Most Schema Markup Is Written for Rich Results That No Longer Exist
Open the structured data on a typical marketing site and you will find a fossil record. FAQPage markup added in 2021 when it still produced expandable results. HowTo blocks from a guide nobody has touched since. A sitelinks search box definition for a feature Google retired.
None of it is doing harm exactly. It is just doing nothing, while the team believes structured data is handled.
The purpose of schema has quietly changed. It used to be a way to win visual real estate on a results page. Now its most valuable job is telling machines, including the ones assembling AI answers, precisely what your content is about and who stands behind it. At Sympler we rebuild structured data around that second purpose, and the markup that results looks quite different from what most sites are running.
What Schema Does and Does Not Do for AI Answers
Let me be precise here, because this is an area where confident overstatement is common.
There is no public confirmation from any major AI provider that structured data directly causes your content to be cited in a generated answer. Anyone presenting a causal guarantee is going beyond the evidence.
What structured data demonstrably does is remove ambiguity. It states, in a format built for machines, that this page is an article, published on this date, written by this author, belonging to this organization, about this entity. Systems that would otherwise infer those facts from page text get them handed over explicitly.
That matters because ambiguity is expensive for a retrieval system. Faced with two comparable sources, one that clearly declares its author, organization, and subject, and one that leaves those to inference, the clearer source is the safer citation. You are not buying a ranking boost. You are reducing the reasons to skip you.
The Markup That Still Earns Its Place
Strip your implementation back to what actually works today.
Organization: The Foundation Everything Else Hangs From
This is the single most underused type on most sites. Your Organization markup should live on your homepage and declare your legal name, logo, contact points, and, critically, your sameAs references pointing to the profiles you control elsewhere.
Those sameAs links are how you connect scattered mentions of your business into one recognized entity. A brand with clean Organization markup and consistent profiles is much easier for any system to resolve confidently than one leaving the connections implicit.
Article and BlogPosting: Provenance and Freshness
For every editorial page, and particularly the ones you have invested in writing well for search, declare headline, publish date, modified date, author, and publisher. The modified date carries real weight for topics where currency matters, and it only helps if you actually update it when you revise the content rather than letting it drift.
Author markup deserves particular attention. Link the author to a real person or organizational entity with its own presence, not a bare text string. An author credential that resolves to something is worth considerably more than one that resolves to nothing.
BreadcrumbList: Cheap Context
Breadcrumbs communicate where a page sits in your information architecture. That hierarchy tells a system whether a page is a broad overview or a narrow detail page, which affects which questions it gets considered for. It is inexpensive to implement and still supported in search results.
LocalBusiness and Service: For Anyone With a Service Area
If you serve specific places, declare it. Name, address, service area, hours, and the services you offer. Local queries increasingly get answered by short AI summaries rather than long lists, and being unambiguously resolvable is the price of entry.
What to Stop Doing
| Markup | Current reality | Recommendation |
|---|---|---|
| FAQPage | Google restricted these rich results to authoritative government and health sites in 2023 | Keep only if genuinely useful as data, not for visual results |
| HowTo | Deprecated for rich results in 2023 | Remove, keep the human readable steps |
| Sitelinks SearchAction | Google retired the sitelinks search box | Remove |
| Self serving Review | Google does not support reviews you write about yourself | Remove from your own pages |
| Speakable | Limited, publisher focused availability | Ignore unless you are a news publisher |
A note on FAQPage. The rich result is largely gone, but the markup still describes your content accurately for machines. Keeping it is defensible. Keeping it while expecting expandable results in Google is not. Decide on the honest basis.
The Difference Between a Graph and a Pile of Snippets
Here is the distinction that separates implementations that work from implementations that merely validate, and it has nothing to do with which types you chose.
Most sites have fifty pages that each independently assert a company name as a text string. Fifty separate claims, fifty opportunities for a machine to conclude they might be fifty different things. A properly built implementation has fifty pages pointing at one canonical entity, with articles, services and locations attached to it.
Only one of those is a graph, and graphs are what these systems are built to consume. Both versions validate perfectly, which is precisely why this failure persists: every tool you check with says the markup is fine, and the markup is fine, and it is also not doing the job you assume it is doing.
Moving from the first state to the second is a structural exercise rather than a markup exercise. It depends on how your templates are built, how your entity is defined, and how consistently that definition is carried across everything you publish. Our entity SEO work treats this as the backbone rather than a technical afterthought, because it is the part that compounds.
Where Structured Data Projects Go Wrong
Sequencing matters more than effort here. Teams routinely spend a quarter on markup and finish with something no more useful than what they started with, usually for one of three reasons.
They add before they subtract. New markup layered over deprecated types produces a longer fossil record, not a clearer signal. An honest inventory comes first, and it almost always surfaces something nobody remembers adding.
They write markup by hand. Hand written markup drifts within weeks. Anything produced per page rather than per template is a maintenance burden with an expiry date, and it is usually why an implementation that was correct in March is quietly wrong by June.
They validate the source instead of the rendered result. This causes more silent failures than everything else combined. A theme update ships, the markup stops appearing for crawlers, and nobody notices for a quarter because nothing visibly changes on the page. Structured data belongs in your release checklist and in whatever monitoring covers your website infrastructure, for the same reason you monitor uptime: the failure is invisible until somebody goes looking.
How to Tell Whether It Is Working
Set the right expectations before you start, because the feedback loop here is indirect.
Valid markup does not produce a traffic spike. What you should watch instead is whether your brand gets described accurately when systems talk about you, whether the correct pages get associated with the correct topics, and whether your knowledge panel information, where one exists, becomes more complete and accurate over time.
Those are slow, qualitative signals. That is the honest picture. Structured data is infrastructure, and infrastructure is judged by the absence of problems rather than the presence of wins.
Frequently Asked Questions
Does structured data improve rankings?
Not directly. Google has consistently described structured data as a way to help it understand content rather than a ranking factor, and the leaked Search API documents did not change that picture. The indirect benefits are real, including eligibility for supported result types and clearer entity resolution, but treating it as a ranking lever leads to disappointment.
JSON-LD or microdata?
JSON-LD, without hesitation. It is Google’s stated preference, it lives in a single script block rather than being tangled through your HTML, and it survives template changes far better. Microdata implementations break constantly because any markup change risks breaking the annotations.
Should I mark up every page?
Mark up page types, not pages. Your article template, your service template, your location template, your homepage. If you are hand writing markup for individual pages you have built a maintenance burden that will decay within months.
Will FAQPage markup still help with AI answers?
Possibly, though nobody outside those companies can say with confidence. Question and answer pairs are structurally convenient for extraction, so there is a plausible mechanism. What is certain is that the Google rich result is largely gone. Keep the markup if the content genuinely is a set of questions and answers. Do not add fake questions to game a result type that no longer exists.
How often should this be reviewed?
Twice a year for the specification itself, since supported types change, and after every significant template or platform change. The most common failure we see is not bad markup, it is markup that silently stopped rendering during an unrelated deployment.
Get Your Markup Audited
Structured data is one of the few areas where the durable value comes from cleanup rather than addition. It is also one where a site can look entirely healthy from the inside while telling machines something quite different from what its owners intend.
Our team maps current implementations against what actually earns its place today, as a standalone audit or alongside broader AI SEO work and content strategy. Reach out to Sympler and we will show you what your markup is telling machines right now, which is often not what you assume.
