Schema markup for AI search in 2026: what helps, what doesn't, and what to add first

Does schema markup help you get cited by AI search?

Not directly, according to the best evidence available. In May 2026 Ahrefs tracked 1,885 pages that added JSON-LD between August 2025 and March 2026 and compared them with about 4,000 similar pages that did not. Thirty days later, citations in Google AI Mode and ChatGPT moved by around 2 percent, which the researchers could not distinguish from zero, and AI Overview citations dipped slightly. Their conclusion: if the rest of your SEO is done well, JSON-LD is not the unlock.

Google says the same thing in its guide to AI features. There is no special schema.org markup needed to appear in AI Overviews or AI Mode, no AI text file to create, and no extra technical requirement beyond being indexed and eligible to show with a snippet. Pages get cited because they rank for the searches the engine runs and contain a passage that answers the question cleanly, which is the work we describe in our GEO guide.

So why do cited pages have more schema?

The same Ahrefs research found that pages cited by AI were almost three times more likely to carry JSON-LD than pages that were not. That is correlation, not cause. Sites that bother with clean structured data also tend to have fast templates, clear headings, solid internal linking and a team that keeps content up to date. The schema is a symptom of a well-run site rather than the reason it gets quoted.

There are also limits to what that study could show. Every page in it was already cited heavily, so it says little about pages AI systems have not discovered yet. It pooled all schema types together, measured only 30 days, and tested only markup present in the raw HTML. For a new or small site, the more useful question is not whether schema earns citations but what it does for the systems that process it.

What structured data still does well

Schema earns its place through three jobs that have nothing to do with tricking an AI into quoting you.

• It disambiguates your brand. Organization markup with your legal name, logo, addresses and sameAs links to your LinkedIn and other profiles tells every engine that these pages, people and profiles are one entity, which matters when your name is shared with other businesses
• It keeps you eligible for rich results. Google still shows enhanced results for types such as Product, Review snippet, Breadcrumb, Article, Event, Video and local business details, and those listings win clicks from both classic and AI-heavy results pages
• It feeds engines that say they use it. Microsoft's Fabrice Canel said at SMX Munich in 2025 that schema helps Bing's language models understand content, and ChatGPT search and Copilot lean heavily on Bing's index

The schema types worth adding for a service business

Most service businesses need a short list, done properly, rather than every type in the schema.org vocabulary. This is the set we put on agency and B2B service sites, including our own.

• Organization (or LocalBusiness for a single location): name, logo, URL, contact point, each office as a structured PostalAddress with an ISO country code, and sameAs links to official profiles
• WebSite: the site name and URL, linked to the Organization with a stable @id so both are recognised as one publisher
• Service: one per service page, with the provider, area served and service type, so each offer is described as a distinct thing
• Article or BlogPosting: headline, author, publisher, dates and image on every post, with dateModified updated when the content really changes
• BreadcrumbList: the path from home to the current page, which also helps Google show a clean URL trail
• FAQPage: only for questions visibly answered on the page, and with realistic expectations, as the next section explains

FAQ schema after Google ended FAQ rich results

Google stopped showing FAQ rich results on 7 May 2026, after limiting them to government and health sites since 2023, and it is removing the FAQ report from Search Console. HowTo rich results went in 2023. Neither change penalises pages that keep the markup, and Google says you can leave it in place.

We still mark up FAQs that appear on the page, for two reasons. Other engines and tools can read FAQPage markup even though Google no longer shows a rich result for it. More importantly, the visible question-and-answer section is what does the GEO work: a direct question followed by a short, self-contained answer is exactly the kind of passage AI engines lift. The markup is a small extra on top of content that is valuable anyway.

Mistakes that make structured data worthless or risky

Bad schema rarely helps and can cost you rich results through a manual action. Google's rule is simple: markup must describe content users can see on the page. These are the problems we find most often in audits.

• Markup that contradicts the page, such as a price, rating or opening hours that differ from the visible text
• Self-serving review stars on Organization or LocalBusiness markup, which Google has not shown as rich results since 2019
• Several plugins each outputting their own Organization node with different names or logos, so the entity looks like three businesses
• Schema injected only by JavaScript, which some crawlers and AI fetchers never execute
• Dates bumped without real changes to look fresh, which undermines trust in dateModified across the site
• Copying a competitor's markup with fields such as aggregateRating that your page cannot back up

How to add schema in the right order

Treat structured data as the last layer, not the first. If a page is not indexed, loads slowly or buries its answer, no markup will fix it. Once the basics hold, this is the order we follow.

• 1. Fix crawlability and indexing first: robots.txt, canonicals, sitemap and a clean server-rendered page
• 2. Add one Organization and one WebSite node sitewide, each with a stable @id, and reference them from every other node instead of repeating the details
• 3. Add Service markup to each service page and BlogPosting to each post, linked to the Organization as provider and publisher
• 4. Add BreadcrumbList and FAQPage where the page content supports them
• 5. Validate with Google's Rich Results Test and the Schema.org validator, then watch Search Console's enhancement reports for errors after each release

What we did on our own site

Digistartup runs on Next.js, so our schema is generated from the same data as the visible page and rendered in the HTML, never added later by a tag manager. We publish one Organization node listing our UK, Canada and Pakistan offices, one WebSite node, Service markup on each service page, a Blog node listing every post, and BlogPosting plus FAQPage markup on articles like this one. All of it points to the same @id, so engines see one consistent business.

None of that replaced the content work. The pages that get our clients mentioned in AI answers are the ones with specific, first-hand detail and clear answers near the top, supported by structured data that makes the business behind them unambiguous. If you are running GoHighLevel or another CRM alongside your site, the same principle applies to the data in it, which is why we usually start client projects with a CRM audit.

Frequently asked questions