Why don't AI models understand what my business does?

AI models analyze your site through entity extraction mechanisms. If your site lacks clear structured data (schema.org), consistent descriptions across all pages, and logical links between your services, the AI cannot build a reliable model of your business. It's like giving 50 different résumés to a recruiter.

What is schema.org markup and why does it matter for AI?

Schema.org is a standardized language that AI models understand to identify who you are (Organization, LocalBusiness) and what you do (Service, Product). Without this markup, LLMs have to interpret your site in raw HTML — they often miss the context. With it, you give them your business's exact identity card.

Does structured data actually help get cited by ChatGPT?

The studies are nuanced. An Ahrefs analysis of 1,885 pages showed that adding JSON-LD did not significantly increase citations in AI Overviews (-4.6%), AI Mode (+2.4%), or ChatGPT (+2.2%). However, sites with schema markup also tend to have better content and more backlinks. Schema alone isn't enough — it's one lever among others, but important for your entity's overall coherence.

How do AI models decide which business to recommend?

AI models use 'query fan-out': they break your question into several sub-queries and compare results. The Ahrefs study (1.4M prompts) shows that cited pages have a cosine similarity score of 0.656 with the AI's sub-questions, versus 0.484 for non-cited pages. To get recommended, your site must answer precisely the questions the AI asks itself — not just the main query.

Should I create an llms.txt file to help AI?

Useful but not magic. John Mueller (Google) confirmed that llms.txt helps discoverability but doesn't let LLMs differentiate site quality. An Ahrefs study of 137,000 domains shows 97% of llms.txt files received zero queries. Create it, but don't rely on it as your main solution.

How to Help AI Understand Exactly What Your Business Does?

The bottom line: AI models don’t read your site like a human does. They extract entities (who you are, what you sell) and build a model of your business. If that model is fuzzy, inconsistent, or incomplete, AI won’t cite you — even if your site looks great.

What you’ll learn:

How AI models (ChatGPT, Google AI Mode, Perplexity) decide to cite a business
What the 4 layers of information AI uses to understand you are
How to structure your site so an AI can answer “what’s this company about?”
The mistakes that make your business invisible to LLMs
A concrete checklist to audit your own Entity Readiness

Before you continue: This article is for TPE/SME owners who already have a website and want to appear in AI answers. If you don’t have a site yet — or if it isn’t indexed on Google — start with my guide on whether your site is well-built.

How does an AI “see” my business today?
What are the 4 layers of information AI uses to understand me?
What is the Entity Readiness Score and how do I calculate mine?
Why doesn’t AI understand my site even if it looks great?
Are structured data really essential for AI?
How should I structure my site so AI understands my business?
What is the Brand Control Quadrant and how does it help?
How to check if my site is entity-ready without paying for an audit
Key takeaways
Go further

I’ve spent hours analyzing TPE websites that deserve to be cited by AI — competent artisans, solid local businesses, skilled service providers. And every time, the same problem: the AI doesn’t understand what they do.

The problem: Your site might have a vague description on the homepage (“We accompany professionals in their digital transformation”), a menu listing services without explaining them, and pages that talk about everything without defining anything precisely. For a human, that works. For an AI, it’s noise.

The solution: There are well-documented mechanisms to help AI build a clear model of your business. It’s not magic — it’s a series of structural decisions you can make starting this week. 13 minutes of reading.

The proof: I stumbled onto this while digging into the Google LLM patent (WO2025063948A1, filed in 2023) which describes exactly how Google builds its understanding of businesses through AI. Since then, I’ve applied these principles to the sites I audit — and the results in terms of AI citations are clear.

How does an AI “see” my business today?

To understand what needs to change, you first need to understand how an AI analyzes your site.

Unlike a human who reads left to right and grasps the full context, an AI uses a process called entity extraction. It’s a 4-step mechanism described in the Google LLM patent:

Identify the entity — The AI spots information blocks that correspond to something concrete: a business name, a location, a service, a product.
Interpret the information — It connects these blocks to understand relationships: “this person works at this company,” “this service is offered in this city.”
Extract attributes — It collects details: hours, prices, specialties, certifications.
Enrich with third-party data — It cross-references with what it finds elsewhere: Google reviews, mentions on other sites, professional directories. What’s fascinating — and concerning — is that the AI does NOT do literal copying. As the patent states, it generates an interpretation, not a copy. If your site is ambiguous, poorly structured, or contradictory across its pages, the AI builds a flawed model. Or worse: it builds nothing at all.

“AI systems need to understand how information is related to each other to synthesize, recommend, and act. A perfect schema markup in isolation is useless if the relationships between entities are fragmented.” — Bill Hunt, CEO at Bisan Digital (Search Engine Journal, 2024)

What this means for you: Every page on your site must tell the SAME story about your business. Not a different version on the homepage, another on the contact page, and a fuzzy version on the services page.

What are the 4 layers of information AI uses to understand me?

Myriam Jessier, a technical SEO expert who worked on the Brand Control Quadrant at Semrush, explains that LLMs analyze your brand through 4 distinct layers:

Layer	What	Example
Known brand	Your owned assets: site, Google profiles, social media	Your website content, your LinkedIn page
Latent brand	Culture, client stories, shared content	Customer testimonials, articles mentioning your brand
Ghost brand	Internal documents, presentations, knowledge bases	PDF files, wikis, public technical documentation
Fractured brand	Scattered data on third-party sites	Reviews, directories, citations beyond your control

From my experience with the TPEs I work with, the most underestimated layer is the known brand — specifically your site’s consistency. 80% of AI comprehension problems come from a site that isn’t “entity-ready.”

Note: I audited a drywall installer who had 7 service pages, all with different descriptions of his business. On one page he said “drywall installer,” on another “partition specialist,” on a third “interior renovation.” For a human, it’s the same thing. For an AI, those might be 3 different entities. Result: no local AI cited him.

What is the Entity Readiness Score and how do I calculate mine?

A concept I created after analyzing about thirty TPE sites: the Entity Readiness Score measures how well your site allows an AI to build a reliable model of your business. It’s based on 5 criteria.

Criterion 1: Identity clarity (30 points)

Does your site say EXACTLY who you are from the very first page, without ambiguity?

✅ Your H1 contains your precise trade (“Plumber and heating engineer in Toulouse” — not “Innovative sanitary solutions”)
✅ Your meta description describes your activity in natural language
✅ Your service pages name the specific service provided, not a vague category

Criterion 2: Multi-page consistency (25 points)

Do all your pages talk about the same business?

✅ Same name, same address, same phone number on every page (consistent NAP)
✅ Your activity descriptions are consistent from one page to another
✅ Your “About” page provides a clear summary of your business

Criterion 3: Schema.org markup (20 points)

Do you have the structured data that AI models natively understand?

✅ Organization or LocalBusiness schema with the right attributes
✅ Service or Product schema for each offering
✅ sameAs links to your social profiles and directories

Tip: The Ahrefs study (1,885 pages) shows that JSON-LD alone doesn’t significantly increase AI citations. But it’s essential for entity coherence — it’s the foundation, not the strategy.

Criterion 4: Entity relationships (15 points)

Are the pieces of information on your site logically connected to each other?

✅ Your services are linked to your location (schema areaServed)
✅ Your blog articles are linked to your business (schema author on articles)
✅ Your products are linked to their categories (no disconnected silos)

As Bill Hunt of Bisan Digital explains, “before citing, AI must find — and discoverability depends on entity relationships.” That’s the Integrity Graph principle.

Criterion 5: Freshness and enrichment (10 points)

Does your site show signs of life and real activity?

✅ Regular content updates (blog, news, portfolio)
✅ Visible and varied customer reviews
✅ Project or achievement pages with concrete examples

Why doesn’t AI understand my site even if it looks great?

This is the question my clients ask me most often. The answer comes down to one word: expected format.

LLMs (large language models) have been trained on billions of documents, but they excel at recognizing structured patterns — not guessing. If your site looks like an ad brochure with vague marketing phrases, the AI won’t have enough confidence to cite you.

From my experience, the most common mistakes are:

Mistake #1: The Dictionary Gap — Your site uses words that nobody types into ChatGPT. “Thermal comfort solutions” instead of “plumber and heating engineer.” AI models look for direct semantic matches with user queries. If you don’t use the words your customers type, the AI won’t make the connection.

Mistake #2: Intentional vagueness — Many businesses deliberately write vaguely to “keep all doors open.” “We help businesses with their digital transition” — that could mean anything. An AI needs certainty to cite you, and vagueness is the enemy of certainty.

Mistake #3: Information silos — Your site has pages that don’t talk to each other. Your “Services” page doesn’t mention your city. Your “Contact” page doesn’t describe your services. Your blog has articles with no links to your offers. The AI can’t build a global model.

Note: A landscaper had 3 distinct services on 3 different pages — tree trimming, flowerbed creation, lawn seeding. No page explicitly said “landscaper.” His H1 said “Outdoor design expert.” Result: for AI models, he was an “outdoor design expert” — a job that doesn’t exist in their knowledge base. It took 3 title changes for him to start getting cited by Google AI Mode.

Are structured data really essential for AI?

This is the hottest debate in GEO in 2026.

Position A: Schema markup (JSON-LD) is the only way to speak the machines’ language. Google Knowledge Graph, which powers knowledge panels and AI search, is based on entities defined by schema.org. Without it, you’re invisible.

Position B: An Ahrefs study (1,885 pages tested) shows that adding JSON-LD did not significantly increase citations in AI Overviews (-4.6%), AI Mode (+2.4%), or ChatGPT (+2.2%). Sites with schema markup also invest in better content and more backlinks — these factors better explain the correlation.

The reality on the ground: Both are right about their point. Schema.org markup alone won’t get you 50% more citations. But without schema.org markup, you’re letting AI guess your identity — and it often gets it wrong.

“AI search is no longer based on keywords but on entities — structured representations that LLMs identify.” — Brian Dean, founder of Backlinko (Entity SEO Guide, 2024)

The technical minimum to set up

{
  "@context": "https://schema.org",
  "@type": "LocalBusiness",
  "name": "Your business name",
  "description": "Precise description of your main activity",
  "url": "https://yoursite.com",
  "telephone": "+33 1 23 45 67 89",
  "address": {
    "@type": "PostalAddress",
    "streetAddress": "123 Example Street",
    "addressLocality": "City",
    "postalCode": "75000",
    "addressCountry": "FR"
  },
  "sameAs": [
    "https://www.linkedin.com",
    "https://www.facebook.com/your-business"
  ]
}

This is the bare minimum. If you don’t have it — do it. If you do — verify it’s correct with the structured data testing tool.

How should I structure my site so AI understands my business?

Your site’s structure is the second most important lever after structured data. Here’s the structure I systematically apply to the sites I work with:

1. Homepage = full identity card

Your homepage must answer the 3 questions an AI asks when it arrives:

Who are you? (name + status)
What do you sell? (services/products)
Where do you operate? (geographic area)

Recommended format:

H1: [Precise trade] in [City] | [Business Name]
Introduction: 40-60 words that exactly describe your activity
Services section: each service named with its exact name
Location section: your service area

2. Service pages = entity description

Each service should be a distinct page with:

An H2 that is a natural question (e.g., “How does a drain unclogging service work?”)
A precise description of what’s included
The corresponding schema:Service link
A link to your service area (schema:areaServed)

3. About page = entity summary

This is the most underrated page for GEO. It should contain:

Your story in one sentence
Your specialties clearly listed
sameAs links to your profiles
Your precise offering

A documented AEO strategy on Reddit (0→15K users in 8 weeks) showed that adding an “entity anchor page” — an /about page with Organization+Person schema — was one of the 5 levers that took a site from anonymity to being recommended by Perplexity and Gemini.

Tip: Use the “Quick Answer block” format: 40 to 60 words at the top of the page that directly answer the question “What is [your service]?” in natural language. This is the block AI models preferentially cite.

What is the Brand Control Quadrant and how does it help?

Myriam Jessier presents the Brand Control Quadrant as a tool to take back control of your brand narrative in the AI era. Here’s how to apply it to your business:

Quadrant	Action	Concrete example
You control & you know	Update and optimize	Website, Google profile, LinkedIn profiles
You control & you ignore	Audit and reveal	Public PDF documents, accessible presentations
You don’t control & you know	Monitor and correct	Customer reviews, directories, third-party citations
You don’t control & you ignore	Discover and reduce	Scattered data, old mentions

Most TPEs actively manage only the first quadrant. The problem: LLMs train on ALL layers, including the most uncontrollable ones.

How to check if my site is entity-ready without paying for an audit

Here are the 3 tests I systematically run on the sites I audit. You can do them yourself in 15 minutes.

Test 1: The “Who are you” in 10 seconds

Open your site’s homepage, scroll until you see a trade or service name. Time it. If in 10 seconds you can’t tell exactly what the company does, neither can the AI.

Test 2: NAP consistency

Open 3 different pages on your site (homepage, contact, service). Write down the name, address, and phone number on each page. If even one character differs — even a period or an abbreviation — you have an entity consistency problem.

Test 3: The dictionary gap test

Type your trade into Google and look at the autocomplete suggestions. These are the terms internet users actually search for. Compare them with the words you use on your site. If you say “thermal solutions” when users search for “plumber,” you have a semantic gap.

What you need to do:

Check that your H1 contains your trade in natural language
Test NAP consistency across 3 different pages
Identify your semantic gap using Google suggestions
Add schema.org Organization or LocalBusiness
Link your services to your location (areaServed)
Create or optimize your /about page with entity summary
Add sameAs links to your social profiles

Summary — checklist:

#	Action	Done?
1	H1 in natural language with your trade	☐
2	NAP consistency verified across 3 pages	☐
3	Semantic gap (Dictionary Gap) identified	☐
4	Schema.org Organization/LocalBusiness added	☐
5	areaServed linked to your services	☐
6	/about page optimized with entity summary	☐
7	sameAs links added	☐

Interpret your score:

0-2: You’re potentially invisible to AI. Start with actions 1 and 4.
3-5: You have a decent foundation. The remaining actions will bring noticeable gains.
6-7: You’re entity-ready. Now monitor your citations in ChatGPT, Perplexity, and Google AI Mode.

Key takeaways

AI extracts entities, not text — your site must speak the language of models, not just look nice.
The Dictionary Gap is your #1 enemy — if you don’t use the words people type, the AI won’t make the connection.
Consistency beats perfection — a consistent site (same info everywhere) beats a beautiful site that contradicts itself.
Structured data is the foundation — not a complete strategy, but essential so you don’t leave AI guessing.
Query fan-out demands precision — the AI asks sub-questions on your behalf. Your pages must answer them one by one.

When your site helps an AI understand exactly what you do, you don’t just gain one more citation. You build an asset that works for you 24/7 — and speaks to every discovery channel: humans, search engines, AND AI.

Go further

What is GEO and how do AI models decide to recommend my business? — The basics of Generative Engine Optimization
How to become the #1 source cited by Perplexity and Google AI? — Advanced AI citation strategies
How to be visible on Google AND cited by AI at the same time? — SEO and GEO: two sides of the same coin

An AI that doesn’t cite you isn’t its fault. It’s your site that doesn’t speak its language.