How AI knowledge graphs decide whether your brand exists (And what to do if you are missing)

Search and AI products may use knowledge graphs, model knowledge, search indexes and retrieved web sources to resolve brands. There is no single universal AI knowledge graph. This guide focuses on practical entity clarity: consistent facts, accurate structured data and credible external corroboration.

SearchScore data: across the fixed panel of 102,873 domains in the Q3 2026 index, 67.6% declared no Organisation schema and 91.3% published no Person schema. Those are structured-data adoption gaps; they do not mean an engine has “nothing to attach a reputation to” or that the brands are invisible as entities. Source: SearchScore SAVI Report, Q3 2026.

What is an AI knowledge graph?

A knowledge graph is a structured representation of entities (people, organisations, products, concepts) and the relationships between them. When an assistant describes a brand, the answer may come from model knowledge, retrieved pages, an entity system or a combination. Do not infer a specific internal knowledge-graph record merely because the description is accurate.

There is no single universal knowledge graph. Multiple overlapping systems feed into AI models:

• Wikidata: an open structured knowledge base used widely across the web; its role varies by product and should not be described as the foundation for most AI systems.

• Google Knowledge Graph: Google’s proprietary entity database, built from web data, Wikipedia and user behaviour.

• Search and entity systems: providers including Google and Microsoft maintain proprietary search/entity infrastructure, but the precise relationship to each assistant’s answer selection is not fully public.

• Proprietary LLM training data: the information absorbed during model training from the public web, books, academic papers and licensed datasets.

An AI system may check several of these sources before deciding whether to mention your brand. If none of them have heard of you, the AI will not either.

How do brands enter knowledge graphs?

Brands do not submit applications. They earn their way in through consistent, verifiable signals across the web. The primary entry points are:

Wikipedia and Wikidata

Wikipedia and Wikidata can be useful entity references when a subject legitimately qualifies and the entries are well sourced. They are not guaranteed shortcuts into every AI system. Follow each platform’s notability, sourcing and conflict-of-interest policies rather than creating an entry solely for GEO.

Consistent entity signals across the web

Your business name, category and description need to appear consistently across dozens of sources. Google Business Profile, LinkedIn, Crunchbase, industry directories, review sites, social media profiles. When the same name and description appear in 20+ independent places, AI systems gain confidence that your brand is a real, established entity rather than a fleeting mention.

Structured data (organisation schema)

Organisation schema can expose business facts and relationships in a standard machine-readable form. Systems can also infer entity information from visible content and external sources, so use schema to reduce ambiguity rather than as the sole mechanism by which AI recognises a company. See our technical GEO guide for implementation details.

Press coverage and directory listings

Independent third-party mentions are validation signals. When a news outlet, industry publication or trusted directory lists your company, it creates an external entity reference that knowledge graphs can cross-reference. The more independent sources that mention your brand, the more confident AI systems become that you are a legitimate entity.

The 5 entity signals AI systems use to verify you exist

AI systems do not just check one source. They cross-reference multiple signals to decide whether a brand is a real, citable entity. These five signals carry the most weight:

1. Name consistency. Your company name must be identical across all sources. “Acme Corp” on your website, “Acme Corporation” on LinkedIn and “Acme Corp Ltd” on Companies House creates three different entities in a knowledge graph. Pick one canonical name and use it everywhere.

2. Described category. What type of entity are you? A software company? A marketing agency? A restaurant? AI systems need to categorise you to know when to mention you. State your category clearly and consistently.

3. Location. Physical location (or headquarters country) helps AI systems verify you exist. Include your address in Organisation schema and ensure it matches your Google Business Profile and directory listings.

4. Relationships to known entities. If your company is mentioned alongside known entities (industry associations, well-known clients, established events), AI systems can verify you through those connections. Partnerships, certifications and co-mentions all help.

5. External validation. Independent sources that confirm your existence and describe your business. Press articles, directory listings, review platforms, industry databases. Each one is a vote of confidence that strengthens your entity presence.

Why inconsistent branding kills entity recognition

This is the most common problem SearchScore audits uncover. A company trades as “Acme Corp” on its website, files as “Acme Corporation Limited” at Companies House, lists itself as “Acme Corp Ltd” on LinkedIn and appears as “Acme” in press coverage. To a human, these are obviously the same company. To a knowledge graph, these are four separate entities with no confirmed relationship.

The fix is simple but requires discipline: choose one canonical name and use it verbatim everywhere. On your website, in Organisation schema, on social profiles, in directory listings, in press releases. Every variation weakens your entity presence. Consistency strengthens it.

The same principle applies to your description. If your homepage says “AI-powered marketing platform” and your LinkedIn says “digital marketing solutions provider,” those are two different descriptions that create ambiguity. Pick one and stick with it.

The role of organisation schema: your machine-readable business card

Organisation schema (JSON-LD) is the most direct way to tell AI systems what your business is. Place it on your homepage. It should include:

• name: your canonical business name (exactly as used everywhere else)

• url: your primary website

• description: one clear sentence about what you do

• logo: a URL to your logo image

• foundingDate: when the company was established

• address: headquarters location

• sameAs: links to your social profiles and directory listings

Organisation markup is useful when it accurately describes the business, but it is not mandatory for AI visibility. Keep it consistent with visible content and credible external profiles. Our ChatGPT SEO guide covers implementation.

How llms.txt accelerates entity recognition

llms.txt is a plain-text file at your domain root that sets out what a site is and which pages matter, in your own words. Whether any AI engine reads it is unproven: no provider has confirmed that it does, and our own pre-registered study found the association with visibility is largely explained by the kind of site that publishes one. What did change in August 2026 is that Chrome’s Lighthouse now audits the file, so the format has a published pass condition for the first time.

A strong llms.txt includes: your company name, what you do, who you serve and markdown links to your most important pages. Write it so that a reader who sees only this file could describe your business correctly. That is a low-cost thing to get right, and it is worth doing on its own terms rather than on a promise about crawlers that nobody has substantiated.

The validation loop: how AI confirms you are real

A retrieval or entity system may compare information across multiple sources, but providers do not publish one universal “validation loop”. The defensible principle is simpler: contradictory business facts create ambiguity, while independent corroboration makes claims easier to verify.

This is why a single Wikipedia page is not enough on its own, and why having a website but no other web presence leaves you vulnerable. The validation loop rewards breadth and consistency. More independent sources confirming the same information equals stronger entity recognition.

6 things to do this week to strengthen your entity presence

1. Fix your canonical business name

Choose one name. Use it everywhere: website, schema, social profiles, directories, press releases. No variations. No abbreviations unless they are the official trading name.

2. Add organisation schema to your homepage

Include name, url, description, logo, foundingDate, address and sameAs (social profile links). Use JSON-LD format. Test with Google’s Rich Results Test tool to verify it parses correctly.

3. Review public entity sources where appropriate

Search Wikidata and other relevant public sources for your company. Correct factual errors through the platform’s rules. Create a new entry only when the subject and sources meet the platform’s requirements; do not treat Wikidata creation as a guaranteed high-impact GEO action.

4. Audit your directory listings for consistency

Check Google Business Profile, LinkedIn, Crunchbase, Yelp, Trustpilot and any industry-specific directories. Ensure your company name, description and category are identical across all of them. Inconsistencies here directly weaken entity recognition.

5. Optionally maintain llms.txt

If you choose to publish llms.txt, keep the summary factual and the links current. Treat it as low-cost agent-readiness housekeeping; no provider has established it as an immediate authoritative entity feed.

6. Run a SearchScore audit

Run a free audit to see your current AI visibility score and which entity signals are missing. The audit checks your presence across ChatGPT, Perplexity, Gemini and Google AI Overviews, and tells you exactly what to fix.