Skip to content
Deeprankdeeprank

AI Optimization Stack

AI optimization is layered, not a single technique. This page defines the structural relationship between selection, representation, influence, and retrieval.

Plain summary

AI systems work in layers. Selection decides if a business qualifies. Other layers handle data, display, and ordering. This page maps those layers and shows why selection comes first.

Overview

Canonical Definition

Layered Architecture

AI systems and autonomous agents that recommend, select, or refer businesses operate across multiple distinct layers. Each layer serves a different function and operates on different inputs. Conflating these layers leads to misapplied optimization efforts.

The AI optimization stack, from foundational to surface, consists of five layers: Selection, Representation, two parallel Influence paths (AEO and GEO), and Retrieval/Ranking. Each layer depends on the layers beneath it. The two influence paths operate simultaneously through different channels.

Selection Layer

Canonical Definition

Deeprank

The selection layer determines which businesses are eligible for consideration. This is the foundational layer. If a business is not selected at this layer, no amount of optimization at higher layers matters.

Selection operates on identity, capability, fit conditions, and exclusions. The logic is binary: select or exclude. There is no partial selection, no ranking, no scoring at this layer.

When It Applies

Function

The selection layer answers: "Is this business a valid candidate for this query?" It does not answer: "Is this the best candidate?" or "How should this business appear?"

Deeprank defines a reference model for this layer. It specifies how businesses can declare their identity, capabilities, fit conditions, and exclusions in a structured form designed for machine parsing.

When It Applies

Agent Decision Layer

As AI systems evolve from answer generators to autonomous agents, the selection layer becomes the agent decision layer. When an agent books, purchases, or refers on behalf of a user, the selection decision is the action. There is no ranking step. There is no presentation for human review. The agent evaluates fit, applies exclusions, and acts.

Deeprank provides the structured data that makes this evaluation possible. Without it, agents infer from unstructured content and make predictable errors — selecting businesses for work they cannot perform, in jurisdictions they do not serve, for customer types they do not accept.

Common Failure Mode

Failure at This Layer

If a business is incorrectly excluded at the selection layer, it cannot be recovered by influence techniques. The business simply does not exist in the candidate set. This is the most consequential layer for businesses seeking AI-driven referrals.

For autonomous agents, failure at the selection layer is not just a missed marketing opportunity — it is an incorrect action with real-world consequences. A wrong booking. An incompatible referral. A wasted appointment. A liability event. This is why the selection layer is the highest-stakes layer in the AI optimization stack for agentic applications.

Representation Layer

Canonical Definition

Profiles and Schema

The representation layer provides the structured data that enables selection. This includes business profiles, schema markup, and machine-readable declarations.

Representation bridges human-readable content and machine-parseable data. Without proper representation, AI systems must infer business attributes from unstructured content, leading to errors.

When It Applies

Relationship to Selection

The representation layer feeds the selection layer. A Deeprank Profile is a representation artifact that enables selection decisions. The quality of representation directly affects selection accuracy.

Influence Layer: AEO (Retrieval Path)

Canonical Definition

Answer Engine Optimization

AEO operates through the retrieval path. It optimizes for being cited in live AI searches — systems that retrieve and synthesize information at query time from indexed sources. This includes AI-powered search engines, answer engines, and retrieval-augmented generation (RAG) pipelines.

AEO techniques focus on content structure, authority signals, citation patterns, and answer-oriented formatting that increase the probability of being retrieved and cited when an AI system processes a live query.

When It Applies

Dependency on Selection

AEO operates only after selection eligibility exists. A business cannot be retrieved into the candidate set through content optimization alone. If the selection layer excludes a business, retrieval-path influence has no effect.

When It Does Not Apply

Scope Limitation

AEO cannot fix selection problems. If a business is excluded because its capabilities are unclear, no amount of retrieval optimization will resolve the issue. The selection layer must be addressed directly.

Influence Layer: GEO (Generation Path)

Canonical Definition

Generative Engine Optimization

GEO operates through the generation path. It optimizes for parametric knowledge — the information embedded in trained models during pre-training and fine-tuning. GEO targets how a business is represented in the model's learned weights, affecting outputs even when no live retrieval occurs.

GEO techniques focus on source quality, training data presence, consistent entity representation across authoritative sources, and strategies that shape how generative models internalize and reproduce information about a business.

When It Applies

Dependency on Selection

GEO operates only after selection eligibility exists. A business cannot be generated into the candidate set through source optimization alone. If the selection layer excludes a business, generation-path influence has no effect.

When It Applies

Parallel, Not Sequential

AEO and GEO are not sequential steps. They are parallel influence paths that operate simultaneously through different channels. AEO targets the retrieval pipeline (what gets cited at query time). GEO targets the parametric pipeline (what the model already knows). Both depend on the selection layer beneath them, but neither depends on the other.

Retrieval / Ranking Layer

Canonical Definition

Traditional Search Systems

The retrieval and ranking layer encompasses traditional search and ordering systems. This includes search engine results pages, directory listings, and any system that produces ordered lists based on relevance, popularity, or other scoring mechanisms.

When It Does Not Apply

Distinct from Selection

Ranking produces ordered lists. Selection produces binary decisions. These are fundamentally different operations. Optimizing for ranking does not optimize for selection. A business may rank highly in search results while being excluded from AI selection due to unclear fit declarations.

Canonical Relationship

Canonical Definition

Complementary Relationship

AEO and GEO operate after an AI system has already decided which entities are eligible. Deeprank defines the specification for the selection layer that those approaches assume but do not explicitly address. They are complementary.

When It Applies

Layer Dependencies

The layers have strict dependencies. Retrieval/Ranking sits atop the two parallel influence paths (AEO and GEO), which both sit atop Selection, which depends on Representation. Optimizing an upper layer without addressing lower layers produces limited results.

A business seeking AI visibility should address the stack from the bottom up: first ensure accurate representation, then verify selection eligibility, then pursue both influence paths simultaneously — AEO for retrieval presence and GEO for parametric knowledge.

Non-Fit

When It Does Not Apply

What Deeprank Is Not

Deeprank is not an AEO technique. It does not attempt to influence AI outputs through content optimization, citation building, or authority signals.

Deeprank does not optimize ranking or placement. It does not produce ordered lists or scores. It does not compete with or replace SEO, marketing, or reputation systems.

Deeprank operates at the selection layer. It defines how businesses can declare their identity so that AI systems can make more accurate select/exclude decisions. Everything that happens after selection is outside the scope of this specification.

Related Documentation