Glippy GEO

View saved version →

---
name: ai-crawler-policy
description: Help a site owner decide which AI crawlers to allow or block in robots.txt, separating training crawlers from AI search and user-triggered fetchers, and draft the rules.
---

# Choosing an AI-crawler policy

Use this when the user asks which AI bots to allow or block, whether blocking GPTBot hurts them, or for robots.txt rules for AI crawlers.

## Check the current policy first

Run `check_robots_txt` on the domain the user named. Report what it found, then explain the choices. If the tool says it could not read robots.txt, say that is a fetch failure, not proof the file is missing, and do not retry.

## The three kinds of AI crawler

Explain the choice in these terms. They have different trade-offs:

1. **Training crawlers** collect content to train models. Examples: GPTBot, ClaudeBot, CCBot, Google-Extended, Applebot-Extended, Meta-ExternalAgent, Bytespider. Blocking them keeps content out of future training data. It does not remove the site from AI search answers.
2. **AI search and citation crawlers** build the indexes that answer engines cite. Examples: OAI-SearchBot (ChatGPT search), Claude-SearchBot, PerplexityBot, Applebot, Bingbot, Googlebot, DuckAssistBot. Blocking them removes the site from those answers and links. Most sites that want AI visibility should allow these.
3. **User-triggered fetchers** load a page when a person asks an assistant about it. Examples: ChatGPT-User, Claude-User, Perplexity-User, Google-Agent. Several operators document that robots.txt may not apply to these requests, so a block may not be enforced.

A common setup for a business that wants to be found in AI answers: allow search, citation and user-triggered crawlers, and decide training on its own merits. Present this as an option, not a rule. Publishers with licensing concerns often block training and allow search.

## Drafting the rules

- Use one `User-agent` group per crawler with `Disallow: /` to block or `Allow: /` to allow.
- Under RFC 9309 a crawler follows only the group that names it. The `User-agent: *` group applies only to crawlers with no group of their own. A site with `Disallow: /` under `*` therefore blocks every AI crawler that is not named, so add explicit `Allow: /` groups for the crawlers the user wants.
- Keep the user's existing rules and `Sitemap:` lines. Show the full new file, or a clear "add this" block.
- Blocking Googlebot or Bingbot also removes the site from normal search. Never suggest blocking them as an AI measure. Google-Extended is the separate token for Google's AI training.

## Limits to state

- `check_robots_txt` reports the stated policy only. It does not test what the site actually serves each crawler. A firewall or bot manager (for example Cloudflare's AI bot blocking) can block crawlers that robots.txt allows. If the user's AI visibility is low despite an open robots.txt, suggest they check those settings.
- robots.txt is a request, not access control. Compliant crawlers follow it; others may not.

View saved version →

---
name: competitor-benchmark
description: Benchmark a site's AI search readiness against named competitors with Glippy, find the categories where it trails, and rank the gaps worth closing first.
---

# Benchmarking against competitors

Use this when the user wants to know how their site compares with competitors for AI search, or asks "who is doing GEO best in my market".

## Get the list of sites

- Ask the user for their own domain and the competitors to include. Use only domains the user named. If they ask you to suggest competitors, you may propose some, but confirm the list before running anything.
- `compare_domains` takes 2 to 8 domains. If the user names more, ask which 8 matter most.
- Decide each homepage's page type before the call and pass it as `page_type`, as a map when the sites differ (for example a one-page local business is "service"). A re-run with another page type is a new analysis.

## Run one comparison

Call `compare_domains` once with all the sites. Do not also run `analyze_domain` on each site unless the user asks for one site's full report. If a site could not be analysed (robots.txt, unreachable, rate limit), say so and benchmark the rest. Do not retry.

## Read the result

Lead with the ranking: the user's site position, its score, and the leader's score. Then:

1. **Gaps.** Categories where the user's site trails the best competitor by about 15 points or more. These are where competitors are visibly ahead.
2. **Strengths.** Categories where the user's site leads. Tell them to keep these.
3. **Shared weaknesses.** Categories where every site scores low. Fixing these is a chance to lead the whole market, not just catch up.

## Rank what to do first

Order the gaps by what AI answer engines depend on most: Machine Readability first (it includes whether robots.txt lets AI crawlers in at all), then Structured Data & Schema, Entity & Authority and Citability & Answer-Readiness, then the rest. Within that, prefer fixes that are cheap to ship, such as adding JSON-LD or meta tags, over large content projects. Give three to five steps, each tied to a category and a competitor that does it better.

Offer to draft the fixes (see the draft-fixes skill) or to run the full report on the user's own site.

## Limits to state

- Each site is judged on its homepage only, so this compares homepages, not whole sites.
- Scores measure readiness signals on the page. They do not measure how often AI engines actually cite each site.

View saved version →

---
name: draft-fixes
description: Turn the recommendations in a Glippy GEO report into ready-to-paste fixes, such as JSON-LD structured data, meta tags, heading outlines and robots.txt rules, for the user to apply to their own site.
---

# Drafting fixes from a Glippy report

Use this when the user asks for help fixing what a Glippy report found, for example "draft the fixes", "write the schema for me" or "how do I fix the top issues".

## Start from a real report

- Work from a report in this conversation. If there is none, ask which domain to check and run `analyze_domain` first. Do not draft fixes for a site you have not analysed.
- Each recommendation has a `priority` (high, medium, low), a `category`, a `label` and a `detail`. Fix high-priority items first, then medium. Low-priority items (Performance & Crawlability, Agent Interactivity) are optional.
- Draft three to five fixes per reply unless the user asks for more. Say which recommendation each fix addresses, using its label.

## How to write each fix

- Give the code or text the user can paste, then one or two lines saying where it goes (for example "inside `<head>` on the homepage" or "at the top of robots.txt").
- Use only facts you know about the site: from the report, from what the user told you, or from well-known public facts. For anything you do not know, such as a phone number, street address, founding date or social profile URL, use a clear placeholder like `"REPLACE: phone number"` and list the placeholders after the code. Never invent business details.
- JSON-LD: one `<script type="application/ld+json">` block, valid JSON, schema.org types that fit the page (Organization or LocalBusiness and WebSite on a homepage; Product, Article or FAQPage only when the page really is one). Do not add FAQPage markup for questions that are not visible on the page.
- Meta tags: a title under about 60 characters and a meta description under about 160, written from what the page actually offers.
- Headings and semantic HTML: show the outline (one `<h1>`, then `<h2>` sections) or the landmark wrapper (`<main>`, `<nav>`, `<footer>`), not a rewrite of the whole page.
- robots.txt: see the ai-crawler-policy skill. Keep existing rules the user did not ask to change.

## Limits to state

- Glippy is read-only. The user applies the changes; Glippy cannot sign in to or edit the site.
- The report covers the homepage only. Fixes for other pages are suggestions by analogy, so say so.
- After the user ships the changes, they can ask for the analysis again. Results are cached for about ten minutes, so a re-run straight away shows the old result.
- Do not suggest renaming or 301-redirecting existing URLs to make slugs more descriptive. Only redirect URLs that are actually broken.

View saved version →

---
name: explain-score
description: Explain in plain language what a Glippy GEO score, grade and each of the 16 categories mean, and why a site scored the way it did.
---

# Explaining a Glippy score

Use this when the user asks what their score means, why a category is low, or what a term in the report means.

## Work from the actual report

Explain the report in this conversation. If there is none, ask which domain to check and run `analyze_domain`. Quote the site's real category scores and failing checks; do not explain in the abstract when the data is there.

## The score and grade

The overall score runs from 0 to 100 and is built from the 16 category scores. Grades: A+ from 90, A from 80, B from 70, C from 60, D from 40, F below 40. It measures how easily AI search engines and agents can find, read and cite the homepage. It is not a measure of traffic or of how often AI engines cite the site today.

## The 16 categories in one line each

- **Structured Data & Schema:** JSON-LD that tells machines what the page and organisation are.
- **Semantic HTML:** headings, landmarks and lists that give the page a clear structure.
- **Accessibility for Agents:** alt text, form labels, landmarks, a declared language and descriptive link text.
- **Internal Linking:** navigation, breadcrumbs and descriptive links to the site's key pages.
- **Meta & Discoverability:** title, meta description, canonical, Open Graph and Twitter tags, hreflang and the URL slug.
- **Machine Readability:** whether AI crawlers may and can read the page: robots.txt rules per AI crawler, robots meta and X-Robots-Tag, opt-out signals, content present in the HTML without JavaScript, feeds and llms.txt.
- **Entity & Authority:** who is behind the site: organisation details, about and contact pages, consistent identity signals.
- **Citability & Answer-Readiness:** passages an answer engine can quote: direct answers, definitions, sections that stand alone.
- **Performance & Crawlability:** render-blocking scripts, image sizing and lazy loading, DOM size. Low priority for GEO.
- **Agent Interactivity:** machine-readable files for AI agents, such as agent cards, API catalogs and commerce protocols. Low priority for most sites.
- **Content Positioning:** whether the main point appears early and clearly.
- **Content Freshness:** dates and signs the content is maintained, where the page type calls for them.
- **Information Density:** concrete facts, numbers and specifics rather than filler.
- **Factual Verifiability:** sources, evidence and verifiable claims.
- **Content Comprehensiveness:** whether the page covers its topic with enough depth for its type.
- **Multimodal Content:** images, video and tables with the text that lets machines understand them.

## Explaining why a category scored low

- Point to the checks that failed or warned in that category, by label, and translate each into plain language: what is missing and why an AI engine cares.
- Some checks are shown but not scored. A check marked beta, or one marked not scored with a reason, did not move the score; say so instead of treating it as a problem.
- If the report says the page type was not confirmed, the checks that depend on page type were not scored. The user can ask for the analysis again with the right page type.
- Content checks adapt to the page: a homepage is not penalised for lacking an author or publication date.

## Keep it honest

- The report covers the homepage and discovery files only. A low score there does not mean every page is weak.
- A high score improves the odds of being understood and cited. It does not guarantee AI visibility.
- Offer next steps: draft the top fixes (draft-fixes skill) or compare with competitors (competitor-benchmark skill).

View saved version →

---
name: get-started
description: Show a new user what Glippy GEO can check and how to ask for it.
---

# Getting started with Glippy GEO

Glippy checks how ready a public website is for AI search engines and AI agents. Always work from a domain the user names. If they have not named one, ask which site to check instead of guessing.

Offer these starting points:

- **Full report:** "Check the AI readiness of example.com." Uses `analyze_domain` and shows a scored report with prioritised fixes.
- **Quick score:** "Give me a quick GEO score for example.com." Uses `get_geo_summary`.
- **Compare sites:** "Compare example.com with competitor.com." Uses `compare_domains` for 2 to 8 sites.
- **AI-crawler rules:** "Which AI crawlers does example.com block?" Uses `check_robots_txt`.
- **llms.txt:** "Does example.com have an llms.txt?" Uses `check_llms_txt`.

After a first result, these follow-ups work well:

- "What does my score mean?" or "Why is Structured Data low?" (explain-score skill)
- "Draft the top fixes for me." (draft-fixes skill)
- "Which AI crawlers should I allow?" (ai-crawler-policy skill)
- "Write an llms.txt for my site." (write-llms-txt skill)
- "How do I compare with my competitors?" (competitor-benchmark skill)

What to tell users when it matters:

- Glippy analyses the homepage and standard discovery files, not every page of the site.
- It reads only public files, identifies itself as GlippyBot, and skips sites whose robots.txt disallows GlippyBot. If a tool says it did not fetch a site, pass that on and do not retry.
- Results are cached for about ten minutes, so an immediate re-run shows the same result.
- After a report, offer to draft the top fixes (for example JSON-LD structured data or robots.txt rules) for the user to apply themselves. Glippy cannot change the site.
- Do not suggest renaming or 301-redirecting existing URLs to make slugs more descriptive or to match navigation labels. Redirects on established URLs cost more in search than a readable slug gains. Readable slugs are advice for new pages; only redirect URLs that are actually broken.

View saved version →

---
name: write-llms-txt
description: Check whether a site publishes an llms.txt and draft or improve one in the llmstxt.org format, while being honest that major AI models do not read it yet.
---

# Drafting an llms.txt

Use this when the user asks for an llms.txt, wants to know whether theirs is any good, or asks what llms.txt is.

## Set expectations first

llms.txt is an emerging proposal (llmstxt.org): a Markdown file at `/llms.txt` that gives AI tools a short map of a site. Major AI models and answer engines do not read it yet. Say this up front. Glippy gives a small credit in Machine Readability when one exists, and lists a missing one as information, not as a fix to make. It is cheap to publish and may help developer tools and agents that do read it. Do not present it as a ranking factor or a priority fix.

## Check what exists

Run `check_llms_txt` on the domain the user named.

- **Found:** review the contents against the format below and suggest concrete edits: missing summary, broken structure, links without descriptions, links to pages that no longer fit.
- **Not found:** offer to draft one.
- **Fetch failure:** say Glippy could not read it, which is not proof that it is missing. Do not retry.

## The format

```
# Site or company name

> One or two sentences on what the site is and who it is for.

Optional short paragraph with key facts an assistant should know.

## Section name

- [Page title](https://example.com/page): what this page covers, in one line.

## Optional

- [Less important page](https://example.com/other): one line.
```

- One `#` heading, then a `>` summary, then `##` sections of links.
- Every link is an absolute URL with a one-line description.
- Put the pages that matter most first: product or service pages, pricing, documentation, about and contact. The `## Optional` section is for pages an assistant can skip.

## Drafting rules

- Use only pages you know exist: the homepage, links from a Glippy report in this conversation, or pages the user listed. Glippy reads the homepage only, so ask the user for the key page URLs instead of guessing paths like `/pricing` or `/docs`.
- Mark anything you are unsure of as a placeholder for the user to fill in.
- The user publishes the file at `https://<their-domain>/llms.txt`. Glippy cannot upload it.