Does schema markup help with AI search? What it does and does not do
Structured data is worth implementing and it is not the lever people claim. Here is what it actually contributes and which types are worth the effort.
Every article about optimising for AI search recommends schema markup. Almost none of them say what it actually contributes, which leaves people implementing it, seeing no change, and concluding the whole area is nonsense.
Here is the honest version.
What schema does
Structured data is JSON-LD in your page source that states facts in a machine readable format. Instead of a machine inferring your price from text, you declare it.
Three real benefits.
Unambiguous facts. Your address, your opening hours, your price, your rating, stated in a format with no interpretation required. When a model is assembling an answer, an unambiguous fact is more usable than one it has to infer.
Search features. Rich results, review stars, FAQ expansions. This is classic SEO value and it is real.
Consistency signals. Schema stating the same organisation details as your Google listing and your directory profiles reinforces that they are the same entity.
What schema does not do
It does not get you read. If an assistant never fetches your page, your markup is irrelevant. Most buying answers are assembled from third party lists and directories, and no amount of markup on your own site changes which pages get read.
It does not override content. If your page text says one thing and your markup says another, the text wins and you have a credibility problem.
It does not substitute for plain text. A model reading your page reads the visible text. Facts that exist only in markup and not in the copy are a half measure. Write it in both.
That third point is the most common mistake. Teams put the price in Product schema and leave the visible page saying "contact us for pricing". A reader cannot see it, and so neither can most of what reads like a reader.
Which types are worth it
In order of return.
Organization. Name, URL, logo, social profiles, contact. One block, sitewide. Ties your identity together across the web.
LocalBusiness. If you have a physical location: address, hours, phone, area served, price range. High value for local answers, which lean on structured location data heavily.
Product or Service, with offers. Price, currency, availability. Genuinely useful, because pricing is one of the things buyers ask about most and one of the things most often buried.
FAQPage. Questions and answers on a page. Useful, and it maps neatly onto how assistants chunk content. Only mark up questions that visibly appear on the page.
Article. Headline, author, datePublished, dateModified. Freshness is weighted in AI answers, and dateModified states it unambiguously.
BreadcrumbList. Cheap, helps with structure.
That is the list. The long tail of schema types beyond this rarely repays the implementation time.
Rules that keep it honest
Only mark up what is visible. Schema describing content that is not on the page is a policy violation in search and a credibility problem everywhere.
Never fabricate reviews or ratings. This gets caught, it gets penalised, and it undermines exactly the trust you were trying to build.
Keep dateModified truthful. Updating the date without updating the content is a trick that stops working and starts hurting.
Validate it. Use a structured data testing tool. Broken JSON-LD is common and silent.
One Organization block, not five. Duplicate or conflicting entity markup across templates is a frequent and self inflicted problem.
Where to spend the effort instead
If schema is your main AI search initiative, the priority is wrong. In order of what actually moves answers:
1. Check you are not blocking AI crawlers. Three minutes, fixes more problems than anything else here.
2. Get into the third party sources that get cited. Ask your buyer questions, read the citations, go and get listed. This is the work that changes who gets named.
3. Put the facts in plain visible text. What you do, for whom, where, what it costs.
4. Be consistent across every profile you have.
5. Then schema, to state those same facts unambiguously.
Schema at step five is a good use of an afternoon. Schema as step one is why people conclude none of this works.
Checking whether any of it worked
Take a baseline before you change anything: twenty buyer questions across ChatGPT, Claude, Gemini and Perplexity, recording who gets named and what gets cited. Then make changes. Then re measure the same questions.
Without that, you are implementing markup and hoping. Our own Figo runs the measurement weekly from $49 a month, but a spreadsheet and two hours does the same job, and having any baseline at all puts you ahead of almost everyone.
Questions people ask
Does schema markup make AI assistants recommend me?
Not on its own. It helps machines parse your facts unambiguously, which is a genuine benefit, but it does not put you in an answer if you are absent from the sources the assistant reads.
Which schema types are worth doing?
Organization, LocalBusiness if you have a location, Product or Service with pricing, FAQPage, and Article on content pages. That list covers most of the value.
Can I fake review schema?
No. Marking up reviews you did not receive is a policy violation and it gets penalised. It also produces exactly the credibility problem you were trying to solve.
Is schema more or less important than it used to be?
About the same for search, and modestly useful for AI. It has never been the deciding factor and it still is not.
See it on your own competitors
Figo checks their ads, pages, rankings, reviews and AI answers every week, then tells you what to do in plain words. Set up in two minutes.
Keep reading