Schema as AI Context: Tell AI Models Which Company You Are
Schema markup does not earn AI citations. Its real job is identity: telling the models that every name you have traded under belongs to one company.
Key Takeaways
Schema markup will not get you cited by AI. It can stop the models from splitting your company into several names.
- Schema is not a citation lever. A causal study of 1,885 pages that added schema found no meaningful change in AI citations. AI systems read the visible text when they fetch a page.
- AI models remember names, not companies. Asked to compare Danish debt collectors, ChatGPT, Gemini and Claude named a brand retired in 2017 in 51.1% of answers. The company that now runs its Danish business appeared in 1.9%.
- Organization markup tells machines which names are yours. Declare your canonical name, list every variant as
alternateName, and tie your profiles and registry records together withsameAs. - Expect weeks, not days. A documented entity fix showed no movement for 29 days, with the effect projected around week eight.
Below: why schema does not move citations, the identity block for your homepage, and how to handle renames, mergers and brand families.
Why schema does not earn AI citations
Ahrefs tracked 1,885 pages that added schema against 4,000 matched pages that did not, and compared AI citations 30 days before and after. ChatGPT and Google AI Mode moved about 2%, which is statistically indistinguishable from zero. Google AI Overviews fell 4.6%. Every page in the study was already cited by AI, so it says nothing about pages the models have never seen.
The reason is how AI systems read a page. A test by searchVIU found that ChatGPT, Claude, Perplexity, Gemini and Google AI Mode all ignored JSON-LD when fetching a page and read only the visible text. Google states in its guide to AI features that no special structured data is needed to appear in AI Overviews or AI Mode.
So anything you want an AI model to read must be in the visible text. What schema can do is tell search indexes and knowledge graphs which organization a name belongs to, and the models learn from and ground their answers on those sources. How much weight that carries is not published by any AI provider. The problem it addresses, however, is easy to measure.
The models keep track of names, not companies. When one company appears under several names, its share splits between them.
The problem schema can solve: one company, several names
A rename, a merger, a misspelling or a sub-brand gives the models a separate name to recommend, and the names do not merge on their own.
In Danish debt collection, Lindorff merged into Intrum in 2017, and its Danish business was sold to Lowell in 2018. Eight years later, the models still named Lindorff in 51.1% of answers comparing providers and 40.4% of final recommendations. Lowell appeared in 1.9% of the comparisons. The models were recommending a name that no longer answers the phone.
In Danish home and car insurance, the Danske Forsikring name was retired on 1 July 2025, when its policies moved to the Topdanmark name. Fifteen months later it still held 33.3% of the models’ answers about which insurer to stay with.
Share earned by a name you no longer use is share you cannot collect. The fix has two layers: one name, spelled one way, everywhere text is written about you; and an Organization schema that declares which names and records belong to you. It is the recommendation design principle applied to identity: make the right answer the easy answer for the machine.
One block of markup on your homepage, referenced from every other page.
The identity block: Organization schema for AI
Google uses Organization markup to understand an organization’s details and tell it apart from others. Every property should match something a reader can verify on your site or in a public register.
| Property | What it tells the machine |
|---|---|
name | The one name you want recommended, spelled exactly as everywhere else. |
alternateName | Every other name the world uses for you: former names, abbreviations, common misspellings, your domain used as a name. |
legalName | The registered company name, when it differs from the trading name. |
sameAs | Profiles and records about you and only you: LinkedIn company page, business register entry, Wikidata item, industry register, review profiles. |
identifier, vatID | Your registration number. It cannot be misspelled, which makes it the strongest link between your site and the register. |
parentOrganization, subOrganization | Which company owns you, and which brands you own. |
@id | A stable identifier, such as your homepage URL plus #organization. Point every article’s publisher to it, so the whole site describes one entity. |
This is the block on this site. Seedli trades as Seedli, is registered as Seedli Labs ApS, and has a registered secondary name, Seedli ApS:
{
"@type": "Organization",
"@id": "https://www.seedli.ai/#organization",
"name": "Seedli",
"legalName": "Seedli Labs ApS",
"alternateName": ["Seedli Labs", "Seedli ApS", "seedli.ai"],
"url": "https://www.seedli.ai",
"identifier": {
"@type": "PropertyValue",
"propertyID": "CVR",
"value": "46386981"
},
"vatID": "DK46386981",
"sameAs": [
"https://www.linkedin.com/company/seedli-ai",
"https://datacvr.virk.dk/enhed/virksomhed/46386981"
]
}For Danish companies: secondary names registered in CVR (binavne) belong in alternateName, and your CVR page on Virk is the right sameAs link to the register. If your industry has an authorization register, make sure the name in it matches the name you trade under. Only add a Wikidata link if an item about you already exists.
Renames, mergers and brand families need a decision first, and markup second.
Renames, mergers and brand families
You retired a name, or bought one
Put the old name in alternateName on the company that carries the business today, and only if you legitimately inherited it. Say it in visible text too, on your About page and in your press boilerplate (“formerly Example Collect”), because the visible text is what an AI model reads when it fetches your page.
You are part of a group
Use parentOrganization on the brand and subOrganization on the parent, each pointing to the other’s @id or homepage. State the relationship in plain words on both sites as well.
You run several brands
Decide brand by brand: should it earn recommendations on its own, or feed the strongest name? A brand that competes on its own gets its own Organization block with its own sameAs profiles. A brand that feeds the parent goes in the parent’s alternateName. If you decide neither, the models decide for you.
Other schema worth keeping
Two more additions are cheap and consistent with the identity block. Neither will move AI answers on its own.
about and mentions, with links
about declares what a page is about; mentions lists entities referenced in passing. Give each entity a sameAs or url pointing to its official page, so “Gemini” resolves to the Google product and not the star sign.
DefinedTerm, for terms you coined
If you introduce a concept of your own, DefinedTerm records its canonical definition and who created it. The visible definition on the page is what gets quoted, so write that first. Do not mark up standard industry terms.
If a checklist tells you to add speakable markup for AI, skip it. It lives in JSON-LD, which AI systems do not read when they fetch a page.
How to check that it works
Run your homepage through the Schema Markup Validator and confirm the Organization block parses and every article’s publisher points to the same @id.
Then ask ChatGPT, Gemini and Claude about every name you have traded under, and read which company each answer describes. An old name described as a current company is a split still in place.
Give it time. In a study by Seer Interactive, an entity fix showed no movement for 29 days, with the effect projected around week eight. Track each name separately: the fix is working when the old name’s share shrinks while the canonical name’s grows.
What not to do
Do not claim names that are not yours
A competitor’s old brand or a generic category term in alternateName is a false statement, and easy to check against the business register.
Do not let markup contradict the page
Google requires structured data to match the visible text. A former name in your schema that your About page never mentions is a contradiction, and AI systems only see the page.
Do not point sameAs at pages only partly about you
A news article, a directory of fifty firms or a partner page is not your company. sameAs is for profiles and records that represent you and nothing else.
Do not describe your company differently on every page
Pages that each describe the publisher with a slightly different name and no shared @id recreate the split in markup. Define the organization once and reference it.
Do not let the identity block go stale
A rename, an acquisition or a new brand changes it. Review it whenever you review your internal links and update dates.
See which names the models use for your company
Seedli measures every name the models use for you, separately, so you can see where your share is splitting and whether the fix is taking hold.
Get startedThis is part of the Seedli technique series on structural optimisation for AI-mediated discovery. See all techniques and playbooks.