Does schema markup help with AI search?

Yes. Structured data is associated with roughly 1.7 times more AI citations, and it costs one block of JSON in your head. It works because schema states plainly what a model would otherwise have to infer: who published this, what it is, and where else that publisher exists.

Which schema properties actually matter?

name, url, sameAs, description and @id. Name is the exact string you want repeated, sameAs resolves your identity across the internet, and @id joins your pages into one entity instead of many strangers. Skip aggregateRating unless the reviews are real and visible.

What schema should a small site use?

Start with one Person or Organization block with a permanent @id, then add BlogPosting to articles with that same id as author, and FAQPage wherever you answer a real question. Never mark up anything you cannot prove on the page.

Schema Is How You Introduce Yourself to a Machine

¶ CONTEXTTHE COLD ARRIVAL

01A machine arrives knowing nothing

When a crawler lands on your page it has no memory of you. It sees words arranged for a human, and it has to guess three things: who published this, what it is, and whether any of it can be trusted enough to repeat.

Schema is where you stop making it guess. A page without schema is a stranger. A page with schema is a source.

I learned this the embarrassing way. I once ran a tool whose homepage was titled AI Visibility Tool. Accurate, descriptive, and it never once stated the product’s own name in a place a machine could lift. Models had no name to attach to the thing they were reading, so they attached none. A tool that cannot state its own name will be described instead of recommended.

The fix was not more content. It was a small block of JSON in the head, saying plainly: this is the name, this is the maker, here is where else that maker exists.

¶ PLAYBOOKSIGNAL AND NOISE

02Five properties do almost all the work

Most schema guides hand you fifty fields. In practice a short list carries the weight, and two popular ones actively hurt you.

name

The exact string you want repeated. Not a tagline, not a slogan. If your schema name and your page title disagree, you have introduced two people.

url

The canonical home of this entity. One address, chosen once, used everywhere.

sameAs

Your identity anchors. My Person block lists four: GitHub, Telegram, LinkedIn and my agency profile. This is the property most sites skip and it is the one that resolves who you are.

description

One sentence a model can lift whole. Write it as the answer to who is this, not as marketing.

@id

The join key. Every article on this blog names the same author @id, so thirteen entries and one person are one entity instead of fourteen strangers.

aggregateRating

Noise, and dangerous noise, unless the reviews are real and visible on the page.

keyword stuffing in knowsAbout

Noise. A list of forty skills reads as a list of zero skills.

The one people underestimate is sameAs. sameAs is not a list of social links, it is you telling a machine which accounts across the internet are the same person as you. That is identity resolution, and models lean on it hard when they decide whether two mentions describe one entity or two.

The one nobody uses properly is @id. Give your Person or Organization a permanent id, then reference that same id as the author of every article you publish. An @id is what turns scattered snippets into one entity a model can hold.

EXTRACTED — THE SENTENCE THIS ENTRY EXISTS FOR

Schema does not make you interesting. It makes you identifiable.

¶ PLAYBOOKWHAT I ACTUALLY SHIP

03One graph across 21 products

Here is the real structure on my own portfolio, because a pattern you can see beats a pattern you are told about.

[1]

One Person block, one id. My homepage carries a single Person entity with a permanent id, four sameAs anchors and a short knowsAbout list. Everything else in the graph points back at it.

[2]

SoftwareApplication per product. Seventeen application blocks sit on that one page, each naming a real tool with a real description, wrapped in an ItemList of all 21 products so a model can read the portfolio as a set rather than a pile of links.

[3]

BlogPosting plus BreadcrumbList on every entry. Including this one. The author field is not a name string, it is the same id from step one.

[4]

FAQPage wherever a real question is answered. Sixteen questions on the blog index, three on every entry. Structured data lifts citations by roughly 1.7 times, and FAQ blocks are the cheapest place to earn it.

[5]

Nothing I cannot prove on the page. No invented ratings, no review counts I do not have. Markup you cannot verify on the page is not structured data, it is a lie with a syntax.

Start with the smallest useful version. This one block, adapted honestly, outperforms a hundred lines of invented markup.

// the introduction — paste in <head>, replace every value with something true

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Person",
  "@id": "https://yoursite.com/#you",
  "name": "Your Name",
  "url": "https://yoursite.com",
  "jobTitle": "What you actually do",
  "description": "One sentence a model can repeat about you.",
  "sameAs": [
    "https://github.com/you",
    "https://www.linkedin.com/in/you/",
    "https://x.com/you"
  ],
  "knowsAbout": ["Three", "To Six", "Real Topics"]
}
</script>

Paste it, replace every value with something true, and confirm it parses in a validator before you ship. Then say the same name, the same role and the same links everywhere else you exist.

Which is exactly tomorrow’s problem: what happens when your own story disagrees with itself across the web.