Skip to content
visible4ai
E-E-A-T CheckerLearnAI File GeneratorPricing
en|
Sign inAccount
visible4ai
E-E-A-T CheckerLearnBlogAI File GeneratorPricingContact
PrivacyTermsRefundsImprintCancel contracts hereWithdraw from contract

Featured on Product Hunt

© 2026 visible4ai. A product of Siempi AG.

We measure whether AI tools cite your website and show you how to improve.

Back to blog

Which schema markup still matters for AI search, and which you can stop adding

Structured data was built for rich results that mostly no longer exist. Four types do real work for answer engines today, two are decoration, and one is actively misleading if you fill it in wrong.

Saurav Chamoli·10 November 2026·3 min read

Most of the schema.org markup on the web was added between 2015 and 2022 to win a rich result in Google: star ratings under a recipe, an FAQ dropdown under a service page, a how-to carousel. Google removed or restricted most of those displays in 2023 and 2024. The markup stayed. Plugins keep generating it.

So the honest question for 2026 is not "does schema help SEO" but "which of these blocks does anything for the systems that now write the answer". My reading, type by type, from the markup we see when sites run an analysis here.

Still doing real work

Organization (and its children: LocalBusiness, ProfessionalService). This is the one that identifies you as an entity: legal name, address, founding date, the sameAs profiles that corroborate you. Answer engines are confused about who companies are far more often than about what pages say, and this block is the cheapest cure. Fill legalName, url, address, sameAs. Leave aggregateRating empty unless the reviews are on the page.

Person, for authors. A named human with a jobTitle, an affiliation and sameAs links to a LinkedIn or a personal site. This is the machine-readable half of "who wrote this". Pair it with a visible byline; the markup alone, with no author on the page, is a contradiction a system can detect.

Article / BlogPosting, with author pointing at that Person, datePublished, dateModified. Dates matter more than they used to: an assistant deciding between two sources on a topic that changes prefers the one it can date. Keep dateModified honest, it is read.

Product, if you sell things, with real offers, price, priceCurrency and availability. Shopping-style questions ("what does X cost in Switzerland") are among the ones assistants answer most confidently, and they take the number from markup when it is there.

Decoration

FAQPage. Google stopped showing FAQ rich results for almost all sites in August 2023. The markup does nothing for Google now, and answer engines read the questions and answers from the page text, where they are already in question-and-answer form. If the FAQ is good content, you lose nothing by keeping the markup; you gain nothing by adding it.

BreadcrumbList. Fine. Harmless. Helps a crawler understand site structure a little. Not worth an afternoon.

Written by

Saurav Chamoli

Builds visible4ai at Siempi AG and writes about the technical side of being read by AI systems: crawlers, files, markup.

More from the blog

  • 12 November 2026 · 7 min read

    On 10 November OpenAI opened Business Hub: claim your company in ChatGPT, keep its details current, add a product feed, and see how often ChatGPT cites your website. Open wherever ChatGPT is available and the verification is supported; we are testing it with our own site. What launched, the five steps, and what is US-only or missing.

    Mauro Miotti

  • 10 November 2026 · 3 min read

    Swiss SMEs keep their best content in PDFs: price lists, product sheets, the 40-page company brochure. AI answer engines mostly skip them. Why that is, and the two-hour job that fixes it.

    Saurav Chamoli

  • 10 November 2026 · 3 min read

    Swiss and German law makes you publish who you are. That obligation happens to produce exactly the page answer engines use to decide whether a company is real. Seven lines that turn a legal footer into a trust signal.

Actively misleading when wrong

aggregateRating anywhere the reviews are not visible on that page, or where the five stars are self-reported. Google treats this as spam. More to the point, an assistant that reads "4.9 from 212 reviews" in your markup and then cannot find the reviews has caught you in a claim it cannot verify, on the one page where it was checking whether to trust you. Either show the reviews or drop the number.

What all of this has in common

The types that still matter describe entities: an organisation, a person, an article, a product. The types that stopped mattering described presentation: how Google should draw a box. Answer engines do not draw boxes. They decide whom to believe, and the markup that helps is the markup that tells them, verifiably, who you are.

So the test for any block on your site: does it make a checkable claim about a real thing? legalName: Muster Treuhand AG, with a register link in sameAs, passes. @type: FAQPage around marketing copy does not.

A note on validators

A block can be syntactically perfect and useless. Google's Rich Results Test tells you whether it parses, not whether it says anything. Read your own JSON-LD once as if you were a stranger trying to learn who runs the site. If the answer is not in there, that is the fix.

The free analysis on this site reads the structured data on your pages and flags the gaps above. The patterns for answer-friendly markup are in Answer blocks and schema.

Mauro Miotti

ChatGPT now lets you claim your business. What to do this week, and what not to expect
Your PDF brochure is invisible to AI assistants. Here is what to do with it
Your Impressum is the most trusted page on your site. AI assistants read it too