$seo-ai-search

---
name: seo-ai-search
description: Review or improve a website's organic search and AI-search discovery, including technical SEO, indexing, content opportunities, crawler policies, citations, and conversion paths. Use for SEO/GEO/AEO strategy, search-console investigations, or applying AI-search research to a real business. Not a replacement for the site's editorial or deployment workflow.
---

# SEO and AI Search

Help the right audience discover a useful answer, understand why the business is relevant, and take an appropriate next step. Distinguish search visibility, AI citation, service/product recommendation, and business outcomes; they are not interchangeable successes.

## Establish the Actual Website

Before prescribing tactics, identify the relevant context from the request, project documentation, implementation, and available evidence:

- Audience, language, market, reader tasks, and the outcome the business wants.
- Business model, approved access model, and actual offer: a managed service, self-service product, publisher, retailer, marketplace, or another model. Distinguish existing capabilities from aspirations; preserve intentional subscriptions, paywalls, and private content.
- Lifecycle: first launch, improvement of an established site, migration, or investigation of a visibility/traffic change. Preserve existing search equity and verified setup where applicable.
- Canonical content owners, publication states, routes, rendering stack, hosting, and existing measurement tools.
- Crawl, retrieval, and training preferences. Do not assume visibility requires permission for model training.

Ask only when a missing decision materially changes the recommendation. For a narrow indexing investigation, establish only the context needed to diagnose it; do not turn every request into a full strategy project.

Inspect current project instructions and authoritative content plans before proposing changes. A skill is not a second content inventory, copy guide, roadmap, or source of deployment authority. When documents conflict, trace their owners and revisions rather than executing the most convenient checklist.

## Choose the Relevant Depth

- For search-engine setup, site-type checks, migrations, rendering, metadata, redirects, sitemaps, search-console alerts, crawler access, or release checks, read [technical-and-indexing.md](references/technical-and-indexing.md).
- For content opportunities, AI visibility, service/product discovery, useful resources, distribution, measurement, or adapting case studies, read [strategy-and-measurement.md](references/strategy-and-measurement.md).
- Read both for an end-to-end review; do not run every check for a focused question.

The references include primary documentation and the Railway/Ryze research that informed this skill. Recheck current provider documentation when behavior, support, permissions, or limits could have changed.

For a launch or indexing setup request, cover property access/verification, discovery/submission, and follow-up as well as technical readiness. Do not assume accounts are configured or treat an indexability audit as completed search-engine onboarding.

## Work from Evidence to the Smallest Useful Action

1. Establish what is observed: affected URLs or queries, source, date range, report freshness, and the user-visible or commercial consequence.
2. Follow the evidence from its canonical source through rendering, hosting/CDN behavior, and the search or AI surface where the problem appears. Separate local source, built output, deployed response, crawler access, indexing, and serving.
3. Classify the gap: technical eligibility, answer quality, distinct unmet task, business/offer clarity, relevant next action, external recognition, or measurement. Do not use content production to compensate for a technical block, or a technical rebuild to compensate for weak value.
4. Prefer an existing page or workflow when it can solve the need cleanly. Propose a new page, tool, integration, or automation only when it has a distinct present benefit and a maintainable owner.
5. Name the action, owner, expected benefit, evidence strength, dependencies, and how success will be checked. Rank by useful audience reach and business fit as well as demand, effort, risk, and ongoing maintenance. Use qualitative judgment unless a scoring system adds real value.

Keep observed findings, hypotheses, proposed experiments, user decisions, and completed work distinct. A saved setting is not proof of delivery; a sitemap submission is not proof of indexing; a citation is not proof of a qualified visit or sale.

## Preserve the Reader's Task

- Give a direct, sufficiently complete answer, then evidence, conditions, examples, limits, and useful continuation. Do not force a fixed word count, heading template, exact keyword, or promotional adjective into every opening.
- One page should resolve a distinct task or question. Keyword variants, cities, codes, industries, and source documents do not each justify separate pages by themselves.
- Separate audiences when their problem, outcome, or offer differs. A shared resource may still serve several audiences; do not create parallel copies automatically.
- Use crawlable contextual links and helpful hubs. Do not add intermediary pages simply to complete a taxonomy or hide unavailable destinations.
- For freely published guidance, make commercial continuation optional and relevant rather than withholding the promised answer. Respect the site's approved paid or registration-based offer; SEO is not authority to remove a gate or expose restricted content.
- Preserve truthful authorship, verification, sources, publication history, and update dates. Never invent reviewers, proof, customer outcomes, prices, or product functionality for SEO.

## Keep AI Search Proportionate

Conventional SEO and AI discovery share foundations. Diagnose additional needs instead of starting a separate AI-only architecture.

- Machine-readable alternates and `llms.txt` can make retrieval convenient; do not present them as ranking requirements or guarantees.
- Agent-readable content is different from an agent-operable product. MCP servers, plugins, anonymous trials, APIs, and one-command setup require a real product workflow, not merely a desire for citations.
- Do not copy SaaS distribution, review directories, comparison pages, or programmatic page volume onto a business where those tactics do not fit.
- Meaningful freshness follows changed information or a useful publication, not a schedule of cosmetic date changes.
- Prefer a focused pilot and existing measurement surfaces before a large page-generation programme or custom visibility platform.

## Authority and Handoff

An investigation or plan is read-only unless the user also authorizes changes. Publication, deployment, account settings, submissions, outreach, tracking, spending, and recurring monitoring require the authority appropriate to that action; invoking this skill does not grant it.

For authorized implementation, follow the project's existing editorial, consent/privacy, validation, and release process. Check representative affected HTML and any derived metadata/alternates; confirm live behavior after deployment only when deployment is in scope. Do not change visual design or product meaning as a hidden SEO fix.

Lead the report with the conclusion and highest-value next action. State evidence and uncertainty, what is already working, what should change and why, how it will be verified, and any user decisions needed. Report inaccessible accounts and unverified coverage explicitly rather than claiming that everything is healthy.