agentleFS
kristian-pressokristian-presso/claude-marketing-skills/skills/ai-search-visibility/SKILL.md

ai-search-visibility

Skill0 starsChanged 32 days agoMITView on GitHub

Audits a site or page for how well AI assistants and answer engines (ChatGPT, Claude, Perplexity, Google AI Overviews, Copilot) can find, read and cite it, and rewrites content so it is answer-shaped and quotable. Use when someone asks why AI tools do not mention their business, wants a GEO or AEO audit, asks about llms.txt or AI crawler access in robots.txt, wants a page made more citable, or wants to track AI share of voice. Triggers on "AI search visibility", "GEO audit", "AEO", "get cited by ChatGPT", "show up in AI Overviews", "answer engine optimisation", "make this page answer-shaped", "AI share of voice".

11 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use schema-markup and seo-audit.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • silvainfmSkill

    schema-markup

    When the user wants to add, fix, or optimize schema markup and structured data on their site. Also use when the user mentions "schema markup," "structured data," "JSON-LD," "rich snippets," "schema.org," "FAQ schema," "product schema," "review schema," or "breadcrumb schema." For broader SEO issues, see seo-audit.

    Also in 4 projects that use this

  • AnthropicSkill

    seo-audit

    Run a comprehensive SEO audit — keyword research, on-page analysis, content gaps, technical checks, and competitor comparison. Use when assessing a site's SEO health, when finding keyword opportunities and content gaps competitors own, or when you need a prioritized action plan split into quick wins and strategic investments.

    Also in 5 projects that use this

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 3 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 11 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If they give a URL with no other instruction, run Audit.

    line 23
  2. Very specific:

    If they ask for a rewrite but you can see a crawl blocker (robots.txt, noindex, content behind JavaScript), the first line of your answer …

    line 25
  3. Distinctive:

    Run the bundled script against the live site, not the repo:

    line 36
  4. Very specific:

    If you cannot run scripts (for example on claude.ai), fetch /robots.txt, /llms.txt, /sitemap.xml and a few key pages with whatever fetch …

    line 47
  5. Very specific:

    Search the person's top three target questions yourself (web search tool if you have one) and look at which pages get cited.

    line 75
  6. Distinctive:

    If the person has an SEO or AI-visibility tool connected (an MCP server or API with brand or AI-citation tracking), use it to populate the …

    line 145

“Common” is measured: kept by at least 6 in 10 of the 11 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

AI search visibility

More buyers now ask an assistant before they search. Being cited in that answer is a different job from ranking a blue link, and most of the failures are silent: the page looks fine in a browser, nothing errors, and the crawler that feeds the answer was never let in.

This skill works in a fixed order. Crawlable before quotable. Better content on a page no AI crawler can read changes nothing.

Pick the mode

The person wantsMode
"Why are we not showing up in AI answers?" / an audit of a site or URLAudit
A draft or live page made more citableRewrite
A repeatable way to track whether it is workingMeasure

If they give a URL with no other instruction, run Audit. If they paste prose, run Rewrite.

If they ask for a rewrite but you can see a crawl blocker (robots.txt, noindex, content behind JavaScript), the first line of your answer names the blocker and says the rewrite changes nothing until it is fixed. Then do the rewrite.

Mode 1: Audit

Work through the four layers in order and stop to report if a lower layer is broken. A blocked crawler makes every finding above it moot.

Layer 1: can the AI crawlers get in?

Run the bundled script against the live site, not the repo:

python3 scripts/ai_visibility_audit.py https://example.com --max 25

It reports, for each main AI agent, whether robots.txt allows it at the root, whether the agent is named explicitly, any rules written for agent names that do not exist, whether llms.txt exists, and per-page meta, headings and JSON-LD health. Standard library only, no install. Use --single for one URL and --json for machine-readable output.

If you cannot run scripts (for example on claude.ai), fetch /robots.txt, /llms.txt, /sitemap.xml and a few key pages with whatever fetch tool you have and apply the same checks by reading. Read references/technical-discoverability.md for the agent table and the checks in full.

The findings that matter most, in order:

  1. Search and user-fetch agents blocked. Blocking a training crawler is a business choice. Blocking the agent that builds the citation index (for example OAI-SearchBot, Claude-SearchBot, PerplexityBot, Bingbot) means no citations from that engine. Say which is which; do not tell anyone to unblock training crawlers.
  2. Rules for agents that do not exist. ChatGPTBot is the common one. It matches nothing.
  3. Key pages missing a meta description, canonical or parseable JSON-LD. When this is sitewide, trace it end to end (content field, page component, layout, deployed HTML) before fixing; it is usually two missing links, not one.
  4. Sitemap and noindex disagreeing. A page in the sitemap that says noindex.
  5. Substance behind JavaScript, logins, forms or cookie walls. Fetch the raw HTML and check the main answer is in it.

Layer 2: is the content answer-shaped?

For the five to ten pages that matter commercially (pricing, the main service pages, the best guides), check each against the answer-shape list below. Read references/answer-shaped-content.md for the patterns and examples.

Layer 3: is it worth citing over a competitor?

Search the person's top three target questions yourself (web search tool if you have one) and look at which pages get cited. For each, compare: original data, named author, dated claims, specificity, sources linked. Read references/citation-worthiness.md.

Layer 4: does the wider web agree about them?

Answer engines lean on third-party sources: directories, review sites, comparison articles, forum threads. Note where competitors are listed and the business is not, and where the business is listed with an inconsistent name or description.

Audit output

# AI search visibility audit: <site>, <date>

## Verdict
<One paragraph. The single biggest blocker and what fixing it changes.>

## Fix first (silent failures)
| # | Finding | Evidence | Fix | Effort |

## Then (content and authority)
| # | Page | What is missing | Fix |

## Off-site gaps
| Source | Competitors listed? | You listed? | Action |

## Not checked
<Anything you could not see, and why. Never leave this out.>

Lead with a recommendation, not a menu. Order fixes by impact over effort.

Mode 2: Rewrite

Take a draft or a live page and return a rewritten version plus a short list of what changed. Keep the person's voice and facts. Never invent a statistic, quote, date or source to satisfy the checklist; where one is missing, leave a marked placeholder like [SOURCE NEEDED: average setup time] and list it at the end.

The answer-shape checklist:

  • The first 40 to 60 words after the title answer the title's question directly.
  • Each section opens with its answer in one or two sentences, then supports it.
  • At least two headings phrased the way someone would type or say the question.
  • Every section makes sense alone. No "as mentioned above", no pronouns pointing out of the section, acronyms defined again where they are used.
  • Key terms defined in the form "X is Y".
  • Any X versus Y comparison is a table with short cells.
  • Specific numbers with a named source and a date, not "many" or "most".
  • A named author and a visible last-updated date.
  • Time-sensitive claims dated ("as of March 2026").
  • Nothing important behind a form, login or click-to-reveal.
  • A FAQPage or HowTo block suggested only if the page genuinely is Q and A or steps, and only marking up what is visible on the page.

Return: the rewritten text, then a "Changed" list (one line each), then any placeholders the person must fill.

Mode 3: Measure

Rank tracking misses this entirely. The useful metrics are share of voice in AI answers, citation share, how the business is described, which competitor URLs get cited instead, and AI-referred traffic.

Set up a prompt panel: 15 to 50 real questions buyers ask, run the same way on the same surfaces each month, recorded in a table. Read references/measurement.md for the panel method, the free data sources (Google Search Console, Bing Webmaster Tools AI Performance, analytics referrers, server logs) and a monthly review template.

If the person has an SEO or AI-visibility tool connected (an MCP server or API with brand or AI-citation tracking), use it to populate the same metrics. If not, run the panel manually or with a web search tool and say which surfaces you could and could not query.

Rules

  • Fetch the deployed site. Templates lie. Count what is served.
  • Report what you could not check. A clean report that skipped three layers is worse than an honest partial one.
  • No invented evidence. Placeholders, not plausible-sounding numbers.
  • Facts about engines drift. Agent names, report availability and surface behaviour change. Check references/changes-ledger.md and the vendor's own docs before stating one as current.
  • No guarantees. Nobody controls whether a model cites a page. Say "more likely", not "will".

Example

User: Why does ChatGPT never mention us? We're a small venue-hire company. Site is example-venues.co.uk.

Claude: (runs the audit script) The main blocker is in robots.txt: it has a rule for ChatGPTBot, which is not a real agent, and a Disallow: / under a wildcard for anything not named. So OAI-SearchBot, the crawler behind ChatGPT's citations, is blocked, and so are Perplexity's and Claude's search crawlers. Nothing else matters until that is fixed. Second, 14 of 22 pages have no meta description, and the pricing page's JSON-LD block has a trailing comma, so it does not parse. Fix those three, then I will rewrite the pricing and "how to book" pages so the first paragraph answers the question.

Configuration

Two things vary per person:

  • The target questions. The 15 to 50 buyer questions used for Layer 3 and the prompt panel. Ask for them, or draft a list from their site and confirm it.
  • The crawler policy. Whether they want to allow AI training crawlers as well as search crawlers. That is their call; this skill only makes sure the choice is deliberate.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.