On May 7, 2026, Google added a small notice to a developer page. FAQ rich results were gone. No blog post, no announcement, just a line at the top of the documentation retiring a feature that thousands of marketing teams had spent years bolting onto every page they published.
The reaction split immediately into 2 camps, and both were wrong. One declared schema dead. The other declared that FAQ markup was more vital than ever, this time for AI. Neither claim survives contact with the evidence published over the past year.
Here’s the more useful version. Schema markup is a clarity tool, not a visibility lever. It tells machines what a thing is, who wrote it, when it was updated, and how it relates to other things. That helps systems interpret your content correctly. What it does not do is make a page get cited when it otherwise would not. The 2026 research is unusually direct on that point.
That distinction carries weight commercially, because schema work is cheap to over-invest in and easy to get wrong. This guide covers 7 markup types worth implementing for answer engine visibility, the ones that no longer earn anything, and how to tell the difference without relying on vendor claims.
What the 2026 Evidence Actually Says
Three findings from the past 18 months define what is reasonable to believe about schema and AI search.
The correlation is real. The causation is not established. Analyses have repeatedly found that AI-cited pages carry structured data far more often than uncited pages; a gap of roughly 3 to 1 in one widely circulated dataset. But cited pages also tend to sit on larger, better-maintained, more authoritative sites, and those sites implement schema as a matter of course. The markup travels with quality; it does not obviously produce it.
The causal test came back flat. In May 2026, Ahrefs researchers Louise Linehan and Xibeijia Guan published a matched analysis of 1,885 pages that added JSON-LD between August 2025 and March 2026, comparing each against control pages with similar citation levels that did not. Across Google AI Overviews, AI Mode, and ChatGPT, citations barely moved. That is the closest thing the industry has to a controlled experiment, and it points one way.
Platforms confirm interpretation, not preference. Microsoft has stated that schema helps its systems understand content for Copilot, and Google has said structured data offers an advantage in Search. Google has also said plainly that AI Overviews and AI Mode need no special markup. Interpretation and preference are different claims, and only the first is supported.
There is one nuance worth knowing. A controlled test published in early 2026 by Mark Williams Cook embedded an address exclusively inside invalid JSON-LD, invisible on the page, and both ChatGPT and Perplexity surfaced it, suggesting some systems tokenize markup as raw text rather than parsing it semantically.
Interesting, but not a strategy. Invalid markup fails Google’s validation, contributes nothing to entity resolution, and is not a foundation to build on.
Why does the correlation persist if the causation is weak? Because schema is a proxy for site maturity. Companies with structured data usually also have working templates, technical oversight, consistent publishing, and the authority that earns citations.
Measuring the markup is easy; measuring everything travelling alongside it is not. This is why Growth-onomics separates technical hygiene from visibility diagnosis when auditing a site; conflating the two is how teams end up spending a quarter on markup while the actual gap sits in content and third-party authority.
So the honest position is this: implement schema because it removes ambiguity for the platforms that have confirmed they use it, and because it costs little once.
Do not implement it expecting citations to follow.
Quick Comparison
| # | Schema Type | What It Clarifies | Priority |
| 1 | Organization | Who your brand is and where else it appears | Highest |
| 2 | Person | Who your authors and executives are | High |
| 3 | Article / BlogPosting | Authorship, publication date, freshness | High |
| 4 | Product / SoftwareApplication | What the product is, does, and costs | High |
| 5 | BreadcrumbList | Where a page sits in your site hierarchy | Medium |
| 6 | Dataset | That original research exists and how it was made | Medium |
| 7 | ItemList | The structure of roundups and comparison lists | Medium |
1. Organization
If you implement one type, implement this one. Organization markup is the anchor for everything a machine believes about your brand as an entity.
What It Clarifies
Your legal name, alternate names, logo, founding date, location, and most importantly the sameAs property, which links your site to your verified profiles elsewhere: LinkedIn, Crunchbase, G2, Wikidata, and your own social accounts.
Why it’s important for AEO
Entity resolution is the problem sameAs solves. When your brand shares a name with 3 other companies, systems need explicit signals to connect the right profiles, reviews, and coverage to the right entity. Getting misidentified is a worse outcome than being unmentioned, and it is more common than most teams realize.
Implementation Note
Place it once, sitewide, and keep it consistent with what appears on every external profile you list. A sameAs array pointing to abandoned or inconsistent profiles creates confusion rather than clarity. Audit it whenever your company name, positioning, or profile set changes, since stale entity data quietly outlives every rebrand.
2. Person
Person markup gives your authors and executives an identity a machine can track across the web, making expertise signals increasingly important.
What It Clarifies
Who wrote something, what their role and credentials are, which organization they belong to, and where else they publish or appear.
Why it’s important for AEO
Expert commentary is one of the most frequently cited content types in B2B, and citations often attach to a person as much as a brand. Linking an author entity to an organization entity, and both to external profiles, makes it far easier for a system to connect a founder’s podcast appearance, a bylined article, and your product pages into one coherent picture.
Implementation Note
Only mark up real, named people with real credentials and a genuine author page. Generic “editorial team” bylines with Person markup attached add nothing and look like padding.
3. Article and BlogPosting
Article markup is the most routine type on this list and one of the few with a direct line to something answer engines demonstrably care about: recency.
What It Clarifies
Headline, author, publisher, datePublished, and dateModified, plus the article’s primary image and section.
Why it’s important for AEO
Answer engines favor current sources for anything time-sensitive, and dates are how they judge currency. Explicit, accurate dateModified values let a system see that a page was genuinely revised rather than guessing from the copyright footer.
Implementation Note
Do not fake it. Updating dateModified without changing content is a well-known manipulation, and it damages trust with both search systems and readers who notice a “recently updated” page containing three-year-old figures.
4. Product and SoftwareApplication
For SaaS companies, SoftwareApplication (a subtype of Product) is the markup that describes what you actually sell.
What It Clarifies
Application category, operating systems or platforms supported, feature list, and pricing through the Offer property.
Why it’s important for AEO
Capability and cost questions are among the most common things buyers ask assistants, and this markup states the answers in a form that leaves no room for inference. It also reduces the chance of a model attributing a competitor’s limitations to you.
Implementation Note
Every marked-up claim must match visible page content. Pricing in an Offer that contradicts your pricing page is worse than no markup at all. If you do not publish figures, describe the pricing model in visible text and leave the Offer out rather than inventing a number.
5. BreadcrumbList
Breadcrumbs are unglamorous and still worth doing, because they tell a machine where a page sits in your information architecture.
What It Clarifies
The hierarchical path to a page category, subcategory, and the page itself is expressed explicitly rather than inferred from URL structure.
Why it’s important for AEO
Context changes interpretation. A page titled “Permissions” means something different under “Documentation → Security” than under “Blog → Product Updates.” Breadcrumbs make that relationship explicit, which helps systems judge whether a passage answers a technical question or a marketing one.
Implementation Note
Match the breadcrumb to your actual navigation and keep the labels consistent with the terminology used elsewhere on the site. This is a low-effort type to automate at the template level.
6. Dataset
Dataset markup is rare in B2B SaaS and underused by exactly the companies that should have it: those publishing original research.
What It Clarifies
That a dataset exists, what it covers, who produced it, when it was collected, how it is licensed, and where the underlying data can be accessed.
Why it’s important for AEO
Original data is the most citable asset a SaaS company can publish, because it exists in one place and must be attributed. Dataset markup makes the research itself a discoverable object rather than a chart buried inside a blog post, and it documents the methodology signals that make a statistic worth repeating.
Implementation Note
Only use it for genuine datasets like survey results, benchmarks, aggregate product data with a stated methodology and collection period. It is not a wrapper for a blog post containing 3 statistics you found elsewhere. Pair it with a visible methodology section, because the human-readable version is what actually gets quoted.
7. ItemList
Roundups, comparison pages, and “best tools for X” content are among the most heavily cited formats in AI answers, and ItemList describes their structure.
What It Clarifies
That a page contains an ordered or unordered list of distinct items, what each item is, and the position of each within the list.
Why it’s important for AEO
When a buyer asks an assistant for options, the best sources are pages that already contain a considered list. Marking the list structure explicitly reduces the chance that a system misreads a comparison as prose or misattributes an entry to the wrong product.
Implementation Note
Use it where a real list exists and matches the visible page. Pair each item with its own descriptive text; a bare list of names with markup around it is not more useful to a model than the same list without markup.
The Schema Types AI Engines Ignore
This is the shorter and more valuable half of the picture, because unnecessary schema work has a real opportunity cost.
FAQPage, as a visibility lever. Google restricted FAQ rich results in 2023 and retired them entirely in May 2026, with reporting and testing support wound down through the following months. The markup remains a valid Schema.org type and can stay on pages without harm, but expecting it to produce visibility is expecting a feature that no longer exists.
Keep FAQ content; real questions your buyers ask, answered directly because that content earns citations on its own merits. The markup is optional.
HowTo. Deprecated as a rich result and removed from Google’s documentation. Well-structured step-by-step content still performs; the markup around it does nothing.
Speakable. Narrow in scope, limited in support, and irrelevant to the way current answer engines assemble responses for the overwhelming majority of B2B sites.
Self-serving review markup. Review and AggregateRating markup applied to your own product, on your own site, based on testimonials you collected, is not eligible for rich results and does not build credibility with systems that weigh independent sources. Third-party review platforms are where that signal actually lives.
Invented AI properties. Periodically a tool or plugin promises “AI-optimized schema” using properties that do not exist in the Schema.org vocabulary. Systems ignore what they cannot parse, and invalid markup fails validation.
Markup that contradicts the page. The one genuinely harmful category. Structured data describing content that is not visible; prices, ratings, dates, authors risks manual action from Google and misrepresents you to any system that reads it. Every marked-up fact should be verifiable on the rendered page.
A related question comes up constantly: llms.txt. It is not schema, and no major answer engine has confirmed using it. Google has explicitly said no new file or markup is required for AI Overviews. Publishing one costs almost nothing, but treating it as a visibility strategy is unsupported.
How to Implement Without Wasting Time
Schema is a one-time infrastructure job with light maintenance, not a channel. Treat it accordingly.
Do it at the template level. Organization, BreadcrumbList, and Article markup should be generated by your CMS across page types, not hand-added per page. Manual implementation is where inconsistency and decay begin.
Validate everything, then re-validate after releases. Use Google’s Rich Results Test and the Schema.org validator. Broken markup is common after site migrations and template changes, and nobody notices for months because nothing visibly breaks.
Keep markup and visible content in sync. This is the single rule that prevents every serious schema problem. If a fact appears in JSON-LD, it should appear on the page.
Fix crawlability first. Markup on a page that cannot be rendered or indexed accomplishes nothing. When Growth-onomics scopes technical work, structured data sits inside the content and structural refinement step of its AI Optimization framework after the readiness audit confirms that engines can actually reach and parse the pages in question.
Then stop and move on. Once valid schema is in place across your templates, additional markup effort has sharply diminishing returns. The remaining levers are content quality, entity consistency across the web, and third-party authority, which is where the time freed up should go.
Measure the right thing. If you are tracking schema work, track whether your brand is described accurately in AI answers and whether the right entities are being connected, not whether citation counts moved the week after deployment. The Ahrefs data suggests they will not.
Conclusion
Schema markup earns its place in an AEO program as hygiene, in the same category as clean URLs, working canonicals, and a valid sitemap. It removes ambiguity about who you are, what you sell, who wrote what, and when it was last true. Platforms that have confirmed using it interpret your content more accurately because of it.
What it will not do is move a brand from invisible to cited. The 2026 evidence is consistent on that point, and any vendor promising otherwise is selling a correlation as a mechanism. Pages get cited because they answer a real question clearly, carry facts worth attributing, and belong to a brand that credible sources also describe well. Markup helps machines read that; it does not create it.
The practical sequence is simple. Implement the 7 types above once, at the template level, keep them valid and consistent with the visible page, and then redirect the effort toward content that answers what your buyers actually ask and the off-domain presence that supports it.
If you would like a clear read on where your technical foundation ends and your visibility problem begins, the Growth-onomics team can audit both together and show you which one is actually costing you citations.
FAQs
Does schema markup improve AI citations?
Not on its own, based on the strongest available evidence. A May 2026 matched analysis by Ahrefs tracked 1,885 pages that added JSON-LD and found citations barely changed against comparable control pages across Google AI Overviews, AI Mode, and ChatGPT. Correlational data showing cited pages carry more schema is real, but cited pages also tend to sit on stronger sites that implement schema routinely. Schema helps systems classify and interpret content accurately, which is worth having. Treat it as hygiene that supports accurate representation rather than a lever that produces visibility.
Should I remove FAQ schema now that Google deprecated it?
No, removal is unnecessary. Google retired FAQ rich results in May 2026 and has confirmed that unused structured data causes no problems for Search, so existing markup can stay in place. FAQPage remains a valid Schema.org type and continues to be crawled by other systems. What should change is your expectation: FAQ markup is no longer a route to extra search real estate, and there is no evidence it drives AI citations. Keep the FAQ content itself, because real questions answered directly are among the most citable content you can publish.
Which schema type should a B2B SaaS company implement first?
Organization, without much debate. It anchors your brand as an entity and, through the sameAs property, connects your site to your verified profiles on LinkedIn, review platforms, and other authoritative sources. Entity confusion being conflated with a similarly named company is a more damaging and more common problem than most teams appreciate, and this is the markup that addresses it. Person and Article markup follow closely, since they establish authorship and recency, both of which influence how systems judge the credibility and currency of what you publish.
Do ChatGPT and Perplexity actually read structured data?
The honest answer is that we do not fully know. Microsoft has confirmed schema helps its systems understand content for Copilot, and Google has said structured data provides an advantage in Search, but OpenAI and Perplexity have made no equivalent statements about their retrieval pipelines. One controlled test in early 2026 found both extracted an address that existed only inside invalid JSON-LD, which suggests some systems tokenize markup as text. That is a curiosity rather than a strategy. Implement valid schema for the platforms that have confirmed using it and treat the rest as unverified.
Can bad schema markup hurt my AI visibility?
Yes, in one specific way: markup that contradicts the visible page. Structured data claiming prices, ratings, dates, or authors that do not appear on the rendered page risks manual action from Google and misrepresents you to any system reading it. Invalid or broken markup is less dangerous but still wasteful, since it fails validation and contributes nothing to entity resolution. What does not hurt you is unused markup for retired features. Google has confirmed that structured data with no current search appearance causes no problems. The rule that prevents nearly every issue: if it is in the JSON-LD, it should be on the page.
Is llms.txt worth publishing?
It costs little and proves nothing. No major answer engine has confirmed using llms.txt, and Google has explicitly stated that AI Overviews and AI Mode require no special file or markup. If publishing one is a 5 minute task, there is no strong argument against it. There is a strong argument against treating it as a visibility strategy, or against any vendor presenting it as a requirement. The time is better spent making sure your pages are crawlable, your content answers real questions, and your entity information is consistent across the web.
