What changed for FAQ and HowTo markup?
Google restricted FAQ rich results to authoritative government and health sites and reduced HowTo display in 2023, which made many teams abandon both types. That decision optimized for a metric — rich-result stars and dropdowns — while missing what the markup actually does for AI pipelines. AI Overviews, ChatGPT, Perplexity and Gemini all retrieve passages; a page whose questions and answers are explicitly delimited is easier to chunk, embed and match to fan-out sub-queries than an unstructured wall of prose. The decorative value shrank; the structural value grew.
How do these types map onto the AI pipeline?
- Retrieval. Query fan-out turns ‘how do I set up IndexNow’ into sub-queries about keys, endpoints and verification. A HowTo with labeled steps, or an FAQ whose question matches the sub-query, gives the retriever a passage that answers it exactly — the principle behind answer block writing, expressed in markup.
- Grounding. Explicit Q&A structure makes it cheap to verify that your stated answer agrees with the rest of your page and with third-party sources.
- Generation and citation. Well-delimited Q&A passages quote cleanly, which raises the odds your text — and your link — is what gets used.
When should you use each type?
- FAQPage: pages with genuine, visible question-and-answer pairs — product FAQs, support hubs, glossary-style explainers. Each
Questioncarries anacceptedAnswerwhose text must appear on the page. - HowTo: real procedural content with sequential steps — setup guides, recipes, configuration tutorials. Steps, tools, durations and supplies all have fields; use them honestly.
- Neither: marketing pages with fake ‘FAQs’ written to stuff keywords, or editorial articles where Q&A markup would misrepresent the content. Grounding punishes theater.
Implementation checklist
- Mark up only Q&A and steps that are visible to users on the page; hidden markup-only content violates guidelines and fails grounding.
- Keep answers self-contained and answer-first: the first sentence of each acceptedAnswer should resolve the question — see answer block writing.
- Nest FAQPage or HowTo nodes into your single JSON-LD graph — see JSON-LD nesting patterns.
- Date-stamp procedural content and keep steps current; stale instructions are grounding failures waiting to happen (content freshness).
- Measure impact with a fixed 50-100 question set scored monthly on Visibility and Citation Rate — the standard in measuring AI visibility.
Common pitfalls
First, FAQ spam: twenty keyword-shaped questions with thin answers, which now earns no rich results and degrades page quality signals. Second, markup-content mismatch: the acceptedAnswer text differs from the visible answer, or questions marked up that users cannot see — both are policy violations and grounding contradictions. Third, marking up user-generated Q&A threads as FAQPage instead of QAPage. Fourth, abandoning the types entirely because rich results shrank — the AI citation layer still benefits from the structure even when the classic SERP shows nothing. Fifth, expecting markup alone: a mediocre answer in perfect schema still loses to a great answer in plain HTML; markup amplifies, it does not substitute.
How do you write questions AI actually retrieves?
Source them from real demand: sales emails, support tickets, community threads and the fan-out trees around your money queries — see query fan-out. Phrase each question the way a user asks an assistant, not the way a keyword tool abbreviates it: ‘how long does IndexNow take to update Bing’ beats ‘IndexNow timing’. Keep each answer self-contained — forty to eighty words for the core answer, with evidence after the direct statement. Resist volume: eight sharp FAQs outrank forty padded ones, because padding dilutes the page-level relevance that retrieval computes first. Curate quarterly; an FAQ that answers last year’s questions teaches engines your page is about the past.
Where to go next
Strengthen the underlying content with citation-worthy content, map question coverage via query fan-out, and see the full stack in the structured data deep dive.