Markup describes; Google decides
There’s a useful distinction getting lost in the latest schema debate: structured data can describe what a page contains and make it eligible for a search feature. It cannot order Google to show that feature. And neither valid markup nor a study finding little change in AI citations is a guarantee about what happens next.
That distinction matters now that familiar search rewards have receded. Google restricted FAQ rich results to authoritative government and health sites, while HowTo results became desktop-only before Google deprecated the feature. The code may still describe an FAQ or a set of steps perfectly well. The old expansive display is another matter entirely.
For site owners, this isn’t a reason to rip out every JSON-LD block. It is a reason to stop treating a rich result as the payoff for adding schema. Eligibility is not display; machine-readable meaning is not a ranking lever. Search Engine Journal’s report captures both the retreat in visible features and the newer argument over schema’s value in AI search.
The visible rewards have narrowed
FAQ markup once offered a conspicuous way for pages to occupy more room in search results. Google later limited FAQ rich results to well-known, authoritative government and health sites. For most commercial and publisher sites, valid FAQ markup no longer means the expanded result is on offer.
HowTo followed a different path. Google limited HowTo rich results to desktop, then deprecated them. That sequence is worth noticing: a feature can become less broadly visible before it disappears from the supported set. A site that maintained step-by-step markup for the search presentation may have watched its practical return shrink well before the final change.
The broader pattern is not that structured data stopped working. It is that the visible presentation is Google’s product, and Google can change eligibility, device coverage, or support. The source report also notes other structured-data features retired over time, with Google describing some as not commonly used in Search and no longer providing value to users. The practical lesson is modest but important: don’t build a durable traffic forecast on one rich-result format.
A valid schema block is not a display command
Structured data’s job is to make information about a page machine-readable. A FAQPage description can express that a page contains questions and answers; HowTo markup can describe steps. That gives a search engine structured context. It does not force the engine to render a special result, and it does not turn a page into something it is not.
This is the line to hold when someone says, “We implemented schema, so why didn’t we get the rich result?” The right first answer is that implementation and appearance are separate events. Correct markup can support eligibility where a feature is available, but the search engine retains control over whether it appears in a particular result.
That sounds obvious until a team attaches a KPI to the code change. If the KPI is “rich result displayed,” the team is measuring a decision it does not control. Better to check whether the markup accurately reflects visible page content, whether it validates, and whether the feature still exists for that page and audience. Then measure the actual search presentation rather than assuming it from a successful test.
The AI-citation study has a narrower answer
The same caution applies to claims that schema will win AI citations. The report discusses Ahrefs’ analysis of 1,885 pages that added JSON-LD, compared with control pages that did not. It tracked citation changes across Google AI Overviews, AI Mode, and ChatGPT.
The reported changes were +2.4% for AI Mode, +2.2% for ChatGPT, and -4.6% for AI Overviews. The first two movements were too small to distinguish from random variation. The AI Overviews decline was statistically significant, but Ahrefs said it could not confidently attribute that decline to schema.
That is not proof that schema never helps. It is also not evidence that schema reliably generates citations. The study’s pages already had more than 100 AI Overview citations before adding markup, so it does not answer whether schema helps a page that has not yet been cited. It pooled schema types—Article, FAQ, Product, HowTo, and Organization—rather than isolating their individual effects. Its 30-day window may miss slower changes, and simultaneous page edits complicate attribution. The report examined schema in page HTML, not JavaScript-injected schema, which AI crawlers may treat differently.
So the careful conclusion is unglamorous: this dataset did not show a clear citation lift for the pages tested. It cannot promise that markup earns AI citations, and it cannot settle every possible use case. Don’t turn either the positive percentages or the negative number into a universal rule.
Keep schema for the job it can do
A sensible schema strategy starts with the page, not the promise of a search feature. Mark up information that is actually present and useful to describe. Keep it accurate as the page changes. Use supported types where they fit, but don’t add fields just to chase a display that may be restricted or retired.
For a site already carrying FAQ or HowTo markup, the decision need not be dramatic. Ask whether it remains accurate and whether it serves a practical purpose beyond the former rich result. If it does, retaining it may still make sense. If it exists only because someone expected a guaranteed search expansion, revisit the maintenance cost and the expectation—not the basic distinction between description and display.
The source report notes that Google’s broader guidance still presents structured data as a way to make page information machine-readable, and that Google Search Relations representatives at a 2025 Search Central Live event in Madrid said supported structured data types remained worth using. That is compatible with a skeptical view of overpromising. “Worth using” does not mean “will appear” or “will be cited.”
Measure what happened, not what the markup promised
Separate implementation checks from outcome checks. A validator can help identify errors; it cannot certify that Google will show a rich result. Search monitoring can tell you whether a feature appeared for your pages; it cannot turn one appearance into a permanent entitlement. AI citation tracking can describe observed changes, but a before-and-after result alone may not establish that schema caused them.
When testing a change, record what pages changed, what markup types were added, what else changed on those pages, and the period being measured. Avoid blending FAQ, Product, Article, and HowTo into one verdict if the question is about a particular type. The Ahrefs report’s pooled design is precisely why its findings should be read as a bounded result, not a ruling on every schema type.
Google controls the presentation. Publishers control whether their markup is honest, current, and worth maintaining. That’s a less exciting promise than “add schema, get cited.” It is also the one the evidence supports.
markup describes; google decides
markup describes; google decides