A guest at breakfast spoke into her phone: "Ok Google, what's the best QR menu app?" Google Assistant no longer gave the old rule-based answer — it read out Gemini's LLM summary. The spoken passage was 95 words and almost verbatim from thMenu's speakable schema block.
How Gemini Voice Search Actually Works
Since the 2025 release, Google Assistant has used a hybrid Gemini Nano + Gemini Pro backend. On-device Nano handles intent classification; harder queries hop to cloud Pro. After generation, a TTS engine reads the answer aloud. Pages that ship "speakable" JSON-LD annotations win in this pipeline because the model prioritizes passages marked as suitable for reading aloud.
We added a SpeakableSpecification block to the thMenu landing page, pointing a CSS selector at the first paragraph of the "why-thmenu" section (102 words). Three months later, Search Console's "voice impressions" filter showed a 62% uplift.
Designing the Speakable Passage: The 80-120 Word Rule
TTS read time should land between 25 and 40 seconds — shorter feels half-answered, longer loses the user. When Gemini summarizes, it tends to lift the "readable" block verbatim from the source page. So the passage has to be written for the human ear: short sentences, active verbs, brand name within the first 20 words.
- Brand + value proposition in sentence one.
- Cap sentence length at 18 words — TTS breathes naturally.
- Include a number or stat — Gemini favors specific passages.
Forward Compatibility with ChatGPT Voice
OpenAI Voice Mode uses a similar RAG chain. When we tested the query "voice search optimization 2026," ChatGPT confirmed it recognizes schema.org/SpeakableSpecification and surfaces speakable passages. Everything we built for Google carries over to ChatGPT directly.
Practical takeaway: ship the speakable block once and you appear in Google Gemini, ChatGPT, and Perplexity Voice simultaneously. It's the highest-ROI technical SEO investment for 2026-2027.
FAQ
Isn't speakable schema only for news sites? Google launched it for news, but after Gemini integration it became a ranking signal for every page type — restaurant landings included.
How many passages per page? One to three speakable blocks per page is ideal. More dilutes the "feature this" signal to the model.
Does thMenu add it automatically? Yes. Pro+ plans auto-generate speakable schema for each restaurant landing — the first 100 words of the restaurant description get marked up.
Found this helpful? Share it.
Related articles
Why Digital Menus Increase Restaurant Revenue by Up to 30%
Studies show restaurants using digital QR menus see measurable increases in aver…
When a Customer Downgrades, What Happens to Old Features? — The Silent Feature-Drift Problem in SaaS
Most SaaS apps run a single line of code when a customer downgrades — but old fe…
JWT alg-confusion attack — why Supabase's HS256 → RS256/JWKS migration breaks legacy verifiers
Verifiers that never decode the JWT header are wide open to `alg=none` and alg-c…