Why French B2B sites underperform on AI Overviews — and how we fix them
3 May 2026 · 5 min read · TheAIgency
TL;DR. AI Overviews, ChatGPT Search, Perplexity, and Claude all favour language-matched, server-rendered, schema-tagged sources. Most French B2B sites in MA and the EU are English-translated post hoc, missing hreflang, and rendered client-side. The fix is four things: server-render the right language, use hreflang properly, add a language-locked answer capsule under each H1, and translate (not transliterate) schema.
The French B2B gap
Roughly 70-80% of Morocco's B2B commercial-intent searches are in French (per Web Success Morocco's 2026 SEO benchmark). The same is true across the rest of francophone EU and MENA. But most B2B sites that target this audience have one of these problems:
- The English version is the canonical, the French is a translated afterthought.
- Hreflang is missing or mis-paired (one-way mapping).
- Language detection is JS-based — the server renders the wrong language to crawlers.
- Schema is left in English even on French pages.
AI search engines penalise all four. ChatGPT Search runs on Bing's index and Bing's relevance model heavily favours language match. Perplexity's citations show a strong preference for sources whose lang attribute matches the query language. Google AI Overviews extract less reliably from translated-not-localised content.
How we fix it on every site we ship
| Fix | What we do | Why it matters |
|---|---|---|
| Server-render the right language | next-intl with locale-prefixed routes (/fr, /en); SSR locks language at the request level | Crawlers and AI agents see the right language without running JS |
| Hreflang done properly | Per-URL alternates.languages in the sitemap + <link rel="alternate" hreflang> in head, including x-default | Google + Bing correctly cluster the FR/EN pages and serve the right one per user |
| Language-locked answer capsule | Every page leads with an 80-word TL;DR in the matching language, before any prose | That capsule is the chunk AI Overviews and ChatGPT cite — language match drives extraction |
| Schema translated, not transliterated | FAQPage, Service, Article schemas all have FR copy on /fr URLs (not just the prose) | Schema is parsed by AI engines as ground-truth metadata |
What this looks like in practice
On theaigency.ma, every page has both an /en and an /fr URL with content authored in each language (not auto-translated). The sitemap declares the alternates. The schema on /fr pages is in French. The TL;DR pattern is enforced.
The result: in Perplexity citations for "agence IA Maroc"-style queries, language-matched pages dominate. We see this on our own site and on every client site we've shipped under our Web line.
The bonus that comes free
Doing this properly also unlocks bilingual organic Google performance. The same /fr page that wins AI Overviews for French queries also ranks better in Google.fr classic results. There's no trade-off.
What to do if your site is already English-first
The minimum viable fix in priority order: (1) add proper hreflang on every URL, (2) hand-translate (don't auto-translate) the top 10 commercial pages, (3) put a French TL;DR at the top of those pages, (4) translate the schema. That alone usually moves citation share within 4-6 weeks.
If you want this audited and shipped, that's what our Signal tier covers by default.