Skip to main content
Free Check
Methodology

The Machine-First Architecture framework

A four-pillar methodology for making websites visible and useful to AI agents. Authored by Web Performance Tools co-founder Slobodan Manić.

Get your site audited
Identity
Entity control
Structure
Crawl visibility
Content
Citation readiness
Interaction
Agent friction
The shift

Why this exists

ChatGPT, Perplexity, Claude, and Google AI Overviews now research products, compare options, and complete transactions on behalf of users. These are not future scenarios. They are current behavior, at scale, across every B2B category.

These agents read websites differently than browsers do. They operate on raw HTTP responses, partial DOM states, and time-limited crawl windows. A page that looks perfect in Chrome may be functionally empty to GPTBot.

Traditional SEO frameworks optimise for search engine crawlers that render JavaScript and index content over hours or days. AI agents operate in seconds, often without JavaScript execution. The gap between what a browser sees and what an AI agent sees is the gap this methodology measures and closes.

A dedicated methodology was needed: one that measures what AI agents actually see, identifies the specific technical gaps, and provides a structured remediation path.

Framework

The four pillars

Each pillar targets a distinct failure mode in how AI systems consume web content.

Who AI thinks you are

Identity

What it measures

Whether AI systems can correctly identify the company as a distinct, real-world entity, and not confuse it with similarly-named competitors or related companies.

Technical scope

Validation of JSON-LD Organization schema, uniqueness of the root node identifier (@id), audit of the sameAs array against verifiable external entity profiles. Detection of "Signal Drift," where entity grounding has fragmented across multiple inconsistent representations.

Why it matters

AI systems build internal representations of companies by aggregating signals from across the web. If your entity grounding is inconsistent or missing, you get confused with competitors, attributed to the wrong category, or omitted from responses entirely.

What AI crawlers actually see

Structure

What it measures

Whether the commercial content on the site is actually visible to AI crawlers, or hidden behind JavaScript that crawlers don’t execute.

Technical scope

Character-level comparison between raw HTTP fetch payload (no JavaScript execution) and fully hydrated DOM (via headless browser, networkidle state). Time-stratified measurement via the Agent Visibility Curve: content visibility measured at 1s, 3s, 5s, and 10s from time-to-first-byte, mapping against the operating windows of real AI crawlers.

Why it matters

AI crawlers have time budgets. A site that hydrates fully in 12 seconds is functionally invisible to a crawler that operates on a 3-second window. The Agent Visibility Curve shows exactly which crawlers can see how much of your site.

Whether you get cited

Content

What it measures

Whether the site’s content is structured and dense enough to be extracted and reused by AI systems, or whether it’s vague marketing prose that gets discarded.

Technical scope

Detection of root-level /llms.txt configuration. Analysis of the first 200 words of primary structural blocks (<main> or <article>) for explicit entity definitions, dense factual content, and machine-readable structure. Modularity audit: can sections of the page be extracted and quoted independently, or is meaning trapped in implicit context?

Why it matters

Large language models select around 380 words from any given page when deciding what to cite. If those 380 words are marketing taglines instead of substantive content, you don’t get cited.

Whether agents can take action

Interaction

What it measures

Whether AI agents can actually take action on the site (filling forms, clicking buttons, completing flows) or whether the conversion paths are technically inaccessible to agents.

Technical scope

Audit of primary conversion CTAs for non-semantic tags (<div> or <span> elements with client-side onClick handlers, instead of native <button> or <a href> elements). Verification that form error and success states scale programmatically via standard DOM attributes (aria-invalid="true", ARIA live regions, semantic state markers).

Why it matters

AI agents that act on behalf of users (Anthropic’s computer use, OpenAI’s agentic browsing, the wave following them) are already completing commercial transactions. Sites whose CTAs are technically inaccessible to agent automation will lose to sites whose CTAs work.

Application

How the framework is applied

The four pillars and the Agent Visibility Curve together form the diagnostic basis for the Agent Readiness Audit. Every audit measures all four pillars across up to 25 pages, produces per-page Agent Visibility Curves, and delivers a prioritized remediation roadmap.

The result is a specific, technical answer to the question: “How visible is our site to AI agents, and what exactly needs to change?”

Learn about the Agent Readiness Audit
Get started

See how your site scores

The free Agent Visibility Check runs one page through the Agent Visibility Curve diagnostic and sends you the result. No commitment required.

Get your free Agent Visibility Check