Use ai-seo to examine how a brand or resource appears in AI-search answers. Begin with explicit questions, platforms, and cited pages so observations can be compared. It is not a switch that guarantees recommendations after adding a special file.
Install and invoke
Run this in your project terminal to install only this skill for Claude Code:
npx skills add coreyhaines31/marketingskills --skill ai-seo -a claude-code
Node.js and npx must be available. For Codex, replace -a claude-code with -a codex. In the agent conversation, explicitly ask it to use ai-seo and supply your material. The installation command belongs in the terminal, not in your content brief.
Build a comparable baseline
Prepare ten to twenty questions customers might genuinely ask, covering categories, comparisons, and specific tasks. For each run, record the platform, date, exact prompt, response, and citation URLs. Distinguish a brand mention from a citation to your site or an explicit recommendation.
Example brief:
Use ai-seo to analyze questions.csv and answers.md.
Report mentions, citations to our site, and recommendations separately rather than merging them into one success rate.
Identify three content gaps, connecting each to the question, original answer, and evidence URL.
If browsing is unavailable, analyze only the supplied samples and do not invent platform responses.
Repeat questions if useful, recording the number of runs and the denominator. One missing mention does not prove a crawl block, and one citation does not establish a stable position. Compare what competing pages actually answer and substantiate instead of copying their headings mechanically.
Improve explainable content
Clarify definitions, requirements, pricing boundaries, and operating instructions. Support conclusions with real evidence and honest update dates. Structured data should match visible content. New comparisons or case studies need verified material, not invented reviews, credentials, or customer outcomes.
Treat crawler settings by platform. Google’s official documentation says Search AI features follow ordinary search fundamentals and relevant Googlebot access and snippet controls. Google-Extended concerns other systems; it is not a required switch for appearing in AI Search features. Google also does not require special AI text files or new schema for these features. Do not equate another platform’s training crawler with its search crawler either.
After changes, repeat the same questions using the same record format. Describe changed answers, cited URLs, and corrected misunderstandings while retaining samples that did not improve. Google includes traffic from these AI features in overall Web performance reporting in Search Console, so an aggregate increase cannot automatically be attributed to this work. When evidence is limited, report an observation rather than a causal result.
Turn observations into a repeatable page-level task
For a first audit, keep three columns: the question a buyer would ask, the page on your site that should answer it, and the page the answer actually cites. “How should I design RLS for a multi-tenant application?” calls for a practical permissions guide, not a brand homepage. Record brand mentions, citations to your site, and actual recommendations separately. If repeated runs disagree, preserve the disagreement instead of selecting the best-looking response.
Then diagnose each priority page. If a cited competitor explains definitions and boundaries while your page offers only a slogan, add a precise explanation, an example, and the conditions under which the advice applies. If your page already answers the question, inspect crawlability, titles, and whether the relevant paragraph can be understood in context. Do not fragment an article into disconnected sentences merely to appeal to an answer engine. Keep the original prompt, answer, date, platform, and cited URLs so the same question can be checked again.
Write the next step as a small table: page, missing answer, evidence needed, owner, and recheck date. Improve one useful guide first, then repeat the same question set. AI answers vary with platform and time, so a changed citation in one run is an observation, not proof that a particular edit caused it.
Example: test whether a migration guide answers real questions
Suppose you run a developer-tools site and want to know whether questions about moving authentication from an old SDK to a new one lead readers to your guide. Prepare five real phrasings of the question, your existing migration article, and AI answers with citation links that you saved manually. Ask: “Use ai-seo to compare these five question-and-answer records. For each one, show whether our migration guide was cited, which competing passage was used, and which actionable step our article lacks. Propose edits to one page only; do not claim to have queried platforms whose answers are not attached.”
A useful result points to a section, such as a missing old-to-new parameter mapping, and to the question that exposed the gap. If the response only says “write authoritative content” or “allow every AI crawler,” require it to explain each recommendation against the five records. After editing, repeat the same questions on the same platform with the same record format. A changed answer is an observation, not proof that a page edit increased traffic.
Check visits and installation after finding a citation
A brand mention in an AI answer does not mean a visitor reached your site. Inspect AI referral landing pages and 404s in your analytics. Ahrefs found referrals to nonexistent URLs in its own data. Log those paths to distinguish answer visibility from visits that actually arrive.
If ../../tools/REGISTRY.md is missing after installation, first check whether you installed one Skill with npx or the whole repository. A community issue reproduces the problem: a per-Skill install does not copy the repository-level tools/ directory. Read the registry in the official repository or use an installation that includes it; the missing file does not mean the AI visibility audit itself failed. A public pipeline loads ai-seo alongside site-structure, copy and schema Skill text as non-interactive system-prompt guidance in one strategy stage. It shows integration, not a separate slash-command run or a traffic result.