Two B2B service firms in Berlin. Same kind of clients, comparable case studies, websites that could be siblings. Ask ChatGPT or Perplexity to name agencies in their niche and one of them shows up by name. The other doesn’t exist as far as the model is concerned. The only meaningful technical difference: one ships a machine-readable Organization schema, the other relies on prose. A company without machine-readable Organization schema isn’t a verified entity to an AI model—it’s just a name that happens to appear on the web.
This is the single most common gap I see in the audits we run. The copy is fine. The positioning is clear to any human who reads it. But the machine doing the recommending never gets to read it like a human—it reads it like a database, and a database needs fields, not vibes. Below is what the model actually does with your site, the exact eight fields that decide whether you register as an entity, and how to ship it on WordPress this week.
Why ChatGPT doesn’t recommend your company
To answer the question people actually type—“Warum empfiehlt ChatGPT mein Unternehmen nicht?”—you have to understand how a model builds knowledge about a company in the first place. It is not one step. It is a pipeline, and prose drops out at the second stage.
Roughly, it goes: crawl the page, parse the structured data, map the entity and its attributes, link those identifiers to known references, and embed the result into a knowledge graph the model can reuse when it writes an answer. Szymon Słowik describes the same five moves—extract, parse, map, link, embed—and then reuse for grounding. His point lands hard: “entity linking gives machines confidence. And confidence is what gets your data reused.”
Here’s the part that trips people up. A sentence like “we’re a Berlin agency founded in 2019” is unstructured text. The model can read it, but it can’t bank it as a fact with high confidence, because it has no anchor—no way to know this Berlin agency is the same one mentioned on LinkedIn, in a directory, in a press piece. Structured fields are different. They’re parsed flexibly, mapped to attributes, and linked to outside references. Prose gets interpreted; schema gets reused.
One honest caveat, because the data deserves it. A large Ahrefs study of 1,885 pages that added schema found that schema is far more common on AI-cited pages—but adding it alone produced no statistically significant citation uplift. Their own caveat: they pooled all schema types together, so Organization and entity schema weren’t isolated from Article or FAQ markup. Read that the right way. Schema is infrastructure, not a magic button. It doesn’t manufacture relevance you haven’t earned. It makes the relevance you do have legible to a machine.
The Organization schema: the 8 fields that decide AI visibility
This is the core of it. Drop this block into your homepage <head>, change the values, and you’ve given the model an entity to bank. Eight fields, each earning its place.
<script type="application/ld+json">
{
“@context”: “https://schema.org”,
“@type”: “Organization”,
“@id”: “https://example.com/#organization”,
“name”: “Acme Digital GmbH”,
“url”: “https://example.com”,
“logo”: “https://example.com/logo.png”,
“foundingDate”: “2019-03-01”,
“founder”: { “@id”: “https://example.com/#founder” },
“contactPoint”: {
“@type”: “ContactPoint”,
“contactType”: “sales”,
“email”: “hello@example.com”,
“areaServed”: “DE”
},
“sameAs”: [
“https://www.wikidata.org/wiki/Q000000”,
“https://www.linkedin.com/company/acme-digital”,
“https://www.crunchbase.com/organization/acme-digital”
]
}
</script>- @type—declares this is an Organization, not a product or article. Without it the model has no category to file you under.
- name—your exact legal or trading name. Pick one canonical spelling and use it everywhere; inconsistency reads as two different companies.
- url—your canonical homepage. This is the home address the rest of the graph points back to.
- logo—Google uses this to decide which image shows in Search and the knowledge panel; it’s also a visual identity anchor for the entity.
- foundingDate—a hard, checkable fact. Models love disambiguating attributes, and “founded 2019” separates you from a same-named firm founded last year.
- founder—linked to a separate Person entity via
@id, not typed inline. This is where expertise attaches (more on that next). - contactPoint—signals an operating business with a real point of contact, and
areaServedreinforces your geography. - sameAs—the array of external profiles that prove you’re the same entity across the web. This is the field that does the heavy lifting.
Google’s own documentation says Organization data on the homepage helps it “disambiguate your organization in search results,” with some properties working behind the scenes purely to tell you apart from other organizations. That word—disambiguate—is the whole game. The model isn’t asking “is this company good?” first. It’s asking “which company is this, exactly?”
Why your founder profile must be a Person entity, not a bio paragraph
AI models attribute expertise to people, not to brands. A logo can’t have experience; a person can. So a founder bio that lives only as an H2 and a paragraph—“Alexander Kirsch-Clayton is the founder of…”—carries no machine-readable identity. It reads beautifully and links to nothing.
The fix is a standalone Person entity with its own @id, referenced by the Organization’s founder field, and crucially carrying its own sameAs pointing at LinkedIn, Xing, and Wikidata where it exists. The guidance from jsonld.com is blunt: always include sameAs with at least one canonical profile, because that’s what’s used to confirm the person is real and to tell them apart from everyone else with the same name.
<script type="application/ld+json">
{
“@context”: “https://schema.org”,
“@type”: “Person”,
“@id”: “https://example.com/#founder”,
“name”: “Alexander Kirsch-Clayton”,
“jobTitle”: “Founder”,
“worksFor”: { “@id”: “https://example.com/#organization” },
“sameAs”: [
“https://www.linkedin.com/in/alexander-kirsch-clayton”,
“https://www.xing.com/profile/alexander_kirschclayton”
]
}
</script>When I first typed my own name into ChatGPT, it showed me exactly how little a bio page does on its own. The prose was there. The connection between “the person” and “the company” was not—because nothing in the markup said so. Two linked entities fix that in an afternoon.
sameAs: the six directories that verify your entity for AI models
In classic SEO, sameAs was nice to have. In LLM search it’s non-negotiable—it’s the property that tells AI systems “this entity is the same as the one described at these other URLs.” Here’s the priority order I use, and why each one earns a slot in the array.
- Wikidata—the strongest disambiguation anchor. It feeds knowledge graphs directly and gives you a Q-number as a stable entity identifier. If you do one thing off this list, get a Wikidata entry and link it first.
- Google Business Profile—local authority and a verified operating presence. The link you add is your profile’s public URL.
- LinkedIn Company Page—the dominant B2B trust signal. Use the canonical
/company/your-slugURL, not a personal profile. - Crunchbase—supplies funding, founding, and startup context that models lean on to describe what kind of company you are.
- Trustpilot or ProvenExpert—a social-proof signal that you’re a real business people transact with. Add the URL of your verified review profile.
- An industry directory—BVDW, Clutch DE, or a recognized vertical listing. Niche, but it reinforces category and geography.
You don’t need all six on day one. Wikidata, LinkedIn, and one review profile already give a model three independent confirmations that you’re a single, real entity. Each one you add tightens the link.
WordPress: Yoast/Rank Math vs. manual JSON-LD
For maybe 80% of sites, a plugin is enough. Yoast SEO outputs an Organization graph on every page and lets you set the logo, social profiles, and founder. Rank Math does the same. If you’re a single entity with no exotic needs, configure the plugin, add your sameAs URLs in the social settings, and you’re done.
You outgrow the plugin when you need more than it models. Yoast itself notes it “does not currently support all content types”—so multiple entities, a standalone Service schema alongside the Organization, or a linked Person with its own sameAs usually means hand-rolled JSON-LD. When you go custom, inject it yourself and make sure the theme isn’t already emitting a conflicting Organization block.
// functions.php — minimal manual fallback
add_action(‘wp_head’, function () {
if (!is_front_page()) return;
$schema = file_get_contents(get_template_directory() . ‘/schema/organization.json’);
echo ‘<script type="application/ld+json">‘ . $schema . ‘</script>‘;
});Verification: Rich Results Test plus a prompt-based cross-check
Two checks, run in order. First, paste your URL into Google’s Rich Results Test. “No items detected” means the markup isn’t being read—a syntax error, a plugin conflict, or it’s not on the page you think it is. “Organization detected” with your fields listed means the machine can parse you.
Second, the cross-check that actually mirrors the question you’re trying to answer. Ask ChatGPT and Perplexity directly about your company by name and watch whether “founded by,” “based in,” and “services” come back correct. This is where measuring matters—and it’s exactly why we built Cited, to track how these models describe a brand over time instead of you re-typing prompts by hand. One realistic note: Perplexity runs a fresh real-time crawl on every query with no static answer store, and schema-enabled pages have shown markedly higher top-three citation rates than pages without. But crawler latency is real—expect four to eight weeks before the first improvements surface.
From invisible entity to cited authority: what 60–90 days looks like
No overselling. Here’s the realistic arc, in three milestones.
- Week 1–2: Organization and Person schema live, validated in the Rich Results Test, no theme conflicts. This part is fast—it’s the foundation everything else stands on.
- Week 2–6: sameAs targets created and indexed—Wikidata entry submitted, LinkedIn and review profiles linked both ways. The external anchors take longest because they live on other people’s crawl schedules.
- Week 6–12: the first Perplexity or AI Overview mention that carries correct company context—founder, location, services—instead of a vague or wrong description, or none at all.
Schema won’t invent authority you haven’t built. What it does is make sure the authority you have gets attributed to you, by name, as a verified entity rather than a name that happens to appear on the web. If you’d rather have someone map the gaps and the fixes for your specific site, that’s exactly what our AI Visibility Audit is for—and you can always get in touch first if you want to talk through scope.
Either way, start with the eight fields. They’re a one-afternoon job, and they’re the difference between being read and being remembered.