IMPLEMENT · Chapter 4

Schema and passage writing for AI citations.

Two layers decide whether a page gets cited inside an AI answer: the structured-data layer that tells the model what the page is and who owns it, and the prose layer that gives the model a quotable paragraph. Get both right and citation share follows. Get either wrong and the page is invisible to AI surfaces no matter how high it ranks.

KEY TAKEAWAYS

Article + Person + Organization is the minimum schema stack. Everything else is situational.
FAQPage and HowTo still help AI surfaces — just not the classical SERP. Keep them only on pages with real Q/A or real steps.
Passage shape beats word count. A 600-word page of three clean passages can out-cite a 2,500-word page of marketing prose.
The @id reference graph is what turns three separate schema blocks into one entity-graph the model can attribute to.

Schema decision matrix for 2026

Type	SERP value	GEO value	SMB verdict
Article	No rich result by itself, but powers the article carousel and Top Stories eligibility.	Strong — gives AI models headline, author, dateModified, and publisher in one block.	Ship on every editorial page. Pair with Person and Organization references via @id.
FAQPage	Deprecated as a SERP rich result for all sites on May 7 2026 (Google Search Central).	Still useful — AI Overviews, AI Mode, ChatGPT, and Perplexity continue to lift Q/A pairs out of FAQPage-marked content.	Keep FAQPage when there is real Q/A on the page. Do not invent Q/A just to add it. Do not add it to pages without genuine Q/A — that risks a spammy structured data manual action.
HowTo	Deprecated as a SERP rich result in September 2023 (still parsed by some engines).	Useful — gives AI models a clean step list with images, durations, tools.	Use only for genuine step-by-step procedures. Otherwise prefer a numbered list inside Article.
Person + Organization	Powers Knowledge Panel resolution. The bedrock entity layer.	Critical. AI engines weight author and publisher identity heavily when deciding which sources to cite.	Ship Person for every author, Organization for the business, with @id references that link the two. Add sameAs to every credible third-party profile.
SpeakableSpecification	Optional add-on inside Article — no standalone rich result.	Hints to voice and AI surfaces which passages to read aloud or extract. Low cost, real upside.	Add to the TL;DR or key-takeaways section of pillar pages.
BreadcrumbList	Replaces URL display in the SERP with a breadcrumb trail.	Helps AI engines understand the site hierarchy and group related pages.	Ship on every page that lives inside a hub or category structure.

Source for SERP-side changes: Google Search Central FAQ deprecation announcement (May 7 2026) and the HowTo deprecation from September 2023. GEO value notes are from field testing across AIO, AI Mode, ChatGPT, and Perplexity in 2026.

The passage-shape rules

Four rules for writing paragraphs that AI engines will lift.

Answer first. The first sentence of each passage states the answer. The next two or three sentences support it. A model reading just the first sentence should know what the passage is about.
40 to 80 words. Shorter and there is not enough context to lift. Longer and the model cherry-picks a fragment that may not represent the page's point.
Name the entities. Use full proper nouns at least once per passage. Pronouns and shorthand confuse the model about who or what the passage is talking about when it is quoted out of context.
One verifiable claim per 100 words. A statistic, a date, a named source, a specific number. Pages that hit this density get cited disproportionately because they are the source of facts other pages might also be quoting.

The @id reference graph in practice

On this site, every Article references a single Person node and a single Organization node by stable @id. The pattern looks like this in shorthand.

The Organization lives at https://josecabal.com/#org. Defined once, in the site-wide schema. Carries name, logo, address, telephone, and sameAs to LinkedIn, Wikidata, and every directory profile.
The Person lives at https://josecabal.com/#jose. Defined once, carries jobTitle, image, knowsAbout, and sameAs to LinkedIn, Wikidata, certifications, and bylined publications.
Every Article references both.author: { "@id": "https://josecabal.com/#jose" } and publisher: { "@id": "https://josecabal.com/#org" }.
Result: AI engines see one entity graph for the site instead of three disconnected schema blocks. The attribution that appears next to a citation displays consistently as "José Cabal" and "José Cabal Consulting" instead of varying or being missing.

Best for

Sites that have a clear author or expert behind the content — the entity graph compounds.
Editorial pages, comparison pages, and pillar pages where citation share is the goal.
B2B sites in technical categories where AI engines visibly prefer well-structured sources.

Fails when

Schema is shipped without an @id graph — each block is read in isolation.
FAQPage is added to pages without genuine Q/A in pursuit of a now-deprecated rich result.
Pages are written for keyword density instead of passage extraction.

Verify before

Run the Article through the Schema.org validator and the Google Rich Results test.
Confirm every Article references author and publisher via @id, not inline.
Read the first paragraph of any page out of context — does it stand alone?

Frequently asked

Is FAQPage schema still worth shipping after May 2026?

Yes, when the page actually contains user-facing Q/A. Google deprecated FAQPage as a SERP rich result for all sites on May 7 2026, but the markup still helps AI surfaces — AI Overviews, AI Mode, ChatGPT, and Perplexity continue to lift Q/A pairs out of FAQPage-marked content. The mistake is keeping FAQPage on pages that have no real Q/A, which now risks a spammy structured data manual action without any upside.

What is a passage in GEO terms?

A passage is a self-contained paragraph that answers a specific question and would still make sense if quoted on its own. Typically 40 to 80 words. It states the answer, then the reasoning, then any qualifiers. The lead paragraph of a page and each H2 section should be a passage.

Does SpeakableSpecification matter for AI search?

It is a low-cost hint, not a ranking factor. Adding speakable schema to the TL;DR and key-takeaways sections of pillar pages signals to voice and AI surfaces which paragraphs to read aloud or extract preferentially. The implementation cost is one JSON-LD block; the upside is marginal but real.

Should every page have HowTo schema for any list of steps?

No. Google deprecated HowTo as a SERP rich result in September 2023. Use HowTo only for genuine step-by-step procedures (cooking, repair, software setup). For most editorial pages, a numbered list inside Article is the right answer. Misapplied HowTo is a structured-data warning waiting to happen.

How important is the @id reference graph?

It is the difference between schema that exists and schema that gets used. Linking Article → author (@id) → Person, and Article → publisher (@id) → Organization, lets AI engines compose a clean attribution. Without @id references, a search engine sees three separate schema blocks; with them, it sees one entity graph for the page.