agentleFS
minhnv0807minhnv0807/fullstack-mkt-skills/skills/en/32-seo-growth-global/SKILL.md

32-seo-growth-global

Skill644 starsChanged 6 months agoMITView on GitHub

Use when the user wants SEO growth, AI SEO, GEO/AEO, Google visibility, ChatGPT or Perplexity citations, schema, structured data, llms.txt, programmatic SEO, local SEO, comparison and alternative pages, directory submissions, Product Hunt, G2/Capterra, backlinks, or a site audit. Trigger on 'SEO audit', 'AI search visibility', 'AI citations', 'schema markup', 'pSEO', 'alternative page', 'submit to directories', or 'backlink plan'. Not for — first-time Google Business Profile setup, see `11-channel-setup-global`; writing the articles themselves, see `36-content-brief-global`.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

14 rules. About 0 are common, 12 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈12

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read .agents/product-marketing-context-global.md if available.

    line 36
  2. Distinctive:

    Run in order — a failure at a lower layer nullifies everything above it.

    line 59
  3. Distinctive:

    Read field data first (Search Console > Core Web Vitals, CrUX), lab data second (PageSpeed Insights).

    line 94
  4. Distinctive:

    When two of your own pages compete for the same query, both underperform.

    line 100
  5. Distinctive:

    Search Console > Performance > filter by query > check the Pages column.

    line 102
  6. Very specific:

    Run site:domain.com [keyword] — several near-identical pages means consolidate.

    line 103
  7. Distinctive:

    Ask at the moment of satisfaction, right after the service is delivered, with a short direct link.

    line 148
  8. Distinctive:

    Do not clone one page and swap the city name.

    line 159
  9. Distinctive:

    Challengers: should still publish genuinely useful guides, but set the expectation correctly — citation and category framing, not near-term …

    line 196
  10. Distinctive:

    When a competitor is cited and you are not, check five things: is their content more extractable, do they cite statistics and sources, is …

    line 269
  11. Distinctive:

    Avoid: bulk link packages, PBNs, comment spam.

    line 325
  12. Distinctive:

    Do not recommend content before clarifying search intent.

    line 377
  13. Distinctive:

    Every page has a CTA and a funnel role.

    line 378
  14. Distinctive:

    If a self-ranked listicle is recommended, expectations are stated by market position.

    line 384

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “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

SEO Growth (Global)

Operating rules: skills/en/references/operating-rules-global.md (esp. rules 2, 3, 7).

This skill covers 6 practical SEO layers: foundation audit, local SEO, search-intent content, AI search visibility, schema/machine-readable files, and backlink/distribution. The goal is qualified demand, not publishing volume.

Information gathering

Read .agents/product-marketing-context-global.md if available. Ask up to 4 questions: website URL, target market, main product/service, and 90-day SEO goal. Use Search Console, GA4, Ahrefs, Semrush or similar exports when available.

Three companion references:

  • references/seo-growth-playbook.md — 90-day roadmap, pSEO, comparison and alternative pages, site architecture.
  • references/schema-json-ld.md — 12 copy-paste JSON-LD blocks with current rich-result status.
  • references/ai-visibility-measurement.md — prompt portfolio, answer audit, AI referral traffic, crawler log check, fix loop.

Pick the mode

User saysMode
"Traffic dropped", "site is not ranking"SEO audit
"We don't show up on Maps / in the local pack"Local SEO
"Get cited by ChatGPT/Perplexity"AI SEO / GEO
"Add schema", "rich results"Schema
"Create many SEO pages"Programmatic SEO
"Build alternative / vs pages"Comparison pages
"Submit to Product Hunt/G2/directories"Directory + backlink

Workflow

1. Audit the foundation

Run in order — a failure at a lower layer nullifies everything above it.

1.1 Crawl and index

CheckPassing thresholdTool
robots.txt does not block money pages0 revenue pages disallowedSearch Console > robots.txt report
XML sitemap exists, declared in robots.txtPresent, auto-updatingSearch Console > Sitemaps
Indexed pages / pages you want indexed>90%Search Console > Pages
Redirect chains1 hop max, no loopsScreaming Frog, Ahrefs Site Audit
Soft 404s0Search Console > Pages > "Soft 404"
Canonicals point at the right page100% self-canonical or pointing at the true originalAny crawler
Leftover noindex from staging0 pagesGrep noindex site-wide

Two issues account for most "the site just won't rank" cases: site-wide canonicals pointing at the homepage (misconfigured theme or plugin) and a noindex left over from staging after go-live.

1.2 On-page

ElementThresholdNote
Title50-60 charactersLonger gets truncated; primary term near the front
Meta description150-160 charactersNot a ranking factor, but drives CTR
H1Exactly one per pageSame topic as the title, not a copy of it
H2/H3Phrased the way people type queriesThis is the scaffolding AI engines extract passages from
ImagesDescriptive alt, meaningful filenamesWebP; no 4MB originals
Internal linksEvery page has at least one inbound linkSee site architecture in the playbook

1.3 Core Web Vitals

MetricGoodNeeds improvementPoor
LCP (Largest Contentful Paint)< 2.5s2.5-4.0s> 4.0s
INP (Interaction to Next Paint)< 200ms200-500ms> 500ms
CLS (Cumulative Layout Shift)< 0.10.1-0.25> 0.25

Read field data first (Search Console > Core Web Vitals, CrUX), lab data second (PageSpeed Insights). A 100 desktop score with a 5s mobile field LCP is common — mobile is what matters.

Usual LCP offenders: uncompressed hero images, auto-playing sliders above the fold, fonts loaded from multiple origins, and synchronous chat or popup scripts.

1.4 Cannibalization

When two of your own pages compete for the same query, both underperform. How to detect it:

  1. Search Console > Performance > filter by query > check the Pages column. If one query rotates between 2+ pages, that is cannibalization.
  2. Run site:domain.com [keyword] — several near-identical pages means consolidate.

Fix in this order: merge into the strongest page and 301 the rest → or split the search intent cleanly across pages → canonical only as a last resort.

1.5 Authority and conversion

  • Authority: backlinks, brand mentions, real reviews, author pages with verifiable credentials.
  • Conversion: every page has a clear CTA and a contact path that matches the market.

2. Local SEO

Applies to any business with a physical location or a defined service area: clinics, restaurants, gyms, studios, home services, multi-location retail.

First-time Google Business Profile setup: see 11-channel-setup-global. This section covers optimization once the profile exists.

2.1 NAP consistency

NAP = Name, Address, Phone. These must be byte-identical everywhere:

Where it appearsCheck
Google Business ProfileThe source of truth — everything else copies from here
Website footerCharacter for character
Contact / location pagesCharacter for character
LocalBusiness schemaCharacter for character
Social profilesCharacter for character
Citations and directories (Apple Business Connect, Bing Places, Yelp, industry directories)Character for character

"Suite 200" vs "Ste. 200" vs "#200" counts as three different addresses to an aggregator. Pick one string and enforce it.

2.2 Google Business Profile optimization

ItemTargetCadence
Primary categoryThe most specific accurate category, never a generic oneReview quarterly
Secondary categories2-5 genuinely relevantQuarterly
PhotosExterior, interior, team, product3-5 per month
PostsOffers, events, updates2-4 per month
Q&ASeed and answer the questions customers actually askMonthly review
HoursUpdate holiday hours two weeks aheadPer holiday calendar
Services / menuComplete, with pricesQuarterly

2.3 Reviews

  • Aim for a steady monthly flow, not bursts. Twenty reviews in one day is an anomaly signal.
  • Ask at the moment of satisfaction, right after the service is delivered, with a short direct link.
  • Respond to 100% of reviews including five-star ones. Reply to negatives within 24 hours in a resolution tone, not a defensive one.
  • Never buy reviews. The downside (profile suspension) outweighs the upside, and AI engines increasingly read review text, not just the score.

2.4 Location pages on your site

One page per location, never all of them crammed onto the contact page:

  • URL: /locations/[city] or /[service]-[city].
  • Unique content per page: address plus embedded map, hours, location-specific phone number, real photos of that location, parking and access notes, the team at that site.
  • A dedicated LocalBusiness block per location (block 9 in references/schema-json-ld.md).
  • Do not clone one page and swap the city name. Those are thin pages and they do not rank.

3. Map keywords by funnel

FunnelKeyword typePage type
TOFUproblem, how-to, checklistblog, guide, free tool
MOFUcomparison, template, use case, "how much does X cost"comparison, use-case page, pricing
BOFUalternatives, review, demo, pricinglanding page, pricing, case study

Local businesses add a geographic layer: [service] [city], [service] near me, [service] [city] cost.

4. Optimize for AI search (GEO/AEO)

4.1 The four-rung visibility ladder

Drop the phrase "rank on ChatGPT" — nobody controls that. Report on four rungs instead, because each is earned differently:

RungMeaningWhat governs itHow you see it
1. RetrievedThe model read your content while composing an answerCrawlability, parseable structure, query relevanceServer logs; largely invisible
2. CitedYour page appears as a source in the answerContent structure, statistics, clarity, freshnessPrompt-tracking tools, AI Overview source lists
3. MentionedYour brand is named in the answer textEntity recognition plus how the web describes youPrompt-tracking tools
4. RecommendedYour product is on the shortlist the buyer considersAggregate web consensus — reviews, communities, analysts, press, videoThe framing around the mention, not just its presence

Rungs 1-3 are what on-site optimization earns you. Rung 4 mostly is not on your site at all. Reporting a single "AI visibility" number hides that gap. A rising citation count with a flat recommendation rate is a specific, diagnosable problem: the rest of the web does not corroborate you yet.

There is also a negative rung: recommended against. On requirements-heavy prompts, models increasingly name products a buyer should avoid, with sources. So track the sentiment around each mention (favorable / neutral / hedged / negative), not just the count.

4.2 The uncomfortable truth about "best [category]" listicles

Publishing your own "Top 10 [category] tools" guide and ranking yourself first is the most widely recommended AEO tactic. It does not work the way it is sold.

The model treats your guide as a source about the category. It extracts the competitor names, evaluation criteria and pricing you compiled — then makes its recommendation from web-wide consensus, where established players dominate. The common outcome: your page earns the citation while the recommendation goes to a competitor. You did the research that helped the model describe them.

It splits by market position:

  • Category leaders get both outcomes. Their guides are cited and their brand is recommended, because analysts, review sites and forums already validate them. For a leader, publishing the definitive buyer's guide is high leverage: you define the criteria the whole category is judged on.
  • Challengers should still publish genuinely useful guides, but set the expectation correctly — citation and category framing, not near-term recommendation — and shift budget toward the offsite signals that actually govern rung 4.

The test before funding another self-ranked listicle: if a model ignored everything on our domain, would the rest of the web still put us on the shortlist? If not, that gap is the priority.

4.3 robots.txt for AI crawlers

User-agent: GPTBot
Allow: /

User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: ClaudeBot
Allow: /

User-agent: anthropic-ai
Allow: /

User-agent: Google-Extended
Allow: /

User-agent: Bingbot
Allow: /

User-agent: CCBot
Disallow: /
BotOperatorRole
GPTBotOpenAICrawls for model training; obeys robots.txt. Independent of search: allowing OAI-SearchBot while disallowing GPTBot is a valid combination
OAI-SearchBotOpenAIDecides whether the site can appear in ChatGPT search answers; obeys robots.txt, changes take about 24h to apply. Opted out means not shown in those answers
ChatGPT-UserOpenAIFetches a page when a user acts in ChatGPT (e.g. opens a link); robots.txt rules may not apply. Not used to decide search inclusion
OAI-AdsBotOpenAIChecks landing pages submitted as ChatGPT ads; does not train models. Only relevant if you run those ads
PerplexityBotPerplexityPerplexity always cites sources, making it the most measurable channel
ClaudeBot / anthropic-aiAnthropicanthropic-ai is the legacy token, ClaudeBot the current one
Google-ExtendedGoogleNot a crawler. See the note below
BingbotMicrosoftFeeds both Bing and Copilot
CCBotCommon CrawlTraining-only corpus; produces no citations, safe to block if you prefer

OpenAI bot roles [checked 2026-10, developers.openai.com/api/docs/bots]. Re-check that page on the planning date: bot lists change, so do not copy user-agent strings or version numbers from memory. Confirm the bots really fetch your key pages with the crawler log check in references/ai-visibility-measurement.md.

What most guides get wrong: Google-Extended does not control whether you appear in AI Overviews. AI Overviews are built on the regular Search index. Blocking Google-Extended only affects use of your content in Google's other generative products. To control snippet usage in AI Overviews you use nosnippet, max-snippet or data-nosnippet — which also costs you ordinary snippets.

Three additional checks:

  1. Silent CDN/WAF blocking. Cloudflare and several other providers ship an "AI crawlers" toggle that is on by default. Your robots.txt can say Allow while the request never reaches origin. Check server logs or the CDN dashboard, not just robots.txt.
  2. Platform limits. Hosted site builders often restrict robots.txt editing. Confirm what is actually editable before promising it in a plan.
  3. llms.txt. Served at /llms.txt, listing your most important pages in markdown with short descriptions. It is a community-proposed convention with no formal commitment from any major platform. Optional: effect unproven, no vendor commitment. Cheap enough to ship, but do not oversell it or bill it as a major line item.

4.4 AI visibility audit table

For a fixed prompt portfolio, per-answer audit columns, metrics, AI referral traffic and a fix loop, use references/ai-visibility-measurement.md. Quick version: test 10-20 of your highest-value queries and log the results:

QuestionPlatform (ChatGPT / Perplexity / AI Overview)Are we cited?Which competitor is cited insteadAction
[query 1]Yes / No
[query 2]Yes / No

Query types to cover:

  • "What is [category]"
  • "Best [category] for [use case / segment]"
  • "Is [your brand] any good"
  • "[your brand] vs [competitor]"
  • "How much does [category] cost"
  • "[service] in [city]" (local businesses)

When a competitor is cited and you are not, check five things: is their content more extractable, do they cite statistics and sources, is it more recently updated, do they have schema you lack, and are they referenced by third parties (press, communities, review sites).

5. Content blocks AI can extract

AI systems extract passages, not pages. Every block below must stand alone without the paragraph before it. Weave these into normal content — do not write a separate "AI version" of the page, which risks scaled-content policy problems.

BlockQuery it servesStructure
Definition"what is X"Sentence 1 defines it. Sentences 2-3 expand. Final sentence says why it matters
Step-by-step"how to X"One intro line plus a numbered list, one concrete action per step
Comparison table"X vs Y"Criteria column, a "Best for" row, and a 1-2 sentence bottom line
Pros and cons"is X worth it"Pros, cons, then a conditional verdict
FAQrelated questionsQuestions phrased as users ask them; 50-100 word answers
Listicle"best X"Read section 4.2 before building these
Statistic citationevidence-seeking queriesSpecific number, source, year, link
Self-contained answerall types40-80 words, no "it", "this", or "as mentioned above"
Pricing"how much does X cost"Real range, the variables that move it, one worked example
Who it is and is not forevaluation queriesExplicitly name who should not buy — this is the differentiator
Freshness and authorshipall typesUpdated date plus a real author with stated expertise

Four formatting rules: lead each section with the answer instead of burying it; write H2/H3 the way users phrase queries; use tables over prose for comparisons; use numbered lists over paragraphs for procedures.

6. Schema and machine-readable files

PageSchema
CompanyOrganization, WebSite
BlogArticle / BlogPosting
ProductProduct + Offer (+ AggregateRating when real reviews are on the page)
SaaS / appSoftwareApplication
FAQFAQPage — see caveat
Ordered procedureHowTo — see caveat
BreadcrumbBreadcrumbList
Location pageLocalBusiness (use the specific subtype: Dentist, Restaurant, ProfessionalService...)
Event, webinarEvent
CourseCourse

Two things to state plainly to the client:

  • FAQPage: Google FAQ rich results are deprecated for all sites from 2026-05-07 (reported: Search Central notice dated 2026-05-08, doc not fetched). Search Console FAQ report removal is reported for June 2026 and API removal for August 2026. The markup is still valid schema.org and harmless: keep it when the FAQ is genuinely on the page, but promise no rich result and do not sell it as extra SERP real estate.
  • HowTo rich results were removed entirely by Google in Sept 2023. Low priority.

Full code for 12 JSON-LD blocks, testing tools and common failure modes: references/schema-json-ld.md.

Never conclude "this site has no schema" from a single fetch. Simple fetch tools do not execute JavaScript, and a large share of sites inject JSON-LD through GTM or client-side rendering. Verify with the Rich Results Test or DevTools before it goes into an audit — including when auditing competitors.

7. Backlinks and distribution

Prioritize sources that can also produce leads, not raw link counts:

  1. Product Hunt / BetaList / startup directories.
  2. G2 / Capterra / TrustRadius / Trustpilot when real customers exist.
  3. Alternative and comparison pages (see the playbook).
  4. AI tool directories when the category fits.
  5. Partner, co-marketing and industry association pages.
  6. Earned media and practitioner communities — these are what move rung 4 of the visibility ladder.

Avoid: bulk link packages, PBNs, comment spam. Beyond the penalty risk, they generate none of the web consensus AI engines actually weigh.

Output template

# SEO Growth Plan — [Brand]

## 1. Diagnosis
| Area | Status | Evidence | Priority |

## 2. Technical baseline
| Metric | Current | Threshold | Fix | Owner |

## 3. 90-day SEO strategy
- Market:
- ICP:
- Main search intent:
- Primary KPI:

## 4. Page roadmap
| Page | Funnel | Keyword intent | CTA | Owner |

## 5. Local SEO (if applicable)
| Item | Current | Action | Cadence |

## 6. AI SEO / GEO actions
| Question | Platform | Cited? | Competitor cited | Action |

Current ladder rung: [retrieved / cited / mentioned / recommended]
Mention sentiment: [favorable / neutral / hedged / negative]

## 7. Schema + technical checklist
| Page | Schema/file | Action | Validation |

## 8. Backlink/distribution plan
| Destination | Asset needed | Submission copy | Deadline |

## 9. Measurement
| Metric | Baseline | Target | Tool |

Related skills

  • 08-competitor-research-global: SEO competitors, content gaps, and the input data for comparison pages.
  • 11-channel-setup-global: first-time Google Business Profile setup.
  • 12-landing-page-brief-global: BOFU landing pages and comparison pages.
  • 13-data-analysis-global: GA4/Search Console/export analysis.
  • 15-social-listening-global: real community questions for FAQ and TOFU content.
  • 36-content-brief-global: turn the page roadmap into writing briefs.

Quality checklist

  • Do not recommend content before clarifying search intent.
  • Every page has a CTA and a funnel role.
  • Technical findings use specific numbers, never "improve page speed".
  • Local SEO: NAP reconciled against Google Business Profile before proposing edits.
  • Schema matches visible page content; no aggregateRating on pages without reviews.
  • Any schema conclusion (yours or a competitor's) verified with a JavaScript-rendering tool.
  • AI SEO claims stay honest: report the four-rung ladder with sentiment, never promise rankings.
  • If a self-ranked listicle is recommended, expectations are stated by market position.
  • Directory submission points to a useful destination page.
  • Plan includes baseline and KPI.

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.