1,885 pages added structured data. What moved wasn't the code.

Structured data, the descriptive code that sits behind a page and tells software what the page is about, is the most commonly sold technical deliverable in this market. One agency packages it at $1,997 if you install it yourself and $2,997 done for you, nine types of it, across a whole site.

In May 2026 Ahrefs ran the closest thing this field has to a controlled experiment on it. They took 1,885 pages that added structured data between August 2025 and March 2026, paired them with 4,000 similar pages that did not, and compared citations in the thirty days before and the thirty days after. That pairing is what makes the test worth anything: if the changed pages climb and the untouched ones don’t, the change is what did it.

Google AI Overviews      a small drop, big enough that it is probably not chance
Google AI Mode           no difference
ChatGPT                  no difference

Their verdict: adding it “produced no major uplift in citations on any platform.” The dip in AI Overviews, they say honestly, they cannot explain.

A second study arrives at the same place by a different road. Kurt Fischman analyzed 730 citations across 1,006 unique pages and 75 commercial queries. At first glance it looked as though the code was hurting the odds of being cited. But pages carrying it already rank higher on Google, and once that head start was set aside, the code itself accounted for nothing. What mattered was position: every step further down the page cost about a quarter of the odds of being cited.

Two independent methods, one conclusion. And Google’s own documentation says the same thing in a sentence: “There’s also no special schema.org structured data that you need to add.”

Now the exception, which is the actually useful part of this article.

Fischman found one kind of page that was cited far more often: the ones where the product and review fields had actually been filled in. Real prices. Real specifications. Actual values in the actual slots.

The code was not doing the work. The facts inside it were.

That lines up precisely with the strongest controlled study we have. Across 252,000 trials, stating a price landed in the top tier, alongside actually being about the subject, while tidy formatting landed at the bottom, where the effect cannot be told apart from doing nothing at all. And it lines up with a third finding, from reading 112 paragraphs by hand: among the paragraphs nobody cited, not one contained a new claim or a hard number. Not few. Zero.

Three methods. One direction. The thing being rewarded is not the container. It is whether there is anything specific inside it.

Ahrefs’ caveats, because they matter. They studied pages that were already heavily cited, a hundred citations or more, so on a page nobody visits the answer could be different. Sites that add this code usually change other things at the same time. All nine types were counted together. And they only watched for thirty days.

So should you add structured data? Yes, and we do, on every engagement.

But for the honest reason, which is not the one it is usually sold with. Structured data helps Google’s ordinary search results, it costs very little once, software can check it for you, and it gives you a disciplined place to record the facts about your business in a consistent form. It is good maintenance. It is not a lever for getting named by AI, and a four-figure price attached to it is being charged for the wrong thing.

The reframe that changes what you’d actually buy. If the effect lives in the facts rather than the format, then the work is not “add this code to 200 pages”. It is this: decide what your prices are, what your response times are, what areas you cover, and what is and isn’t included, and then write those down, on the page and in the code, in the same words.

That is a harder afternoon than running a generator. It is also the one with evidence behind it.