Schema markup will not force ChatGPT or Claude to recommend you. But the right structured data removes ambiguity, makes your facts easy to extract, and gives assistants cleaner signals to reason over. Here is which types matter and how to use them well.
Daniel Arons · Jun 2026 · 7 min read
Structured data has been part of the SEO toolkit for over a decade, mostly as a way to earn rich results in Google: star ratings, FAQ accordions, recipe cards, and the rest. The rise of AI assistants has quietly changed the stakes. When someone asks ChatGPT for the best project management tool for a small agency, or asks Claude which CRM fits their stack, the model assembles an answer from whatever it can read and trust about you. Schema markup is one of the cleanest ways to hand a machine the facts it needs without making it guess.
It is worth being precise about what schema can and cannot do, because the hype usually overshoots. Markup does not contain a magic property that makes a model recommend you. What it does is reduce ambiguity. It tells a parser, in an explicit and standardized vocabulary, that this string is your company name, that number is a price, this block is a customer review, and these are the questions your product answers. That clarity makes your content easier to extract and harder to misread. It is a strong supporting signal, not a switch you flip. This guide covers which schema.org types are worth your time, what each one communicates, how to implement them cleanly, and where structured data fits in the larger picture of getting recommended by AI.
Why schema matters for AI recommendations at all
Large language models do not browse your site the way a human does. When an assistant retrieves information about you, whether from training data, a live web fetch, or a connected search index, it works with text and markup, not the rendered visual page. Schema.org markup, usually expressed as JSON-LD in a script tag, sits in that text as a structured, labeled summary of what the page is about. Instead of inferring that 'Aethon AI' is an organization from surrounding sentences, a parser can read it directly from an Organization object. Instead of hunting for pricing in prose, it can read a Product with an explicit offer.
This matters because extraction errors are quiet and costly. If an assistant misreads your category, conflates you with a similarly named company, or pulls an outdated price, it can pass that error straight into a recommendation. Clean, accurate markup lowers the odds of that happening. It does not guarantee a citation, but it makes the version of you that the model reasons over more accurate and complete. For the prose half of the same problem, our guide on how to write content AI will cite picks up where markup leaves off.
“Schema does not make a model recommend you. It makes the version of you that the model reads more accurate, more complete, and harder to misinterpret.”
The schema types that earn their place
You do not need every type in the schema.org vocabulary. A small set of well-chosen, accurate types does far more than a sprawling implementation full of half-filled fields. Here are the ones most relevant to brands that want to be understood and recommended.
Organization
Organization is the foundation. It establishes who you are as an entity: your legal name, logo, official URL, and the social and external profiles that corroborate your identity through the sameAs property. This is how you reduce the risk of being confused with another company that shares part of your name. Linking out to your LinkedIn, Crunchbase, or Wikipedia entry through sameAs helps anchor your identity across the web, which matters when a model is trying to decide whether two mentions refer to the same organization. Place this once on your homepage or in a sitewide header.
Product and Service
Product describes a specific thing you sell, with properties for name, description, brand, and offers, which can carry price and availability. Service is the parallel type for offerings that are not physical products, with properties like serviceType and the provider that delivers it. For software and B2B companies, the choice between them depends on how you frame what you sell. The value is concrete: when an assistant compares options in your category, an explicit Product or Service object gives it labeled facts to compare rather than prose it has to interpret. Only include offers data you will keep current, because stale pricing in markup is worse than none.
FAQPage
FAQPage marks up a list of questions and their answers using the Question and Answer types. This maps almost perfectly onto how people query AI assistants, because both are question-and-answer shaped. Well-structured FAQ markup gives a model clean, self-contained question-answer pairs it can lift directly. Use it only for genuine questions with genuine answers visible on the page. Do not stuff it with marketing copy disguised as questions, and make sure the marked-up text matches what a visitor actually sees.
Review and AggregateRating
Review captures an individual review with its author, rating, and body, while AggregateRating summarizes many reviews into an average and a count. Social proof is a real input into recommendations, and structured review data makes that proof legible to a parser instead of leaving it buried in testimonial sections. The rule here is strict honesty: the markup must reflect real reviews that are genuinely present and verifiable. Fabricated or self-generated ratings are a credibility risk.
BreadcrumbList
BreadcrumbList describes where a page sits in your site hierarchy. It looks humble next to the others, but it does quiet, useful work. It communicates the relationship between your pages, helping a system understand that a deep article belongs to a broader topic cluster and that your site has organized depth on a subject. That context helps an assistant treat you as an authority on a theme rather than as a scattering of unrelated pages.
Implementing schema cleanly
The mechanics matter less than the discipline, but a few practices separate markup that helps from markup that quietly hurts. Use JSON-LD as your format. It keeps your structured data in a single script block rather than scattering attributes through your HTML, which makes it easier to maintain and audit. The broader ecosystem treats it as the preferred format, and it is the cleanest to generate programmatically.
The cardinal rule is that your markup must match your visible content. Schema is meant to describe what is on the page, not to assert claims that a visitor would never see. If your FAQ markup contains answers that do not appear anywhere on the page, or your AggregateRating shows a number your reviews do not support, you are creating a mismatch that erodes trust and can be penalized. Mark up what is real and present, nothing more.
Keep your data current. Pricing, availability, and ratings change, and markup that drifts out of sync with reality is a liability. If you cannot commit to maintaining a field, leave it out. Validate everything before you ship it, then keep an eye on it after launch. And resist the urge to mark up every conceivable type. A focused implementation of Organization, your Product or Service, and a clean FAQPage will serve you better than a sprawling one stuffed with sparse, half-accurate objects.
Common mistakes that undercut your markup
Most schema problems are not exotic. They are the same handful of errors repeated across thousands of sites. The first is marking up content that is not on the page, which crosses from description into misrepresentation. The second is letting data go stale, especially prices and ratings, so your markup quietly contradicts your live site. The third is over-marking: applying types that do not fit, nesting objects incorrectly, or filling required properties with placeholder values just to satisfy a validator.
A subtler mistake is treating schema as a substitute for substance. Markup is a wrapper around your facts; it cannot manufacture authority, clarity, or trust that the underlying content lacks. If your page does not clearly explain what you do, wrapping it in perfect JSON-LD will not fix the ambiguity. The structured data and the prose have to agree and reinforce each other. A final common error is implementing once and never checking again. Schema breaks silently when templates change or content is updated, so periodic validation should be part of your routine.
“Markup is a wrapper around your facts. It cannot manufacture authority that the underlying content lacks, and it cannot fix a page that does not clearly say what you do.”
Where schema fits in the bigger picture
Structured data is one signal among many. Getting recommended by an AI assistant draws on your overall authority, the clarity and depth of your content, how often and how favorably others mention you, and the consistency of your facts across the web. Schema strengthens several of those inputs by making your facts explicit and machine-readable, but it works in concert with them, not in place of them. A site with thin content and immaculate markup will still struggle, because the markup has little of substance to describe.
This is the practical core of what answer engine optimization tries to address, and it connects to the broader idea behind contextual AI presence mapping: understanding how assistants currently perceive and represent your brand, then improving the underlying signals so that perception is accurate and favorable. Schema is one lever in that system. It is high-leverage because it is concrete and within your control, but it is most powerful when the content it describes is genuinely strong and your facts are consistent everywhere a model might find them.
Treat schema markup as table stakes rather than a silver bullet. Implement Organization, your relevant Product or Service, an honest FAQPage, and any Review or BreadcrumbList data that genuinely applies. Keep it accurate, keep it current, and make sure it always matches what a visitor sees. Then put your real energy into the substance underneath, because that is what an assistant ultimately weighs when it decides whether to recommend you. If you want to see how an assistant currently reads and represents your brand, and where your signals are strong or ambiguous, you can request a demo or read more about how Aethon works.
Frequently asked questions
Does schema markup guarantee that ChatGPT or Claude will recommend my brand?
No. Schema markup removes ambiguity and makes your facts easier for a parser to extract and trust, but it does not contain any property that forces a recommendation. It is a supporting signal that works alongside content quality, authority, and consistency, not a switch you flip.
Which schema types should a B2B software company prioritize?
Start with Organization to establish your identity, then add Service or Product to describe what you sell, and a genuine FAQPage for the questions your buyers actually ask. Add Review or AggregateRating only if you have real, verifiable reviews, and BreadcrumbList to express your site structure.
What format should I use for structured data?
Use JSON-LD. It keeps all your structured data in a single script block rather than scattering attributes through your HTML, which makes it easier to maintain, audit, and generate programmatically. It is also the preferred format across the broader ecosystem.
What is the most common schema mistake to avoid?
Marking up content that is not actually visible on the page, or letting data like prices and ratings go stale so the markup contradicts your live site. Your structured data must match what a visitor sees, and you should validate it regularly since it can break silently when templates or content change.
If schema does not force a recommendation, is it still worth doing?
Yes. Schema is high-leverage because it is concrete and within your control. Clean, accurate markup lowers the odds that an assistant misreads your category, confuses you with another company, or pulls outdated facts, which makes the version of you the model reasons over more accurate and complete.
Written by
Daniel Arons
Co-founder & CEO, Aethon AI
Daniel co-founded Aethon AI in November 2025 to close the gap between how marketers measure AI visibility and what AI is actually doing with their brands. Before Aethon, he spent eight years building digital marketing programs in New York across SaaS, financial services, and consumer brands. He holds an MPA from Baruch College and a BA in Public Relations from SUNY Oswego.