Does Schema Markup Increase AI Citations? What Ahrefs' 1,885-Page Test Means for SaaS Teams

As of 2026, the best controlled test finds that adding JSON-LD schema markup does not clearly increase AI citations. Ahrefs followed 1,885 pages that added it and compared them with matched control pages. Citations barely moved in Google AI Mode or ChatGPT, and slipped slightly in AI Overviews, so treat schema as housekeeping, not a growth lever.

This article is for B2B SaaS marketers and founders who run their own SEO and have heard that structured data is the key to AI visibility. It explains what the test did, what it can and cannot show, and how to run a fair test on your own pages. Cekpr covers search for B2B software, and measurement is where claims about AI visibility most often fall apart.

Why Did People Think Schema Drives AI Citations?

The belief started with a correlation. An earlier Ahrefs scan of 6 million URLs found that AI-cited pages were nearly three times as likely to have JSON-LD as pages that were not cited. About 53 percent of cited pages carried it.

That pattern looks like proof, but it has a simpler explanation. Sites with strong content, clean technical work and real authority tend to add schema as one of many practices. The markup travels with quality, so it shows up in the cited group without causing the citation.

This is a classic measurement trap. When two things occur together, the causal arrow can point either way, or both can trace back to a third factor. A controlled test is how you separate them.

How Did the Ahrefs Test Work?

Ahrefs published its schema and AI citations study on May 11, 2026. The team used its Brand Radar tool to find pages cited in Google AI Overviews, then used crawler history to spot pages that added JSON-LD between August 2025 and March 2026. That gave 1,885 treated pages.

For each treated page, the team picked about three control pages from other domains that never added JSON-LD and had similar citation levels beforehand. It counted citations for 30 days before and 30 days after the change. The primary method was difference-in-differences, which compares the change in treated pages against the change in controls.

That design matters. Raw numbers can mislead, because AI Mode citations rose about 43 percent for treated pages, but controls rose almost as much. A platform-wide trend lifted everything, and only the comparison shows it.

What Did Ahrefs Conclude, and What Are the Limits?

Ahrefs wrote that adding schema produced no major uplift in citations on any platform. It also said that if a page is already getting picked up, adding schema is unlikely to push it higher. The AI Overviews dip was real in the data, but both treated and controls were already declining, so the authors could not blame schema.

The limits are important, and Ahrefs lists them. Every page in the sample already had 100 or more AI Overview citations, so the findings apply to pages AI already sees. They say little about pages that AI ignores.

The window was only 30 days, so slow effects could be missed. Pages that added JSON-LD often changed other things at the same time. Different schema types, including Article, FAQ, Product, HowTo and Organization, were pooled, and only static JSON-LD was tested.

A fair summary: schema is not a proven lever for pages that are already visible. It is not shown to hurt, either.

What Does Google Say About Structured Data and AI Features?

Google's documentation on AI features and your website says there is no special schema.org markup to add for AI Overviews or AI Mode, and no new machine-readable files. A page needs to be indexed and eligible to appear with a snippet. The page says there are no additional technical requirements.

The same page says structured data should match the visible text. That is the rule that still applies. Markup that describes content users cannot see is a problem for rich results and for trust.

Search Console reporting follows the same logic. Google says traffic from AI Overviews and AI Mode is included in the overall Performance report under the Web search type. Our guide to measuring AI search visibility in Search Console shows how to track what that traffic does.

Should B2B SaaS Teams Stop Using Schema?

No. Markup still has uses beyond AI citations. It can support rich results, clarify what an organization or product is and help with consistent labeling. Google's structured data introduction explains that the markup helps Google understand page content.

The point is about priorities. A SaaS team with thin product pages and few third-party mentions should not expect JSON-LD to fix that. Spend time on the things AI systems actually draw on: clear answers, original data, accurate claims and credible sources.

Technical access still matters too. If crawlers cannot reach a page, nothing else counts. Our setup guide for robots.txt and AI crawlers on SaaS sites covers that layer, and our article on whether your SaaS site needs llms.txt reaches a similar conclusion about special files.

How Can You Test Schema on Your Own Site?

Run a small, honest experiment. Ahrefs suggests five to ten treated pages and five to ten matched control pages. The steps below turn that into a plan. Work through them in order:

Write down your hypothesis before you start. Decide what size of change would count as real, and what you will do either way. Without that, it is easy to read noise as a win.

Keep the sample in perspective. Ten pages will not show small effects, and AI citations fluctuate by day. If the treated group beats the controls by a wide margin, schema may be helping. If not, you have saved yourself a project.

What Else Is Worth Measuring?

Citations are one signal, and a noisy one. Track whether AI answers describe your product correctly, how often your brand appears for your main category queries and whether visits from AI sources turn into signups. Those tell you more than a raw citation count.

Accuracy matters too. If a model gets your pricing or features wrong, fix the pages it draws from. Our guide on fact-checking AI-assisted content offers a review process that helps keep those pages reliable.

What Should a Content Team Do Instead?

Put effort where the evidence points. AI systems need pages they can read, trust and quote. That favors direct answers near the top, original numbers with a named source, and claims that match the product.

Cekpr's editors read the Ahrefs authors' own caveats as the main takeaway. The test says nothing against structured data. It says that a single technical tweak rarely changes visibility on its own, and that quality signals work together.

Build a short checklist for each money page. Does the page answer the main question in the first paragraph? Does it cite a primary source for each statistic? Are the product details current? Is the page crawlable and indexed? Each yes is worth more than a markup block.

Also look outside your own site. Third-party mentions, reviews and community discussions shape what AI systems say about a brand. Several vendor studies suggest AI platforms lean on sources beyond brand-owned pages, though their methods differ, so treat the exact shares as indicative. Earn mentions where buyers actually look.

How Do You Avoid Misreading Your Own Results?

Three habits protect you. First, always use a control group, because platform-wide swings hide inside raw numbers. Second, change one thing at a time, since overlapping edits make cause impossible to find. Third, measure long enough to see past daily noise.

Be careful with vendor dashboards too. Citation counts differ by tool, by query set and by day. Pick one source, stick with it for the whole test and note its limits in your write-up.

Share results with the team in plain language, including the null ones. A test that finds nothing still saves money, because it stops a rollout that would not have paid off.

Conclusion

Schema markup is not the unlock for AI citations that some vendors claim. In a large controlled test, adding it changed little for pages that were already cited. Keep your structured data accurate, but put your effort into content, sources and access, and test any claim on your own pages before you rebuild a site around it.