Will Adding Schema Markup Make AI Recommend Your Business?

Your structured data is implemented. The tests pass. The agency has closed the project, and you’ve been told your website is now better prepared for AI.

What problem did you actually solve?

That isn’t a criticism of the work. The markup may be excellent. It’s a question about what the work was for, and whether anyone established that before it began.

Why This Work Is Easy to Approve

Schema markup is an easy recommendation to accept. It’s widely adopted web technology, not a fad. Google documents legitimate uses for it. It produces a defined deliverable: something built, a test passed, a project closed.

Next to vaguer advice about reputation, evidence or authority, it feels concrete.

It often arrives as a finding in an audit: a line item marked missing, a score that rises once it’s fixed. Audit tools are very good at finding what can be counted. A missing property can be counted. A specialty that AI systems consistently misunderstand can’t, at least not by the same tool.

The people recommending it are often right, too. Plenty of websites have markup that is missing, incomplete or wrong, and fixing that is reasonable work.

The appeal is real. That’s exactly why it deserves a closer look.

What the Technology Does, and What the Evidence Shows

Structured data describes a page’s content in a standard, machine-readable format: this is an organization, this is its address, this person wrote this article. Google says it uses structured data to understand the content of a page and to gather information about the wider world, including people and companies. Its documentation for Organization markup says it can help Google tell a business apart from others with similar names.

Those are documented uses in Google Search. They are not promises about AI recommendations, and public guidance from other major AI assistants does not establish a comparable relationship between schema markup and their recommendation behavior.

Google’s own guidance on its AI features is direct on this point. Discussing what it takes to appear in AI Overviews or AI Mode, it says: “There’s also no special schema.org structured data that you need to add.”

A May 2026 study from Ahrefs, an SEO software company, provides a useful test of the assumption that adding markup improves AI visibility. Its researchers tracked 1,885 pages that added JSON-LD, the most common format for structured data, and compared them with about 4,000 similar pages that didn’t. Over the following 30 days, changes in citations from ChatGPT and Google’s AI Mode were statistically indistinguishable from zero. Citations in AI Overviews fell slightly against the comparison pages, a result the authors call real but unexplained. Their summary: “schema had no clear positive or negative effect.”

The study is useful partly because of how it began. Ahrefs first looked at about six million URLs and found that pages cited by AI were more likely to carry schema. A finding like that is easy to turn into a sales argument. It is also what you would expect if well-resourced websites are both more likely to implement markup and more likely to be cited for other reasons. So the researchers tested the change itself.

The study’s limits matter as much as its result. It measured citations, not whether a business was recommended to a buyer. The pages were already heavily cited before the markup was added. It was an observational comparison against matched pages, not a randomized experiment. It pooled every type of markup, and it looked at a short window. It doesn’t show that structured data is useless, and it can’t rule out benefits for businesses that AI systems don’t yet see.

In the pages Ahrefs studied, adding markup did not produce a reliable increase in AI citations. That finding is relevant to claims about AI visibility, but citations and recommendations are different outcomes. For what a technical reviewer can check in a structured data implementation, and what can be concluded from it, see Structured Data and AI Visibility: What Should You Actually Examine?.

The Question That Should Have Come First

Whether schema works in general is the smaller question. The larger one is whether it was the right response to your problem.

Consider three businesses with weak AI visibility.

The first is recognized by name, but AI systems describe it by the work it did a decade ago rather than the specialty it has built since. That is partly a question of entity clarity, and partly a question of what other sources still say.

The second has strong credentials that never appear in the sources AI systems draw on when buyers ask for help.

The third is accurately described, technically sound, and still rarely appears when a buyer asks an assistant to compare firms like it.

Those are three different problems, with different causes.

For the first business, markup on its own website can state the current specialty clearly. If the outdated description comes from directories and old press coverage, the markup doesn’t change what those sources say.

For the second, markup can declare a credential. If the evidence of that credential doesn’t exist in the places AI systems look, declaring it on the firm’s own site may add little.

For the third, the firm is already understood. Its absence from comparisons may have more to do with what evidence exists for the specific questions buyers ask than with how its pages are described.

Structured data could have a role in any of these situations. But the reason for implementing it, and the improvement it might reasonably produce, would be different in each case. That relevance has to be established, not assumed.

Whatever the problem, markup can express information about a business, but it cannot independently substantiate that information. It can state a credential or a specialty in machine-readable form. It can’t make the credential more substantial or the expertise deeper, which is the distinction behind AI can make your authority more visible, but it can’t give you authority.

A treatment being available doesn’t establish that you have the condition it treats.

If nobody defined the problem, the choice of solution wasn’t a diagnosis. It was a default.

Finishing a Task Isn’t Progress Against an Objective

An executive doesn’t need to know how JSON-LD is written. They do need to know what the spending is supposed to accomplish.

There’s a real difference between website maintenance and an AI visibility initiative. Correcting inaccurate markup, keeping business details current and maintaining the search features a site already relies on are reasonable technical upkeep. They don’t need an elaborate justification.

Work commissioned specifically to improve AI visibility makes an additional claim. It says something should change in how AI systems understand, include or recommend the business. That requires establishing what is happening now, what needs to change, why this intervention was chosen, and what evidence would demonstrate progress toward the objective.

The timing matters. Once work is underway, budgets are committed, schedules are established and other activities may depend on its completion. Discovering later that the wrong problem was addressed can affect far more than the original task. Defining the objective and how success will be evaluated before implementation begins makes it possible to judge the work against its intended purpose, rather than simply its completion.

A completion report can document the implementation: pages marked up, errors fixed, tests passed. Those are legitimate measures of the work performed.

The project may be completed exactly as specified. The markup validates. The deliverable is complete. The business may still be misunderstood, overlooked or incorrectly compared when a prospective customer asks AI for help.

The Decision Behind the Deliverable

If nobody established what needed to change about how AI understands and represents your business, how did you decide what to implement?