Directory › Get Found in AI › Schema markup

Does schema markup help AI search? What the 2026 data actually shows

Somebody has probably told you that adding special code to your website will get you recommended by ChatGPT. A large study this year put that claim to the test. The answer is not what the industry has been selling.

Written by Michael Putnam, who runs local businesses in Southwest Florida.

First, what schema markup even is

If you have never heard the word, you are not behind. Schema markup is a block of hidden code that sits in the background of a web page. Visitors never see it. Its job is to label things for machines: this is the business name, this is the phone number, this is a question and here is its answer.

Think of it as the label on a filing cabinet drawer. The papers inside are the words your customers actually read. The label just tells a machine what is in there.

It has been around for years and it is genuinely useful. The argument that got out of hand is a newer one: that adding this code is how you get named when someone asks ChatGPT or Google for a local recommendation.

What everyone has been told

If you have read anything about AI search in the last couple of years, you have seen numbers like these. Businesses with schema see 45 percent higher AI citation rates. Pages with question-and-answer schema are 2.8 times more likely to be cited in AI answers.

Those figures got repeated everywhere, including in guidance I was working from myself. Plenty of agencies built a service around them.

The trouble is that they describe a pattern, not a cause.

What the study did

In May 2026, the team at Ahrefs published research that set out to test the claim properly.

They started where everyone else started. Looking across six million web pages, they found that pages cited by AI were about three times more likely to have schema markup on them. That is the number the industry has been quoting, and it is real.

Then they asked the harder question: does adding schema cause the citations, or do the two things just tend to turn up together?

To find out, they tracked 1,885 pages that added schema markup between August 2025 and March 2026. For each one, they lined up similar pages that were getting cited at about the same rate but never added schema, roughly 4,000 of them in total. Then they measured what changed on Google AI Overviews, Google AI Mode, and ChatGPT.

This is the difference between noticing that people who own running shoes tend to be fit, and actually checking whether buying running shoes makes anyone fitter.

What they found

Nothing. On ChatGPT and Google AI Mode, the pages that added schema moved by about two percent, which across thousands of pages is indistinguishable from random noise. On Google AI Overviews, citations actually fell slightly, by 4.6 percent, though the researchers were careful to say they could not confidently pin that on the schema itself.

Four separate statistical tests, all pointing the same way. Adding the code changed nothing.

A second experiment in the same report went further. Researchers checked whether five major AI systems actually read schema markup when fetching a page in real time. None of them did. They pulled the visible text off the page and ignored the hidden code entirely.

So why did the pattern look so convincing?

Because of who bothers with schema in the first place.

The kind of business that adds structured code to its website is also the kind that writes proper content, keeps its site maintained, earns links from other websites, and ranks well in ordinary search. AI systems were responding to all of that. The schema was just along for the ride.

It is the same reason expensive running shoes correlate with fitness. The shoes are not doing the work. The people who buy them were already running.

One important limit, in fairness

Every page in that study was already being cited by AI before the schema was added. Most had hundreds of citations already.

So what the research shows is that schema adds little for pages the AI systems already see. It does not prove schema is useless for a page nobody has found yet. That is a different question and nobody has answered it properly.

Worth knowing if you have no web presence at all. Not worth betting your marketing budget on.

The other thing that changed this year

In May 2026, Google switched off question-and-answer rich results. Those were the expandable questions that used to appear under some search listings.

For years, that feature was the main practical reason to add question-and-answer schema. It is gone now for almost everyone. Your code can be perfect and the feature simply will not appear.

Which leaves AI citations as the last remaining argument for piling on more schema, and that argument has now been tested and come up short.

Should you rip the code out of your website?

No. Leave it alone.

Schema still helps ordinary search engines understand your business details, and it costs nothing to keep. It is plumbing. Plumbing matters, it is just not the reason anyone visits the house.

What should change is where you spend your time and your money. If someone is charging you monthly to add more schema and promising AI visibility in return, ask them what evidence they have. Ask them to explain the mechanism. If they cannot, you have found something to stop paying for.

What actually works

Three things, and none of them involve code.

1. Recent Google reviews

Still the single biggest factor in whether Google and AI tools recommend you. What matters is not the total. It is the flow. A handful of fresh reviews every month beats two hundred from three years ago, because both Google and AI weigh recent opinion far more heavily.

The habit that works: ask every customer while the job is still fresh in their mind. Never offer anything in return, and never ask only the happy ones. Both will get you into trouble.

2. Being mentioned somewhere you do not own

This is the one most business owners have never considered. Analysis of over 23,000 AI citations found that roughly 77 percent of what AI says about a business comes from somewhere other than that business's own website. Directories, review sites, forums, local news.

Your own website vouching for you counts for very little. Of course it says you are great. Somewhere else saying it counts for a lot more.

That is why being listed in a real directory, mentioned in a local group, or written about by someone else does more for you than another page on your own site.

3. Plain sentences that answer the actual question

Since AI systems read the visible words and ignore the hidden code, the visible words are what you should work on.

Write the way a customer asks. If somebody would say "do you handle insurance claims," put that question on your page and answer it directly underneath. Lead with the answer, then explain. AI tools lift opening sentences.

And be specific. "We provide quality service" gives a machine nothing to quote. "We repair and install residential and commercial air conditioning in Venice, North Port, and Sarasota" gives it everything.

What to do this week

• Ask your last five customers for a Google review. Not a campaign, just five messages.
• Look at the first sentence on your homepage. Does it say plainly what you do and where? If it opens with something like "in today's competitive market," rewrite it.
• Write down the five questions customers actually ask you before booking, then put those questions and their answers on your website in plain text.
• Find one place online that is not your own website where your business could be mentioned by name.

None of that costs money. All of it is worth more than another line of hidden code.