# AI SEO Audit

Runs a deterministic AI-SEO audit on a page through context.dev — scoring ~30 checks across six categories into a 0–100 total, returning failing checks and an agent-ready fix prompt.

> For the complete documentation index, see [llms.txt](/llms.txt). Markdown variants are available by appending `.md` to any URL or sending an `Accept: text/markdown` header. An agent skill is available at [/.well-known/agent-skills/site-skill.md](/.well-known/agent-skills/site-skill.md).



<DocsBaseSwitcher base="mastra" agent="ai-seo-audit" />

<AgentPreview
  agent="ai-seo-audit"
  framework="mastra"
  inputFields="[
  {
    name: &#x22;url&#x22;,
    label: &#x22;Page URL&#x22;,
    placeholder: &#x22;https://example.com/blog/how-it-works&#x22;,
    type: &#x22;text&#x22;,
  },
]"
/>

## Summary [#summary]

The **AI SEO Audit Agent** evaluates a page for how readable it is to AI answer
engines — ChatGPT, Claude, and Perplexity — rather than classic blue-link
search. It fetches the page through context.dev (never touching the site
directly), then runs a **deterministic rubric** of \~30 checks across six
categories to a 0–100 total. The model doesn't eyeball the score — `audit_page`
computes it in `lib/rules.ts`, so the same page always yields the same
result. The agent presents the breakdown, lists the failing checks by impact,
and returns an **agent-ready fix prompt** you can paste straight into a coding
agent. Reach for it when you want a page to get cited by AI assistants.

## Installation [#installation]

<CodeTabs>
  <TabsList>
    <TabsTrigger value="cli">
      Command
    </TabsTrigger>

    <TabsTrigger value="manual">
      Manual
    </TabsTrigger>
  </TabsList>

  <TabsContent value="cli">
    ```bash
    npx shadcn@latest add @agentcn/mastra/ai-seo-audit
    ```
  </TabsContent>

  <TabsContent value="manual">
    <Steps>
      <Step>
        Install the following dependencies:
      </Step>

      ```bash
      npm install @mastra/core context.dev
      ```

      <Step>
        Copy and paste the following code into your project.
      </Step>

      <ComponentSource src="registry/mastra/ai-seo-audit/config.ts" title="config.ts" />

      <ComponentSource src="registry/mastra/ai-seo-audit/instructions.md" title="instructions.md" />

      <ComponentSource src="registry/mastra/ai-seo-audit/tools/audit_page.ts" title="tools/audit_page.ts" />

      <ComponentSource src="registry/mastra/ai-seo-audit/lib/audit.ts" title="lib/audit.ts" />

      <ComponentSource src="registry/mastra/ai-seo-audit/lib/rules.ts" title="lib/rules.ts" />

      <ComponentSource src="registry/mastra/ai-seo-audit/lib/agent-fix-prompt.ts" title="lib/agent-fix-prompt.ts" />

      <ComponentSource src="registry/mastra/ai-seo-audit/lib/audit-types.ts" title="lib/audit-types.ts" />

      <ComponentSource src="registry/mastra/ai-seo-audit/lib/url.ts" title="lib/url.ts" />

      <ComponentSource src="registry/mastra/ai-seo-audit/skills/seo-audit/SKILL.md" title="skills/seo-audit/SKILL.md" />

      <Step>
        Update the import paths to match your project setup.
      </Step>
    </Steps>
  </TabsContent>
</CodeTabs>

## Composition [#composition]

```text
├── config.ts              # Agent definition (model + config)
├── instructions.md        # Audit-and-report instructions
├── tools/
│   └── audit_page.ts      # Runs the deterministic audit, returns the scored result
├── lib/
│   ├── audit.ts           # Fetches via context.dev, parses, and scores the rules
│   ├── rules.ts           # The ~30 checks and six category weights
│   ├── agent-fix-prompt.ts # Builds the copy-paste fix prompts from the findings
│   ├── audit-types.ts     # Shared result types
│   └── url.ts             # URL normalization helpers
└── skills/
    └── seo-audit/
        └── SKILL.md       # How the agent presents the audit_page result
```

## Customization [#customization]

* **Fetching.** The page is fetched through [context.dev](https://context.dev) —
  the recipe never hits the audited site directly. Set `CONTEXT_DEV_API_KEY`.
* **Tune the rubric.** Edit `lib/rules.ts` to add checks, reweight
  categories, or change the scoring bands — this is the single source of truth
  for the score.
* **Reshape the output.** Edit `instructions.md` or `SKILL.md` to emit
  JSON, a shorter summary, or a different fix-prompt format.
* **Swap the model.** Edit `config.ts`. The model only narrates the
  deterministic result, so any capable model works.