Structured data and AI answers: what actually gets used
Markup does not persuade an answer engine to quote a page. It removes the guesswork about who wrote the page, when, for whom, and where it sits on the site.
Structured data does not persuade an AI answer engine to quote a page, and no schema type exists that makes a citation more likely on its own. What markup does is remove guesswork: it states the headline, the author, the publisher, the dates and the page's position in the site in a form that parses identically for Google, Bing and every model that reads the page. On a software blog four types still earn their place — BlogPosting or Article, BreadcrumbList, Organization with WebSite, and SoftwareApplication where the product genuinely is software.
An answer engine takes facts from structured data, not judgements. A BlogPosting node says this page has this headline, this author, this publisher and this date; a reader of the HTML would have to infer all four from layout, and different readers infer differently. The value is consistency rather than emphasis: the same four facts come back the same way whether the page is being indexed, summarised or quoted. Nothing in the vocabulary lets a page assert that it is authoritative, and a node claiming so would be ignored by every consumer.
The prose still carries the argument. A model quoting a page lifts a passage from the body, and the markup only tells it whose passage it is. That is why markup is a finishing step rather than a strategy, and why a page that does not rank for a query is not helped into an answer by adding JSON-LD to it.
Four schema types still change something for a software blog, and two widely recommended ones no longer do. Google restricted FAQ rich results on 8 September 2023 to well-known government and health sites, and withdrew HowTo in the same announcement before removing it entirely. FAQPage markup on an ordinary site is still valid and still parses; it just draws no rich result, which makes "add FAQPage and get the expandable questions in search" the most persistent piece of stale advice in this area.
A cache is invalidated when its source changes. A block of JSON-LD pasted into somebody's codebase is not. The only thing that can notice is a record of what was handed out.
Market4 turns one release note into a changelog page, a blog post, a mail-out and a week of social posts — and then tells you which of them brought anyone back.
| Type | What it still does | What it does not do |
|---|---|---|
| BlogPosting / Article | Carries headline, author, publisher and dates in a parseable form | Draw a rich result of its own for most sites |
| BreadcrumbList | Changes the visible result row by showing the trail instead of a bare URL | Affect ranking or crawl priority |
| Organization + WebSite | Feeds entity reconciliation through sameAs, and sitelinks | Create a knowledge panel on request |
| SoftwareApplication | States what kind of software the product is, with a real category | Earn the app rich result without a rating we do not have |
| FAQPage | Publishes question-shaped answers as machine-readable Q and A | Draw expandable questions in search, since September 2023 |
Markup that contradicts the visible page is treated by Google as misleading structured data, and the penalty is not scoped to the page that carried it. A manual action for misleading markup removes rich-result eligibility for the whole site, which is a poor trade for a headline that was tidied up in the JSON-LD but not in the h1. The rule that avoids the entire category is short: every value in the markup should be a value a visitor can see on the page, and a fact the page does not display should not appear in its graph.
Generated markup is safer than hand-written markup for this reason alone. When the JSON-LD is built from the same record that renders the page, the headline in the graph cannot drift from the headline in the heading. Hand-written markup drifts the first time somebody edits a title and forgets the script tag underneath it, and nothing in a browser or an editor makes that drift visible.
No. Structured data makes a page's facts machine-readable, which helps a model attribute a passage correctly once it has already chosen to use the page. Selection happens on the strength of the prose and the page's standing in search results. Marking up a page that no engine currently retrieves changes what is available to be quoted, not whether anything is quoted.
It is worth adding for machine readability and worthless as a rich-result play. Since 8 September 2023 Google has shown FAQ rich results only for well-known government and health sites, so an ordinary site gets valid markup and no visible change in search. The reason to keep emitting it is that question-and-answer pairs are the easiest shape for an answer engine to lift whole.
No. Google withdrew HowTo rich results in the same August 2023 announcement that restricted FAQ, and later removed the feature entirely. There is nothing left to opt into, so the markup adds bytes and maintenance with no counterpart in search. Write the steps as an ordered list in the page instead, which is what both readers and models use.
Google classifies it as misleading structured data and can issue a manual action, which removes rich-result eligibility for the entire site rather than for the offending page. The common cause is not deception but drift: a headline edited in the body and left alone in the JSON-LD. Generating markup from the same record that renders the page removes the failure mode.
Breadcrumb markup is the schema type that visibly changes a search result. It needs at least two rungs, three properties per rung, and a trail the page really shows.
No setting anywhere grants a citation. What a small site controls is the shape of its passages and its own server logs. What makes a paragraph cheap to quote, how to tell whether assistants are fetching your pages, and why a cited page can still show zero referrals.