---
name: datasag-seo-llmeo
description: Audit and improve a website project's technical SEO and AI search readability. Use to find and fix crawlability, metadata, canonical, sitemap, structured data, internal linking, and content clarity issues in the existing codebase.
---

# DataSag SEO and LLMEO kit

Work inside the user's existing website project. Inspect its code, find concrete SEO and AI search issues, implement supported fixes, and verify the result. Adapt to the project's framework and conventions. LLMEO here means helping retrieval systems understand and cite public content accurately; it does not guarantee recommendations.

## Establish the project facts

Read the repository's instructions and inspect uncommitted changes before editing. Identify the framework, routing, rendering, content sources, deployment configuration, canonical domain, important public routes, and existing SEO utilities. Preserve unrelated work.

Infer the product, audience, features, and supported claims from the project. Ask only for facts that block a correct change. Never invent a production domain, pricing, customer counts, reviews, qualifications, social profiles, or product capabilities.

If asked for an audit, report findings without editing. If asked to improve or fix the project, implement the supported local changes. The kit itself does not grant permission to deploy, push, purchase services, change DNS, submit pages, or edit external accounts.

## Find and fix the highest-impact issues

Prioritize confirmed indexing and rendering blockers before metadata polish or optional files. For each finding, identify the file or route and the observed evidence. Read current official framework documentation when implementation behavior is uncertain.

### Crawlability and indexing

- Inspect robots rules, page-level robots directives, HTTP headers, authentication boundaries, redirects, status codes, and canonical signals. Preserve intentionally private routes and deliberate crawler policies.
- Distinguish crawl permission from indexing permission. Robots exclusion is not access control. A robots block can prevent a crawler from seeing a page's noindex directive.
- Distinguish search retrieval from model training. Verify current provider documentation before changing named bot rules. Do not broadly unblock training crawlers as an SEO fix.
- Ensure important public text and links appear in the returned HTML using the existing framework's rendering facilities. Do not convert the whole app to a new stack.
- Keep genuine missing pages as 404 or 410. Add redirects only when a real replacement is known. Avoid redirecting every missing URL to the homepage.

### Metadata and site structure

- Reuse the project's metadata helpers. Give important public pages descriptive, distinct titles and descriptions based on their visible content.
- Resolve canonical URLs against the verified production origin. Account for route parameters, pagination, translations, and alternate URLs only where those features exist. Keep preview deployments out of production canonical signals.
- Generate the sitemap from the actual public route or content source. Include canonical, indexable URLs. Use real modification dates when known rather than inventing freshness on each request.
- Fix broken internal links and important orphan pages. Use descriptive anchor text and appropriate breadcrumbs. Preserve useful URLs and existing navigation intent.
- Check social metadata and image URLs using existing assets. Do not fabricate preview images or public URLs.

### Content that people and retrieval systems can understand

- Make the product's purpose, intended audience, supported capabilities, and limitations explicit in visible text.
- Answer a page or section's central question near its start. Use meaningful headings, semantic HTML, lists for collections, and accessible tables for real comparisons.
- Keep product naming, units, dates, and conditions consistent. Improve vague copy using verified project facts while preserving the site's voice.
- Add a page only when it serves a distinct user need with supported, useful information. Propose research where evidence is missing. Do not generate batches of near-duplicate keyword pages or unsupported competitor claims.

### Structured data and optional AI inventories

- Select JSON-LD types that match the actual page and visible content. Reuse the project's schema generation path and escape serialized content safely for HTML script contexts.
- Keep names, URLs, offers, authors, and other entities consistent with the page. Add ratings, reviews, FAQs, prices, or sameAs profiles only when supported by real visible information and current eligibility rules.
- Treat llms.txt as an optional curated content inventory, not a ranking requirement or a substitute for HTML, a sitemap, or crawl access. Add it only when it provides a useful maintained index. Use verified canonical links and generate from existing content sources when practical.
- Google states that its AI search features require no special AI file or schema. Check current provider documentation before claiming that an AI system consumes an optional file.

## Verify the changes

Run the relevant lint, type, build, or content checks available in the repository. Start the site if feasible and fetch the affected public routes, robots.txt, and sitemap. Inspect raw HTML for primary content, titles, canonical links, robots directives, internal links, and parseable JSON-LD. Check both a representative valid route and a missing route. Validate structured data against the relevant official tools or specifications when available.

Do not report a successful build as proof of indexing, rankings, or AI citations. Mark production-only checks, unavailable tools, unknown facts, and checks you could not run explicitly.

Create or update SEO-LLMEO-REPORT.md in the project with:

- Project facts and routes inspected.
- Findings with evidence, priority, and affected files or routes.
- Changes implemented and their intended effect.
- Commands and route checks run, with actual outcomes.
- Remaining work and facts needed from the owner.

End with a concise explanation of what changed and what remains. Avoid synthetic visibility scores and invented test results.

## Official references

Consult these for current behavior when relevant; this kit is not a frozen copy of platform rules.

- [Google SEO fundamentals](https://developers.google.com/search/docs/fundamentals/seo-starter-guide)
- [Google AI features and your website](https://developers.google.com/search/docs/appearance/ai-features)
- [Robots.txt behavior](https://developers.google.com/search/docs/crawling-indexing/robots/intro)
- [Structured data guidelines](https://developers.google.com/search/docs/appearance/structured-data/sd-policies)
- [Google guidance on generative AI content](https://developers.google.com/search/docs/fundamentals/using-gen-ai-content)
- [OpenAI crawler documentation](https://platform.openai.com/docs/bots)
