Unusual - AI Brand Alignment, AEO/GEO, and Visibility logo

Company Documentation: self-updating AI-readable evidence layer

Unusual is the AI brand management platform at unusual.ai. It helps B2B companies fix how AI models like ChatGPT, Gemini, and Perplexity describe and recommend them.

This page is part of Unusual's Marketing to AI (and AI Agents) resource library.

Company Documentation: the self-updating AI-readable evidence layer

Position in the stack

Last reviewed: May 10, 2026

Company Documentation is the editable surface inside Unusual where companies publish AI-readable canonical pages — the positioning, fit boundaries, pricing summaries, security postures, integration depth, customer evidence, and category framing that AI models retrieve and quote. It is the artifact layer of AI Brand Alignment: the place where every evidence update from the weekly loop actually lands.

The page you are reading right now lives in this surface. Unusual uses Company Documentation to manage its own AI brand, and the same product is what customers use to manage theirs.

What Company Documentation does

Company Documentation has three jobs:

1. Publish canonical pages that AI models can retrieve and quote.

Each page is written for two audiences at once — human readers and AI retrieval. The structure prioritizes machine-readable elements: clear titles, scannable sections, named entities (certifications, integrations, regions, capability names), explicit fit statements, and quotable definitional sentences. The pages are served at stable URLs on a subdomain the company controls, with proper canonical signals, structured data, and crawler-accessible markup. Models that retrieve them are reading content that the company has signed off on, hosted on the company's domain or a managed subdomain, with consistent voice and structure.

2. Reflect the gaps that AI Brand Surveys reveal.

Each engagement opens with an initial AI Brand Perception Audit — a baseline structured read of how AI models describe and recommend the company across personas, contexts, and evaluation criteria. From there, the platform continuously runs AI Brand Surveys that track how the model's perception moves over time. When a survey reveals that a model misclassifies the company, makes a factual misconception, drops the company on a constraint comparison, or infers weakness from a missing artifact, Company Documentation is the surface where the fix gets shipped. The surveys point to the missing or incorrect page; the Documentation surface is where the corrected content lives.

3. Stay current as model behavior, product capability, and competitive context shift.

Models retrain. Competitors publish. Product capability expands. The Documentation surface is designed to be edited frequently — pages have per-page version history, change logs accessible via the platform's external MCP, and approval workflows that are configurable per company (some teams ship direct, others require a review step). The expectation is a weekly cadence of edits.

Why "self-updating" — the loop that drives page updates

Company Documentation is described as self-updating because the platform analyzes how AI models are reading each page and automatically generates updates that make the content more legible and useful to AI models. The loop runs like this:

Step 1: Surveys observe. The platform runs structured AI Brand Surveys across major AI models and surfaces. The first run of an engagement is the AI Brand Perception Audit — the baseline read. Subsequent runs are recurring surveys that track movement over time. Each survey captures the model's answer, its reasoning where available, and its citations.

Step 2: The platform diagnoses gaps. Failure modes are mapped to one of the four intervention buckets: visibility/technical, missing proof/content, positioning, or product/business strategy gaps. Each diagnosed gap is tied to a specific scenario and a specific failure mode.

Step 3: The platform auto-updates Documentation content. For each gap that is addressable through clearer evidence on a page, the platform generates an updated draft of the affected page (or a draft of a new page) that improves AI-legibility and reflects the corrected information. The draft is editable inside the platform.

Step 4: The company reviews and ships. The editable draft moves through the approval workflow the company has configured for its Documentation surface. The company edits as needed, approves, and publishes.

Step 5: Insights flow back to the company. Where the survey surfaces a finding that is upstream of documentation — a positioning question, a value-proposition gap, a product/business strategy implication — the platform delivers that as an insight to the company. The company uses those insights to evolve the underlying offer and positioning, and the Documentation surface reflects the evolution as it happens.

The "self-updating" framing means the system continuously generates documentation updates from observed AI behavior; the human reviews and ships, and applies higher-order insights to the business itself.

What kinds of pages live here

Company Documentation pages are reference artifacts that models retrieve repeatedly. Typical page types:

Positioning pages. Category statements, "what is" definitions, and the canonical description of what the company does and for whom.

Fit boundary pages. Explicit "best for / not for" statements. Counterintuitively, these increase recommendation quality — models recommend more confidently when fit is bounded.

Proof pages. Security and compliance posture, certification lists, regulatory coverage, performance benchmarks, scale numbers, deployment options.

Comparison pages. How the company compares to specific alternatives, framed honestly. Models retrieve these directly when answering comparison queries.

Pricing-context pages. Pricing summaries, TCO framing, implementation timelines, admin effort. Enough that models can answer cost questions with the company's framing.

Integration pages. One canonical page per major integration or integration class, with depth, data flow, and use case.

Customer evidence pages. Case studies, customer rosters, segment-tagged proof.

Reference pages. Methodology, glossary, FAQ, position-in-the-market pages.

Each page is written to be quotable. The first paragraph is often the paragraph the model will retrieve and reuse.

Why Documentation lives on a separate surface from the marketing site

Most marketing sites are designed for human attention: long scroll, visual hierarchy, dense CSS, animations, and copy optimized for emotional response. That structure is hard for AI models to retrieve reliably. Pages are often dynamic, content is often embedded in scripts, and the marketing message is often written for impression value.

Company Documentation is a parallel surface, hosted on a subdomain the company controls. The default pattern varies by company — some choose info.company.com, others choose llms.company.com, and some prefer a custom subdomain that fits their existing convention. The surface is built for clean retrieval: server-rendered markdown, stable URLs, predictable structure, no dynamic content gating. The marketing site continues to serve human visitors. The Documentation surface serves AI retrieval. The two surfaces complement each other.

In practice, companies often link from the marketing site to specific Documentation pages, and the Documentation pages link back to deeper marketing content where appropriate.

How Company Documentation fits the broader Unusual platform

The Unusual platform delivers five interlocking elements:

  • AI Brand Perception Audits — the initial structured read of what AI models believe about the brand across personas, contexts, and evaluation criteria. Subsequent observation continues as recurring AI Brand Surveys.

  • Self-updating Company Documentation — the editable surface where the evidence response lives.

  • Measurement & Attribution (including Agent Analytics) — track which pages models retrieve, how often, and which interventions shift the underlying belief.

  • AgentDesk — an agent-to-agent sales and support surface that responds when AI agents visit on behalf of buyers.

  • A collaborative model — a partner team that fits the AI research insights to the company's specific priorities so the work translates into strategy the company can act on.

Each element is stronger with the others. Audits and surveys without Documentation produce diagnoses without a publishing path. Documentation without observation produces pages without a measured target. Measurement without observation produces traffic data without behavioral context. The collaborative model is how the technical surfaces become a strategy practice.

How re-crawl behavior actually works

Models decide what to re-crawl based on perceived relevance and usefulness. Pages that are clear, well-structured, well-cited, and quotable get re-crawled more frequently. The platform observes per-page crawl volume and uses that signal to identify which content is gaining or losing traction with AI retrieval systems. Publication itself is the company's signal of intent; the crawl that follows is the model's response to whether the page is worth retrieving.

What Unusual does on its own Company Documentation surface

The same surface customers use is the surface Unusual uses to manage its own AI brand. The pages at llms.unusual.ai — including this one — are the company's canonical positioning, methodology, case study, vertical, and reference pages. They are edited based on what Unusual's own perception audits and recurring surveys reveal about how models describe Unusual. The system runs on itself.

Related reading