+
+describe('sitemap — only advertises locales that are actually translated', () => {
+  it('limits the marketing routes to the visible locales', () => {
+    const untranslated = routing.locales.filter(
+      (locale) => !(LOCALES_VISIBLE as readonly string[]).includes(locale),
+    );
+
+    const offending = urls().filter((url) => {
+      const path = pathOf(url);
+      // The glossary has its own, deliberately wider, locale scope.
+      if (path.includes('/glossaire')) return false;
+      return untranslated.some((locale) => path === `/${locale}` || path.startsWith(`/${locale}/`));
+    });
+
+    expect(offending, 'marketing copy is French verbatim in these locales').toEqual([]);
+  });
+
+  it('keeps the glossary on its own translated scope', () => {
+    const glossary = urls().filter((url) => pathOf(url).includes('/glossaire'));
+    expect(glossary.length).toBeGreaterThan(0);
+
+    const outOfScope = glossary.filter((url) => {
+      const path = pathOf(url);
+      return routing.locales.some(
+        (locale) =>
+          path.startsWith(`/${locale}/`) &&
+          !(GLOSSARY_LOCALES as readonly string[]).includes(locale),
+      );
+    });
+
+    expect(outOfScope, 'a glossary URL in a locale with no translated terms').toEqual([]);
+  });
+});
+
+describe('sitemap — shape', () => {
+  it('emits no duplicate URLs', () => {
+    const all = urls();
+    expect(new Set(all).size).toBe(all.length);
+  });
+
+  it('advertises alternates only for locales it actually submits', () => {
+    const submitted = new Set(urls());
+    const dangling = sitemap().flatMap((entry) =>
+      Object.values(entry.alternates?.languages ?? {}).filter(
+        (href): href is string => typeof href === 'string' && !submitted.has(href),
+      ),
+    );
+
+    expect(
+      [...new Set(dangling)],
+      'hreflang pointing at a URL absent from the sitemap sends mixed signals',
+    ).toEqual([]);
+  });
+});