> ## Documentation Index
> Fetch the complete documentation index at: https://docs.zued.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# How zued works

> The audit pipeline: crawl, chunk, dispatch, compare, recommend. How zued turns your URLs into actionable insights.

zued runs a multi-step audit pipeline for every URL in your project. Each step exists for a specific reason. Together, they produce the alignment and technical scores, content gap analysis, and prioritized recommendations you see in your dashboard.

## The pipeline

<Steps>
  <Step title="Crawl your pages">
    zued fetches each URL using a real browser, the same way AI crawlers access your site. This captures the full rendered page, including JavaScript-dependent content, and measures [Technical Score](/metrics/technical-score) signals: bot accessibility, Core Web Vitals, structured data, and meta tags.
  </Step>

  <Step title="Chunk your content">
    Page content is split into meaningful segments that preserve the page structure. These chunks are the unit of comparison. zued doesn't compare the entire page at once, but rather the most relevant chunk for each prompt.
  </Step>

  <Step title="Generate ICPs and prompts">
    From your crawled content, zued extracts [Ideal Customer Profiles](/platform/icps), who your pages are written for. These ICPs shape the [audit prompts](/platform/prompts) generated for each URL. You review and approve all prompts before they're dispatched, editing, adding, or removing any that don't fit your audit goals.
  </Step>

  <Step title="Dispatch to AI engines">
    Prompts are sent to AI engines through their actual web interfaces: Gemini, ChatGPT, Google AI Mode, Microsoft Copilot, Perplexity, and Grok. Responses are collected including the full answer text, cited sources, and brand mentions. This is how real users experience these engines.

    Each prompt runs 10 times on each engine, because AI answers vary from run to run. This repeated sampling is what turns "did we appear" into a reliability measurement, your [Appearance Rate](/metrics/appearance-rate). One sampled response per prompt and engine becomes the analyzed response that goes through the deep comparison steps below; the remaining samples measure how reliably your brand appears, and all of them can be read in the Compare view.
  </Step>

  <Step title="Compare: your content vs. the AI consensus">
    For each prompt, zued maps the AI response to the most relevant content chunk on your page and compares them across four dimensions:

    * **Coverage**: does your content address the same topics the AI response covers?
    * **Angle**: does your content approach the topic the same way?
    * **Structure**: is it organized so AI can extract and cite it?
    * **Evidence**: does it back its claims with the statistics, sources, and examples the AI answer includes?

    This comparison produces the [Alignment Score](/metrics/alignment-score) and surfaces specific content gaps. For prompts that span multiple URLs, alignment runs per URL. Each page gets its own score and gaps.
  </Step>

  <Step title="Cross-URL analysis">
    When a prompt is relevant to multiple URLs in your project, zued runs [cross-URL analysis](/recommendations/cross-url-analysis): it aggregates per-URL alignment data, identifies internal linking gaps between pages, and recommends how your pages can work together as an ensemble, so AI engines are more likely to cite them.
  </Step>

  <Step title="Verify factual accuracy">
    For pages with verifiable claims (prices, specifications, features, guarantees) zued checks whether AI engines reproduce them accurately. Each claim is tested across all engines, producing per-claim verdicts that show which facts AI gets right and which it distorts. This feeds directly into page-level recommendations so you can prioritize fixing claims that AI gets wrong.
  </Step>

  <Step title="Identify information gain">
    Where your content goes beyond the AI consensus (original data, unique expertise, deeper analysis) zued identifies this as information gain. This is what makes AI engines more likely to choose your page over others.
  </Step>

  <Step title="Generate recommendations">
    Every gap and technical issue becomes a specific, scored recommendation. The [prioritization system](/recommendations/prioritization) ranks them by impact, effort, and how many engines are affected, so you always know what to fix first.
  </Step>
</Steps>

## Why this approach works

Most AEO/GEO tools track **outputs**, whether your brand shows up in AI responses. zued works at the **input** level: what your content contains, whether AI crawlers can reach it, how it compares to the AI consensus, and what specific changes would improve both alignment and technical access.

This is why zued audits weekly rather than daily. The AI consensus on a topic doesn't shift day-to-day. What you can change, your content and technical setup, takes time to update and deploy. Weekly snapshots give you time to implement recommendations and measure their impact against a consistent baseline.

## What makes the comparison meaningful

zued doesn't use APIs to reach AI engines. It uses the same web interfaces real users see. This matters because:

* API responses and web responses can differ significantly
* Web responses include search results, cited sources, and current information
* The comparison reflects what your actual audience experiences

The same prompts run every week against fresh content and fresh AI responses. This is what makes trend tracking reliable: you're measuring the same thing each time, not chasing moving targets.