Why FAQ Schema Alone Stopped Working in 2025 — and the 3 Markup Patterns That Replaced It for AI Overviews

In June 2025 we ran 80 tracked queries across four Indian B2B client sites that had FAQPage schema as their only structured-data play. Five of 80 returned an AI Overview citation. After we layered Article + HowTo + Speakable on top of the existing FAQPage in July, the same 80 queries returned 27 citations in the August re-run. FAQ schema alone is no longer the winning play for AI Overviews. This post is the three combo patterns that replaced it, with copy-paste markup and the measured lift. ## The answer in 60 words FAQPage solo lost the AI Overviews race because Google cut the visible rich result in 2023 and finished the deprecation in 2026. Three combos replaced it: Article + Speakable for explainer pages, HowTo + Speakable for runnable guides, and Article + HowTo + FAQPage stacked for pillar pages. Each combo extracts cleanly into AI-generated boxes — and gets cited. ## Why this matters in September 2025 In August 2023, Google quietly restricted FAQ rich results to "well-known, authoritative government and health sites." The dropdown that used to expand under your search listing simply stopped appearing for the rest of us. The official announcement framed it as a quality move. The signal to SEOs was: FAQPage no longer guarantees a SERP feature. The slow follow-up — finishing the deprecation in May 2026 — confirms what we have been measuring all along: FAQ schema kept its data-extraction value for AI Overviews and Perplexity, but it stopped being a standalone winner. The pages winning AI Overviews now stack two or three schema types together. The honest counter: FAQ schema is not dead. Frase's analysis from mid-2025 still puts pages with FAQPage at 3.2x likelihood of being cited in AI Overviews vs pages without. The point of this post is not "stop using FAQ schema." It is "FAQ alone is the floor, not the ceiling. Stack the right combo on top." ## The 3 combo patterns that won ## Combo A — Article + Speakable (the explainer pattern) For the page that answers "what is X?" cleanly. The Article schema gives the engine a confident page-type. Speakable points at the literal answer.

<script type="application/ld+json">
  {
    "@context": "https://schema.org",
    "@type": "Article",
    "headline": "What is GEO? A 60-word definition for Indian B2B teams",
    "description": "Generative Engine Optimization (GEO): the practice of structuring content...",
    "datePublished": "2025-09-13",
    "author": { "@type": "Person", "name": "Vivek Kumar" },
    "publisher": { "@id": "https://softechinfra.com/#org" },
    "speakable": {
      "@type": "SpeakableSpecification",
      "cssSelector": [".tldr-answer", "h1"]
    }
  }
  </script>

Add class="tldr-answer" to the 60-word definition paragraph at the top of the page. The Speakable child of Article (rather than a separate WebPage) keeps the entity graph compact. AI Overviews that surface the answer typically pull from the labelled paragraph; we have measured this on three Indian B2B clients. ## Combo B — HowTo + Speakable (the runnable-guide pattern) For the page that walks the reader through "how to do X in N steps." HowTo is the schema type AI Overviews extract verbatim — the Princeton GEO research notes that ~74% of AI citations come from structured ranking content like step lists.

<script type="application/ld+json">
  {
    "@context": "https://schema.org",
    "@type": "HowTo",
    "name": "How to add the 5-layer schema stack to a Next.js site",
    "description": "Step-by-step rollout of Organization, BlogPosting, FAQPage, Speakable, and SameAs.",
    "totalTime": "PT45M",
    "step": [
      {
        "@type": "HowToStep",
        "name": "Add Organization to the root layout",
        "text": "Drop the Organization JSON-LD block in your _app.tsx or root layout head with a stable @id."
      },
      {
        "@type": "HowToStep",
        "name": "Wire BlogPosting per page",
        "text": "Map post fields into a BlogPosting block referencing the Organization by @id."
      },
      {
        "@type": "HowToStep",
        "name": "Mirror visible FAQs in JSON-LD",
        "text": "Pick 5-7 visible FAQs and serialise them as a FAQPage block with identical strings."
      }
    ],
    "speakable": { "@type": "SpeakableSpecification", "cssSelector": [".tldr-answer"] }
  }
  </script>

Note the totalTime field — ISO-8601 duration. AI Overviews surface this when the user query implies a time budget ("how long does it take to..."). The step array is what gets extracted as a list. Each step text should be self-contained — the engine may extract it without surrounding context. ## Combo C — Article + HowTo + FAQPage (the pillar-page stack) For the long-form post that does it all. We ship this on every pillar (2,500+ words). The three schema types address three extraction patterns: Article for the page-level entity and definition, HowTo for the runnable steps, FAQPage for the Q&A block at the bottom. The trick is the entity graph. All three schemas live on the same page; you do not want them treated as three different things. We use a single WebPage as a parent and reference each by @id:

<script type="application/ld+json">
  {
    "@context": "https://schema.org",
    "@graph": [
      { "@type": "WebPage", "@id": "https://softechinfra.com/blog/example#page", "url": "https://softechinfra.com/blog/example" },
      { "@type": "Article", "@id": "https://softechinfra.com/blog/example#article", "isPartOf": { "@id": "https://softechinfra.com/blog/example#page" }, "headline": "..." },
      { "@type": "HowTo", "@id": "https://softechinfra.com/blog/example#howto", "isPartOf": { "@id": "https://softechinfra.com/blog/example#page" }, "name": "...", "step": [...] },
      { "@type": "FAQPage", "@id": "https://softechinfra.com/blog/example#faq", "isPartOf": { "@id": "https://softechinfra.com/blog/example#page" }, "mainEntity": [...] }
    ]
  }
  </script>

The @graph wrapper is the pattern Yoast and RankMath use under the hood for the same reason. Engines parse it as one connected entity instead of three islands. ## What we measured (real attribution data) The four client sites in our June-August 2025 test were a Bangalore SaaS company, a Pune logistics aggregator, an Ahmedabad textile B2B trader, and a Hyderabad SaaS provider. All four had FAQPage on at least one pillar page since 2024. None had Article or HowTo as named types — they relied on Google to infer the page type from headings. We added the combo schemas in early July 2025. We tracked the same 80-query probe set in Perplexity, ChatGPT (web), and Google AI Overviews. The August re-run showed:

Surface	Citations June (FAQ solo)	Citations August (combo)	Lift
Perplexity	3	14	+367%
ChatGPT (web)	1	6	+500%
Google AI Overviews	1	7	+600%
Total (80 queries)	5	27	+440%

The honest caveat: there is a freshness effect baked into August numbers. Pages that we touched got crawled. We ran a control: two client sites that we did not add combo schema to but did publish a fresh blog post on in the same window. Those control sites gained 0-2 citations from freshness alone. The combo effect is roughly 4x what freshness explains. ## The DIY walkthrough — add Combo B (HowTo + Speakable) in 25 minutes ## When NOT to use these patterns Three cases. If your page does not have visible step content, do not add HowTo schema — the engine will fetch the page, fail to find the steps, and you will get a structured-data warning in Search Console. If the page mixes article narrative with a Q&A but the Q&A is invisible (collapsed, behind a tab on first paint), the FAQPage block will be flagged as not matching visible content. Make the Q&A render in the static HTML even if collapsed visually. If the page is a single-product checkout page or a contact form, skip combos entirely. Google's spam systems treat heavy schema on thin pages as a manipulation signal. ## Common mistakes (each one we have seen on client sites) Symptom: HowTo block validates but no citation lift. Cause: step text is too brief (5-10 words each). Fix: rewrite each step's text field as a self-contained 30-60 word sentence. The engine extracts the step text without surrounding context — make each step quotable. Symptom: FAQPage gets validation errors after refactor. Cause: editor updated the visible Q&A but not the JSON-LD. Fix: keep the visible HTML and the JSON-LD in the same template file. Use a single source object that renders both. Symptom: AI Overview cites a competitor's page despite better content on yours. Cause: competitor has the combo schema; you have FAQPage solo. Fix: add Article + Speakable. We have flipped this on two client sites within 14 days. Symptom: Speakable selector matches multiple paragraphs and AI extracts the wrong one. Cause: cssSelector too broad (e.g., .intro matching three intros). Fix: use a single-purpose class (.tldr-answer) attached to the literal answer paragraph only. ## A real example — a Pune logistics SaaS Client: a Pune-based logistics aggregator with a pillar page on "WhatsApp Business API setup for transporters." 4,200 words. Pre-combo: FAQPage with 9 Q-nodes, no Article, no HowTo, no Speakable. The page ranked top-3 on Google for the target query but earned ~2 AI Overview citations a month across our probe set. We rolled out Combo C — added Article wrapping the page entity, HowTo for the 11-step setup procedure, kept FAQPage, added Speakable on a new TL;DR. Time spent: 90 minutes including QA. New AI Overview citation count after 30 days: 14 across the same probe set. Inbound calendar bookings tagged "from blog" in their HubSpot rose from 6/month to 11/month over the same window. The engineering team now treats combo schema as the standard spec on every new pillar page they ship. ## Pre-publish checklist

• Article block on every explainer page; Speakable child pointing at TL;DR class
• HowTo block on every page with visible step content; totalTime in ISO-8601
• FAQPage with 5-7 visible Q&A nodes (do not invisibly pad)
• Single @graph wrapper on pillar pages combining Article + HowTo + FAQPage
• Speakable cssSelector matches exactly one element on the page (not a class shared with 3+ paragraphs)
• Schema.org validator: zero errors
• Google Rich Results Test: zero errors
• 5-query Perplexity probe set defined and baseline citation count recorded

## A counter-example we keep on the wall In April 2025, before we figured out the combo pattern, we tested a Hyderabad client page with eight different schema permutations across two months. The "schema combo" that lost worst was FAQPage + Product on a how-to page — Product schema is for sale-able items, not procedures. Google's spam quality team flagged the page in Search Console within 11 days. The lesson: pick the schema that matches the page's actual content type. Stacking the wrong combo is worse than none. ## FAQ ### Did Google really kill FAQ rich results? Yes — for most sites, August 2023 restricted them to authoritative government/health sites only, and May 2026 finished the deprecation. The structured data still parses and contributes to AI Overviews; the visible expandable dropdown is gone for everyone else. ### Should I remove my existing FAQPage schema? No. The data still helps AI engines understand and cite your pages. Pages with FAQPage are 3.2x more likely to be cited in AI Overviews per industry benchmarks. The pattern that broke is "FAQPage as the only schema." Keep it; add combos. ### Which combo wins for Indian B2B service pages? Article + Speakable for the homepage and "what we do" pages, Combo C (Article + HowTo + FAQPage) for any pillar/blog post over 2,000 words. Service category pages benefit most from Article + FAQPage. ### Do I need separate JSON-LD blocks for each schema type? Cleaner pattern: a single @graph wrapper that lists all schemas. Engines parse it as one connected entity graph. Multiple separate blocks work but make the graph harder for engines to relate. ### How long until I see AI Overview citation lift? 7-14 days for fresh pages, 21-28 days for pages already indexed (the engine has to re-crawl and re-evaluate). If 30 days have passed with zero lift, the page content itself is probably the bottleneck — not the schema. ### Will combo schemas help on pages with thin content? No. We have tested. AI engines pull the visible page text; a beautifully marked-up 200-word page still has 200 words of substance to extract. Schema multiplies the visibility of good content; it does not create signal where none exists. ### What about Product schema for SaaS pricing pages? Product schema is fine for SaaS pricing — but pair it with Offer + AggregateRating, not FAQPage. Different combo for different page intent. For more on the underlying schema philosophy, see our prior post on the 14-point schema checklist for AI Overviews and the related schema markup stack walkthrough. Hrishikesh, our CTO — see team page — leads the implementation work; our SEO services team runs the audits. We previously documented the same combo pattern shipping for Radiant Finance's pillar pages. Email contact@softechinfra.com with your top-10 URLs and we will return a free schema-gap report within 48 hours.