Schema and passage writing for AI citations.

FAQPage schema is still worth shipping after May 2026, but only when the page actually contains real, user-facing questions and answers. Google deprecated FAQPage as a search results rich result for all sites on May 7, 2026, but the markup itself still helps AI surfaces directly: AI Overviews, AI Mode, ChatGPT, and Perplexity all continue to lift question-and-answer pairs out of FAQPage-marked content when composing an answer. The real mistake is keeping FAQPage schema on a page that has no genuine question-and-answer content behind it, since that now risks a spammy structured data manual action from Google without any corresponding upside left to justify the risk.

A passage, in GEO terms, is a self-contained paragraph that answers one specific question and would still make complete sense if an AI engine quoted it entirely on its own, with no surrounding context. A typical passage runs 40 to 80 words: it states the direct answer first, then gives the reasoning behind that answer, then adds any qualifiers or exceptions that matter. The opening paragraph of a page and the first paragraph under each H2 heading should both be written as a passage in this sense, since those are the spots an AI engine is most likely to quote directly.

SpeakableSpecification functions as a low-cost hint for AI search rather than a confirmed ranking factor. Adding speakable schema to the summary and key-takeaways sections of a pillar page signals to voice assistants and AI surfaces which specific paragraphs are the best candidates to read aloud or extract preferentially when composing an answer. The implementation cost is minimal, just one small JSON-LD block added to the page, and while the upside is modest rather than dramatic, it is a real, measurable signal rather than a purely theoretical one.

Not every page with a list of steps needs HowTo schema. Google deprecated HowTo as a search results rich result back in September 2023, which already removed most of the direct incentive to add it broadly. HowTo schema should be reserved for genuine step-by-step procedures, such as a cooking recipe, a physical repair process, or a software setup sequence, where the steps are the entire point of the content. For most editorial pages that simply include a numbered list as part of a longer explanation, a plain numbered list inside standard Article schema is the right choice, and misapplying HowTo more broadly is a structured-data warning waiting to happen during an audit.

The @id reference graph is the difference between schema markup that technically exists on a page and schema markup that AI engines and search engines can actually use to compose a clean attribution. Linking Article to its author through an @id reference to a Person entity, and separately linking Article to its publisher through an @id reference to an Organization entity, lets an AI engine understand all three pieces as one connected whole. Without those @id references in place, a search engine or AI crawler sees three separate, disconnected schema blocks on the page; with them properly linked, it sees a single coherent entity graph describing exactly who wrote and published the content.

The fastest way to check schema validity is to paste the page URL or the raw JSON-LD directly into two separate tools: Google's Rich Results Test, and the general Schema Markup Validator at validator.schema.org. The first tool checks specifically for Google rich result eligibility, which is a narrower standard, while the second checks general Schema.org compliance that applies regardless of which specific search engine or AI system ends up reading the markup. Running both tools matters because a schema block can genuinely pass one of these checks while still failing the other for a different reason.

Every page should carry BreadcrumbList schema, with essentially no real exception worth making. BreadcrumbList costs almost nothing to add to a page template, helps Google render a visible breadcrumb trail directly inside the search results listing, and gives AI engines a cheap, clearly structured signal of exactly where a given page sits within the site's overall information hierarchy. For these reasons, BreadcrumbList should be built into the base template that every page on a site inherits automatically, rather than something a content author has to remember to add individually, page by page, as new content gets published.

Google's structured data guidelines treat hidden FAQPage content as a genuine policy violation rather than a stylistic preference, and this specific mismatch can trigger a manual action for spammy structured data against the whole site, not just the one page. The rule followed on this site is strict: any FAQPage JSON-LD on a page must mirror text a visitor can actually read directly on that page, rendered open by default rather than hidden behind a click-to-expand accordion, so the schema never claims to contain content the page does not genuinely and visibly display.