Structured data for AI: which markups really count

What structured data even is

Structured data is a fixed vocabulary with which you describe a page's content in machine-readable form. Instead of a system having to guess from the running text whether "19.90" is a price, a date or a room number, you say it explicitly. The common vocabulary is called Schema.org and is understood by Google, Bing and most AI systems. It is a shared vocabulary that the big providers have agreed on.

Technically you usually write these details as JSON-LD. That is a small block of data in the source code that doesn't change the visible page and yet delivers clearly structured facts. A trades business stores its opening hours this way, an online shop the availability of a product, a medical practice its address. The user sees none of it, the machine sees all of it.

The decisive point: markup doesn't invent content. It only makes unambiguous what is on the page anyway. Whoever marks up details that are missing from the visible text risks a penalty instead of an advantage. Structured data is a translation, not a claim.

Why AI systems react to it differently

Classic search engines have used markup for years for rich results: star ratings, price details, FAQ expanders directly in the result lists. This display is well documented and measurable. AI systems like ChatGPT, Perplexity or Google AI answers work differently. They pull facts together and formulate an answer instead of just pointing to ten blue links.

For these systems, clarity is worth its weight in gold. If your price, your company name and your address are cleanly marked up, the probability drops that the AI assembles them wrongly or confuses them with a competitor. Markup is not a magic switch that lifts you into every answer. But it is a safeguard against confusion and hallucination of your core data.

The honest classification is important: no major provider guarantees that structured data raises your ranking in AI answers. It helps the machine understand you correctly. That is the prerequisite, not the guarantee. Whoever promises more is selling you hope.

The markups that really count

Focus on a few types with a direct connection to your business. Organization or LocalBusiness maps your company: name, address, contact, opening hours. For an online shop, Product with price and availability is central. A publisher or blog benefits from Article with author and date. A service provider page with real user questions uses FAQPage sensibly.

These types pay off because they map facts that are actually asked about. A tax advisor, a restaurant and a SaaS provider have completely different core facts, but the principle stays: mark up what brings the user to a decision. Price, location, availability, responsibility, credibility. Everything else is secondary.

As a rough rule of thumb: if you cannot connect a markup type to your business model in one sentence, you probably don't need it. Completeness is not a value in itself.

• Organization / LocalBusiness – company identity, address, opening hours
• Product / Offer – price, availability, condition in commerce
• Article / NewsArticle – author, date, publisher for content
• FAQPage – only for real, visible questions and answers
• BreadcrumbList – page structure for navigation and context
• Review / AggregateRating – only with real, verifiable reviews

Markup that is often overrated

Many businesses stack dozens of schema types on top of each other, on the assumption that more is better. In practice hardly any system reads out the exotic types, and wrongly set markup harms more than it helps. FAQ markup for invented questions, review stars without real reviews or product prices that don't even appear on the page are classic mistakes.

Google has strongly restricted the display of FAQ and HowTo rich results in recent years. Whoever built their whole concept on it now stands without the hoped-for extra clicks. This shows the basic rule: never rely on a single display format that a provider can switch off at any time. Markup should secure your facts, not sneak in an ad format.

Technical gimmicks too, like deeply nested references or rarely used properties, seldom bring measurable benefit. The effort is out of all proportion. Keep it lean.

How to proceed in practice

Start with an inventory. Which facts are decisive for your customers, and do they appear visibly on the page? Only then do you mark up exactly these facts. Use JSON-LD and place it in the head or body of the page. Content management systems and shop platforms often come with plugins that generate basic markup automatically. For standard cases that is usually enough.

Test every markup before you roll it out. The Rich Results Test and the Schema Markup Validator show you whether your code is error-free and which elements a system recognizes. An engineering office that marks up its services sees immediately whether name, location and offering come through cleanly. Check again after every larger rebuild, because silent errors creep in quickly.

Keep markup and visible content in sync. If a price or an opening hour changes on the page, the markup must follow. Outdated structured data is worse than none, because it actively feeds false facts into machines.

• Collect facts that are purchase-decisive and visible
• Choose the fitting schema type, not as many as possible
• Create JSON-LD, via plugin or manually
• Check with Rich Results Test and Validator
• Test again after every relaunch and every price change

How to measure the benefit honestly

Don't expect a miracle curve upward from markup. The benefit shows indirectly: fewer confusions, more correct display, occasional rich results in classic result lists. In the Search Console you see which structured data was recognized and whether errors occur. That is your most important control instance for the technical side.

Whether an AI answer names you correctly cannot be captured in a single number. A pragmatic approach: regularly ask the AI systems typical questions about your offering and check whether price, name and facts are correct. A bicycle dealer, for example, asks about its brands and opening hours. If the answers deviate, you know where your data is unclear.

Treat markup as hygiene, not as a campaign. It is the basis on which other measures work, not a standalone growth lever. Whoever understands this invests the right amount of time in it and no more.

In summary: focus beats completeness

Structured data is not a trick but a clean translation of your facts for machines. The gain lies in clarity: systems confuse you less often, reproduce prices and contacts more correctly and categorize your content right. This applies to the online shop just as much as to the law firm or the trades business around the corner.

Focus on the few types that map your business, keep them error-free and current and test regularly. Do without markup stacks, invented reviews and formats that depend on a change in provider policy. Focus beats completeness. This way you build a foundation that holds up both in classic search and in AI answers – without making yourself empty promises.

Industry differences: where which markup pulls

Not every markup pays off equally for every industry. A local service provider like a hotel, a practice or a workshop benefits most from LocalBusiness, opening-hours and review data. Here it is about the questions AI systems ask on behalf of your clientele: when are you open, where are you, roughly what does it cost. These details are unambiguous, changeable and often queried. Exactly there clean markup pays directly into visibility.

An online shop, by contrast, lives on Product, Offer and AggregateRating markup. Price, availability and review count are the signals that appear in comparisons and purchase recommendations. Whoever is sloppy here or delivers outdated prices risks not only ranking disadvantages but also loss of trust when the AI names false figures. Therefore keep these fields technically in sync with your merchandise management system, not by hand.

Publishers and advisory sites in turn rely on Article, FAQ and HowTo markup. Here it counts that author, date and topic structure are machine-readable, so that your content is recognized as a robust source. Remember the rule of thumb: mark up first what is actually asked about in your industry, not what is technically easiest.

A worked example

Take a small business with 40 subpages. You decide to implement only three markup types cleanly: LocalBusiness on the homepage, FAQ on the five most important service pages and review data where real reviews sit. That is deliberately little. But it is fully maintained and validated error-free. The effort is about one working day of setup plus half an hour of maintenance a month.

After eight weeks you look at your server logs and the referrers. You see that requests from AI assistants land more often on exactly the five FAQ pages and that the dwell time there is higher. At the same time your opening hours appear correctly in assistant answers, because they are unambiguously marked up. That is not a magic trick, but simply the consequence of the machine no longer having to guess your details.

Weigh the benefit honestly: three clean markups beat twelve half-finished ones. Twelve half-finished ones produce validation errors, contradictory details and maintenance effort that eventually gets neglected. Exactly then your markup delivers false signals, and that is worse than none at all.

Frequent questions and misconceptions

Does more markup automatically bring more visibility? No. Structured data is not a ranking booster you can crank up at will. It is a translation aid. If the information is correct and relevant, the markup helps the machine use it correctly. If the information is thin or false, even perfect markup doesn't make it better.

Do I have to write everything myself by hand? No, and you shouldn't either. Most CMS and shop systems generate markup automatically from your content. Your job is to check this output and ensure that it matches the visible page content. Markup that claims something other than what the page shows is a clear warning sign and can harm you.

And what about the limits? Structured data doesn't replace a good page. It doesn't guarantee inclusion in AI answers, it doesn't accelerate poor content and it is no substitute for timeliness. See it as a foundation: invisible, but load-bearing. When the foundation is clean, everything else works better on top of it. When it crumbles, the rest slowly comes down with it.