How AI Platforms Choose Which Businesses to Recommend

When a prospective client in Toronto asks ChatGPT for the best immigration lawyer to handle their skilled worker application, or when a business owner in Calgary asks Perplexity which accounting firm specialises in cross-border tax for small businesses, the AI system that responds is not pulling results from a directory or running a live search. It is generating a response based on patterns in its training data and, in some systems, real-time retrieval. Understanding how those patterns work — and what signals shift them in your firm's favour — is the entire foundation of Generative Engine Optimization (GEO).

Citation North, a Vancouver-based GEO agency, has run hundreds of AI visibility audits for Canadian professional services firms across law, accounting, financial advisory, and healthcare. The patterns that determine which businesses appear in AI recommendations are consistent, learnable, and actionable. This article explains them.

How LLMs Process Queries About Local Businesses

A large language model like the one underlying ChatGPT does not "look up" your law firm when a user asks about it. Instead, it generates a response based on statistical patterns learned during training — patterns that reflect which businesses appeared frequently, authoritatively, and in contextually relevant ways across the billions of web documents, directories, review sites, media articles, and structured datasets that comprised the training corpus.

Think of it this way: an LLM has read most of the public internet. It has encountered your firm's website, your Google Business Profile, any directory listings you have, any media coverage you've received, and any third-party references to your firm across legal databases, professional associations, and review platforms. The model has built an implicit representation of your firm — how well-defined it is, how authoritative it appears, how often it's mentioned in relevant contexts — and that representation influences whether the model confidently names your firm in response to relevant queries or cautiously omits it.

For retrieval-augmented generation (RAG) systems like Perplexity, the process also involves real-time search: the system queries the web, retrieves current results, and synthesises them into a response. This makes Perplexity more responsive to recent changes — a newly published practitioner bio, a fresh directory listing, updated Schema.org markup — than a system relying purely on training data.

Both types of systems, however, share a common dependency: they recommend businesses that are unambiguous, well-corroborated, and contextually appropriate for the query. The signals that create that profile are what GEO addresses.

What Makes a Business Citable by AI

Across Citation North's audit work, six characteristics consistently differentiate businesses that appear in AI recommendations from those that don't:

Businesses that score well across all six dimensions have high AI citability. Businesses that score well on only some — a well-designed website but no structured markup, strong SEO content but no third-party citations — are systematically underrepresented in AI responses relative to their actual quality and expertise.

The Role of Structured Data in AI Recommendations

Schema.org structured data is the most direct, controllable signal a business can send to AI systems. Where a human reader might infer from context that your firm is an employment law practice based in Vancouver with three senior partners, an AI model needs to be told this explicitly — in a format it can parse without ambiguity.

The relevant Schema.org types for professional services firms include:

• LegalService / LawFirm: Identifies the business type as a legal services firm, with practice areas, service area, and jurisdictions served.
• Person (Attorney / Accountant): Names individual practitioners with their role, credentials, practice areas, and affiliated organisation.
• PostalAddress: Explicit, consistent address data that aligns with Google Business Profile and directory listings.
• FAQPage: Structured question-and-answer content that directly provides AI systems with extract-ready responses to common queries.
• BreadcrumbList: Site hierarchy signals that help AI understand the relationship between pages and content.
• LocalBusiness: General business entity definition with opening hours, contact information, and geographic service area.

When these schema types are correctly implemented and consistent across the website, the AI model's representation of your firm becomes sharper and more confident. The model is less likely to conflate your firm with another similarly named practice, less likely to misattribute expertise, and more likely to name your firm when a query falls within your practice area and geography.

The practical implementation of Schema.org for a law firm involves JSON-LD blocks embedded in each page's <head> — invisible to human visitors but machine-readable for AI crawlers. Citation North's AI Foundation Sprint includes comprehensive Schema.org implementation as its core deliverable, precisely because this is the highest-leverage structural change most firms can make.

Entity Authority Signals: How AI Decides Who to Trust

Beyond structured data on your own website, AI models evaluate "entity authority" — the degree to which your firm is recognised, mentioned, and corroborated by sources independent of your own website. This is analogous to the concept of domain authority in SEO, but operates on entity-level signals rather than link-level signals.

Key entity authority signals for professional services firms include:

Provincial law society and regulatory body listings. For Canadian law firms, the Law Society of British Columbia's online directory, the Law Society of Ontario's Lawyer Lookup, and equivalent provincial directories are authoritative external sources. When an AI model encounters your firm's name in training data or via real-time search, its presence in the official law society database significantly increases the model's confidence in recommending that firm. The same principle applies to CPA Canada listings for accountants and provincial college registries for healthcare professionals.

Legal directory presence. Avvo, Justia, FindLaw, the Canadian Legal Information Institute (CanLII), and specialised directories like the BC Legal Directory all contribute entity authority for law firms. These directories are heavily indexed, frequently cited in training data, and serve as corroboration signals that the entity is real, established, and operating in the claimed jurisdiction.

Media coverage and citations. Articles in Canadian Lawyer Magazine, Law Times, and regional news outlets that name your firm and its practitioners carry significant entity authority weight. When an AI model has encountered your firm's name across both your website and independent journalism, its confidence in the entity's accuracy increases substantially.

Review platform presence. Google reviews, Avvo ratings, and platform-specific ratings signals contribute to AI recommendations, particularly in systems with retrieval-augmented components like Perplexity. A firm with substantial, recent, relevant reviews is a more confident recommendation than one with sparse or dated reviews.

Common Reasons Firms Are Invisible to AI

Understanding why AI recommends certain businesses also illuminates why others are invisible. The five most common reasons Citation North identifies in AI visibility audits are:

1. No structured data. The firm's website contains no Schema.org JSON-LD. The AI model must infer entity information from unstructured text, which introduces ambiguity and reduces recommendation confidence. This is the single most common structural issue found in Citation North's audits of Canadian professional services firms.

2. Generic, non-attributed content. The website describes "our experienced team" and "comprehensive legal services" without naming practitioners, specifying credentials, or addressing the specific questions potential clients ask. AI models cannot extract practitioner-level entities or expertise claims from generic copy.

3. Weak third-party citation profile. The firm's external presence is minimal — a Google Business Profile with few reviews, perhaps one directory listing, no media coverage. Without multi-source corroboration, AI models treat the firm's self-description with uncertainty rather than confidence.

4. Name or entity ambiguity. The firm has a common name, has changed its name, has multiple locations with inconsistent addresses across platforms, or shares a name with a practitioner who has left the firm. Any ambiguity in entity definition reduces citation frequency.

5. Competitors have done the work. In most Canadian cities, one or two law firms in each practice area have either accidentally accumulated strong AI visibility signals (through years of active content, media coverage, and directory management) or have deliberately invested in GEO. Their strong entity signals crowd out competitors in AI responses, not because AI platforms prefer them but because they are simply the most clearly defined and corroborated entities in the model's knowledge.

The actionable insight from understanding AI recommendation patterns is straightforward: AI visibility is not random, and it is not out of your control. The signals that influence whether AI platforms recommend your firm are well-defined, measurable, and responsive to deliberate investment. An AI Visibility Snapshot from Citation North identifies exactly where your firm stands on each of these dimensions — and what to fix first for maximum impact on your AI citation rates.