The 46 Signals That Determine If AI Recommends Your Business

When ChatGPT decides whether to recommend your business, it isn't checking your Google ranking. It isn't counting your backlinks. It's pulling from a retrieval system that operates on a completely different set of signals — and most businesses have optimized for none of them.

The MAKIF Audit Framework is our structured approach to evaluating and improving those signals. Forty-six data points across five layers, each one representing a specific reason AI systems do — or don't — include your business in a generated response.

This post breaks down all five layers and what each one actually means for your business.

Why 46 Signals Across Five Layers

Generative AI systems don't make a single decision about whether to cite you. They make a sequence of decisions. Academic research on Retrieval-Augmented Generation (RAG) — the architecture underlying most AI search systems — identifies three distinct stages before your content ever appears in a response: retrieval (did the system find your content?), ranking (did it select your content over others?), and generation (did it actually cite you in the answer?).

Each stage has its own failure modes. A business can be technically crawlable but entity-ambiguous, so the system finds the content but doesn't know which business it belongs to. It can have strong entity signals but thin content structure, so it gets retrieved but not selected. It can get selected but have no citation triggers, so it gets referenced but not quoted or named.

The MAKIF Audit Framework addresses all five stages with 46 signals grouped into five layers. Miss any layer and you have a gap in your AI visibility — even if every other layer is strong.

Layer 1 — Trust: Is This Source Credible?

Before any AI system retrieves or cites your content, it evaluates whether your source is trustworthy. Google's own guidance on this is explicit: trust is the most important dimension of E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness). It's the baseline that all other signals build on.

The Trust layer evaluates signals like whether your content demonstrates first-hand experience, whether authorship is clear, whether your site's production process is transparent, and whether your content serves the user rather than the algorithm.

The KDD '24 GEO research found that for most local business categories, including Law & Government, the most powerful trust signals were statistics and cited sources. Data-driven claims outperformed tone changes every time.

Layer 2 — Entity: Does AI Know Who You Are?

An entity, in AI systems, is not a keyword. It's a node in a graph — a specific, disambiguated thing with defined properties and relationships. Google's Knowledge Graph contains millions of entities. Your business is either in it, with a clear identity, or it isn't.

Entity signals include your Google Business Profile status, your Organization and LocalBusiness schema markup, the consistency of your Name/Address/Phone across every web mention (NAP consistency), and whether your entity has a Knowledge Graph ID.

For local businesses, this layer is frequently where AI visibility collapses. A law firm may have a website and even rank for some keywords, but if its schema markup is missing, its Business Profile is unclaimed, and its address format varies across three different directory listings — AI systems don't know with confidence what it is or where it is.

Layer 3 — Retrieval: Will AI Systems Find Your Content?

Retrieval is the selection step. Before an AI system can cite you, it has to find you. Modern AI search systems use RAG architectures that retrieve content from external sources at inference time — meaning your content needs to be crawlable, indexable, and structured in a way that matches how these systems select documents.

Research from the 2023 RAG survey (Gao et al., arXiv 2312.10997) established that dense retrieval favors topically coherent, clearly structured content over keyword-heavy content. The Retrieval layer signals include technical factors (can AI crawlers access your site?) and structural factors (are your section headings meaningful?).

An llms.txt file — a direct signal to AI crawlers about how to navigate your site — is one of the fastest technical wins in this layer. Most businesses don't have one.

Layer 4 — Citation: Will You Be Named in the Answer?

Getting retrieved doesn't guarantee getting cited. The Citation layer is where content optimization has the most direct, research-validated impact.

The KDD '24 GEO paper tested nine different content optimization methods on 10,000 queries across multiple generative engines including Perplexity.ai. The top three performing methods — adding statistics, adding sourced citations, and adding quotations from credible sources — improved AI visibility by 30–40%.

For Law & Government queries specifically, Statistics Addition was the top-performing method. Embedding specific, sourced data points into your content is the single highest-impact GEO tactic for legal service businesses.

Layer 5 — Measurement: Are You Tracking Any of This?

Only 16% of businesses currently track their AI search performance in any systematic way, according to McKinsey's 2026 AI search report. This layer evaluates whether you have any monitoring in place — whether you're running regular AI query tests, whether you're using Search Console to understand how your content is being surfaced, and whether you have a baseline score to measure improvement against.

Your Baseline

The MAKIF Audit scores your business across all 46 signals and delivers a 0–100 score for each layer, a composite AI Visibility Score, a competitor comparison, and a prioritized action list ranked by impact.

The average score we've seen in our Towson pilot is 0. The gap between 0 and visible is specific, addressable, and measurable. That's what the framework is designed to close.

Sources: KDD '24 GEO Research (Aggarwal et al.) · RAG Survey, Gao et al. (arXiv 2312.10997) · Google E-E-A-T Quality Rater Guidelines · Google Knowledge Graph API Documentation · McKinsey AI Search Report 2026