The first essay on this blog laid out the macro argument: search is shifting from a list of ranked links to a single AI-generated answer, and local businesses are more exposed to that shift than anyone in the conversation is currently acknowledging. If you haven't read it, start there.
This essay goes one level deeper. It answers the question I hear most often after someone reads that first piece:
Fine — but how does the AI actually decide who to recommend?
The answer is more specific, more mechanical, and more actionable than most people expect. Understanding it changes how you think about almost every decision in your digital presence.
What an Entity Is — and Why It's the Right Unit of Analysis
The word "entity" shows up constantly in discussions about AI and search, usually without enough explanation to be useful. Let me be precise.
In the context of large language models and knowledge graphs, an entity is a distinct, identifiable thing in the world — a person, a business, a place, a product, a concept — that can be described by its attributes and its relationships to other entities.
Your business is an entity. It has attributes: a name, a service category, a geographic market, years in operation, certifications, personnel. And it has relationships: it serves homeowners in specific ZIP codes, it competes with other contractors in the same category, it is associated with certain municipalities and counties, it appears alongside certain other entities in content across the web.
The critical insight is this: AI systems don't see your business the way a human does. They don't visit your website, read your About page, and form an impression. They encounter signals about your entity across thousands of sources — your website, your directory listings, reviews, press mentions, forum discussions, social profiles, citation networks — and they aggregate those signals into a representation of what your entity is, what it does, and how confident they can be about those facts.
The strength and clarity of that representation is what determines whether the AI recommends you when someone asks a relevant question. I call this the entity graph — the web of associations the AI has built between your business and everything else it knows about.
How AI Systems Build Your Entity Graph
Large language models are trained on enormous text corpora. During training, the model encounters your business name in context, repeatedly, across many sources. Each encounter reinforces or modifies the associations it builds between your entity and everything else in those documents.
If the model sees "Bucks County Water Damage" mentioned alongside "emergency restoration," "flood damage," "Pennsylvania homeowners," "IICRC certified," and "24-hour response" consistently across many documents — your website, directory listings, review platforms, a local news mention, a few forum posts — it builds a confident, multi-dimensional representation of that entity.
If the model sees your business on a single website with thin content and nowhere else, it builds a weak, uncertain representation. When a user asks "who should I call for water damage in Bucks County," the model either doesn't surface your entity at all, or surfaces it with low confidence — which in practice means it recommends a competitor with a stronger entity graph instead.
Co-occurrence and Association Strength
The primary mechanism is co-occurrence — the repeated appearance of two concepts in proximity within training data. If your business name appears in the same documents as your service category, your geographic market, and specific relevant attributes across many sources, the model builds stronger associations between all of those elements.
This is why citation consistency matters at a level that goes beyond traditional SEO's concern with NAP accuracy. It's not just that your name, address, and phone need to be consistent for Google's local algorithm. It's that every appearance of your business name in a document about water damage in York County is another signal reinforcing the association the AI is building between those three concepts. Ten consistent citations create more association reinforcement than one well-optimized website. A hundred are building a moat.
The Confidence Threshold Problem
AI language models are probabilistic systems. They generate responses based on what they are most confident about — and they don't recommend entities they're uncertain about. This creates a practical threshold problem for local businesses.
Below the confidence threshold — when the AI has weak, thin, or inconsistent information about your entity — it routes around you entirely. The homeowner asking for a recommendation gets a competitor the AI knows more about, regardless of whether your actual work is superior. You don't get a second-place recommendation. You get no mention at all.
There is no partial credit. A business with a decent website and a few citations gets the same AI recommendation frequency as a business with no digital presence — zero — because both are below the confidence threshold.
Retrieval-Augmented Generation and Real-Time Signals
Modern AI search systems don't operate purely from static training data. Retrieval-augmented generation (RAG) systems augment responses with real-time retrieval from indexed web sources. When a user asks a question, the system retrieves relevant current documents and synthesizes a response from both trained knowledge and retrieved content.
This means your entity graph is built from two layers: static associations from training data, and dynamic signals from real-time retrieval. Your Google Business Profile, active review platforms, and recently updated web content are all potential retrieval targets. A business that has maintained active digital presence has more and more current material available than one that set up a website in 2019 and hasn't touched it since. Freshness matters more for GEO than traditional SEO practitioners often assume.
The Five Dimensions of a Strong Entity Graph
A strong local business entity graph has five distinct dimensions. Weakness in any one reduces overall entity confidence and lowers recommendation frequency.
1. Entity Clarity
The AI must be able to answer — with high confidence — the question: What exactly is this business? This requires your business name, service category, and geographic market to be stated explicitly, consistently, and repeatedly across all digital touchpoints. Not approximately. Not variably. Exactly.
"York County Water Damage Restoration — Licensed Pennsylvania Contractor" stated identically on your website, Google Business Profile, Yelp, BBB, Angi, and every other platform is building entity clarity. "York Water Damage" on one platform and "Water Damage York PA LLC" on another is building entity ambiguity — which the AI resolves by routing recommendations to your competitor who was consistent.
2. Attribute Density
The AI must be able to describe your entity with specificity. Businesses that exist in AI knowledge only as a name and a category are weaker recommendation candidates than those with rich attribute profiles: years of operation, certifications, specific services, response time, insurance affiliations, geographic coverage at the neighborhood level.
Every specific, verifiable fact about your business — whether on your own site or in third-party mentions — is attribute density that strengthens your entity graph. IICRC certification status. Named counties and municipalities served. Specific equipment used. How long you've operated in your market. Every factual claim is another node in the graph.
3. Geographic Association Depth
For local service businesses, geographic association most directly determines which queries you appear in. An AI system builds geographic associations between a business entity and specific locations through co-occurrence: how often does the business name appear in documents specifically about that location?
A single "We serve York County" statement on your website is weak geographic association. A content property with dozens of pages specifically about York County homeownership, water damage, flooding history, and contractor licensing — all of which include your business as a recommended resource — is building deep association between your entity and that market. Specificity creates association. Generality creates noise.
4. Third-Party Corroboration
AI systems are appropriately skeptical of what a business claims about itself. Content on your own website is weighted less heavily than third-party mentions in external sources, because self-description is inherently lower-confidence information than external corroboration.
Every time your business name appears in an external source in association with your service category — a review platform, a directory, a news mention, a forum post — is a unit of third-party corroboration that raises the AI's confidence in its representation of your entity. This is why review accumulation across multiple platforms is a GEO practice, not just a reputation practice. Each review is a piece of third-party content that co-locates your name with your service category and geographic market.
5. Temporal Consistency
AI systems weight the persistence of entity signals over time. A business that has maintained consistent digital presence across the same name, category, and geography for several years has built temporal consistency that a newer or recently restructured business lacks. For businesses building entity graphs from scratch, the practical implication is that consistency from day one matters more than you might expect. Every early inconsistency is a signal the AI will have to resolve later — usually in a competitor's favor.
How This Maps to a Practical Content Strategy
Own Your Service + Geography Intersections Explicitly
For every combination of service and geography you serve, there should be a page, listing, or documented presence that explicitly connects those two things. Not one general "we serve the region" page. Specific, detailed content about water damage in Bucks County. About restoration in York City. About pipe bursts in Allentown. About basement flooding in Lancaster County. Each of those intersections is a specific query a homeowner might ask an AI. The businesses that have content specifically answering each are more likely to appear in responses to them.
Structure Content as Questions and Answers
AI systems are question-answering engines. Content organized around specific questions homeowners ask — with direct, factual answers — is more AI-retrievable than content organized around what a business wants to say about itself. "What should a homeowner do in the first hour after a burst pipe?" followed by a numbered, specific answer is more useful to an AI synthesizing a recommendation than a general page about water damage services.
Build Breadth Before Depth (Initially)
The first priority for a business with a thin entity graph is crossing the confidence threshold as quickly as possible. That means broad presence across many platforms — Google Business Profile, Yelp, BBB, Angi, Houzz, and relevant industry directories — before investing heavily in content depth on any single property. Once you're above threshold and appearing in AI recommendations, deepening the graph with more specific content and third-party presence compounds the advantage.
Respond to Reviews — on Every Platform
A review response is entity-specific content published on a third-party platform, indexed by search engines, and available for retrieval by AI systems. It includes your business name, often your service category, and sometimes your geographic market. Every review response you write is a small but real unit of entity graph construction. Consistent response practice compounds over months and years in ways that most businesses don't track but AI systems do weight.
The Competitive Dynamics: Why Local Markets Tip
One of the most important and underappreciated aspects of AI recommendation economics is that local markets don't just favor the best entity graph — they tip toward it.
In traditional search, the top ten results all get some clicks. Traffic distributes across the results page based on position. In AI search, recommendations collapse. The homeowner gets one or two business names. Sometimes three. The difference in exposure between the first recommended business and the fourth is not gradient — it's categorical. One gets recommended. The others don't exist in the interaction.
This means local markets, as they transition to AI-influenced search, will increasingly produce winner-take-most dynamics within specific service categories. The business with the strongest entity graph in a given market will absorb the AI recommendation share that previously distributed across competitors. The others don't get smaller pieces — they get displaced from the recommendation set entirely.
In a market that hasn't yet tipped, the cost of building entity strength is modest and the return on being the first to cross threshold is enormous. In a market that has already tipped, displacing the leader requires overcoming a compounding advantage that grows more expensive to challenge over time.
Most Pennsylvania local home service markets have not yet tipped. The entities being built now are establishing those dominant positions. This is not indefinitely true.
What Nexum Network Is Building and Why It Works
The network of lead generation properties we're building in Pennsylvania is explicitly designed around entity graph construction for the markets each property serves.
Each property is a distinct entity with clear name-category-geography associations, maintained consistently across every platform it touches. The content on each property is structured around the questions homeowners in that market actually ask, with direct factual answers that are retrievable by both Google and AI systems. Each property builds geographic association depth through content specifically about the communities, flooding patterns, housing stock, and contractor landscape in its market.
The citations, review presence, and third-party mentions we build for each property are not just traditional local SEO work — they're entity graph construction. Each citation is another co-occurrence signal. Each review is another piece of third-party corroboration. Each external mention of the property in association with its service category and geography is another reinforcing node in the graph.
The contractors who work with us benefit from entity graphs built with these mechanics explicitly in mind — not just optimized for Google as it existed in 2015. The calls they receive come from both traditional search and AI-assisted discovery, and the balance between those channels will increasingly favor the latter over the next several years.
The Practical Takeaway
Stop thinking about your digital presence as a website. Start thinking about it as an entity graph under construction.
Your website is one node. Your Google Business Profile is another. Each directory listing, each review, each external mention is another. The connections between those nodes — the consistency of naming, the repetition of category and geography associations, the accumulation of third-party corroboration — are the edges of the graph.
The strength of that graph is what determines whether the AI knows who you are when someone asks for help in your market. Building it is methodical, not magical. It requires consistency more than creativity, depth more than cleverness, and time more than budget. The businesses that treat this as an operational discipline — not a one-time project — are the ones that will occupy the high ground in AI search when local services fully tips.
The ones that wait for it to become obviously urgent will find the high ground already occupied.