Use this skill when performing keyword research, search intent analysis, keyword clustering, SERP analysis, competitor keyword gaps, long-tail keyword discovery, or evaluating keywords for snippet opportunity, AI Overview presence, and tri-surface keyword reports. Covers organic (SEO), answer engine (AEO snippets/PAA), and AI citation (GEO AI Overviews/ChatGPT Search/Perplexity) surfaces.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Une commande directe contourne le prompt de vérification. Examinez la source avant de l'exécuter.
Use this skill when performing keyword research, search intent analysis, keyword clustering, SERP analysis, competitor keyword gaps, long-tail keyword discovery, or evaluating keywords for snippet opportunity, AI Overview presence, and tri-surface keyword reports. Covers organic (SEO), answer engine (AEO snippets/PAA), and AI citation (GEO AI Overviews/ChatGPT Search/Perplexity) surfaces.
When this skill is activated, always start your first response with the 🧢 emoji.
Keyword Research
Quick start (5 steps): Seed keywords > Expand with tools/autocomplete >
Classify intent for each keyword > Tri-score (organic + AEO + GEO) >
Cluster by topic with surface annotations. See the common tasks below for details.
Keyword research is the foundation of all organic and AI search strategy. It is the
process of discovering what words and phrases people type into search engines and AI
assistants, understanding why they search (intent), and evaluating which of three
surfaces - organic results, answer engine features, or AI-generated citations -
offers the best opportunity for each keyword.
In 2026, keyword research must account for three surfaces simultaneously:
Organic blue links (SEO) - traditional rankings on Google, Bing, etc.
Answer engine features (AEO) - featured snippets, People Also Ask, voice results
AI-generated citations (GEO) - Google AI Overviews, ChatGPT Search, Perplexity
This skill covers the full research workflow - from seed topic to prioritized,
tri-surface-scored keyword report. It tells you WHAT to target and WHERE the
opportunity is. For HOW to optimize content once you have chosen your targets,
use the companion skills aeo-optimization (snippet/PAA formatting) and
geo-optimization (AI citation optimization).
When to use this skill
Trigger this skill when the user:
Wants to find keywords for a new website, product page, or blog
Asks to analyze search intent for a keyword list
Needs to group keywords into topic clusters or content pillars
Wants to discover competitor keyword gaps or ranking opportunities
Asks to find long-tail variations of a seed keyword
Needs to prioritize a list of keywords by opportunity or difficulty
Wants to understand what SERP features appear for a target keyword
Asks to detect keyword cannibalization across existing pages
Wants to evaluate keywords for snippet opportunity or featured snippet potential
Asks to assess AI Overview presence for target keywords
Needs a tri-surface keyword report scoring organic, AEO, and GEO opportunity
Wants to understand which keywords AI search engines answer vs. defer
Do NOT trigger this skill for:
Paid search (PPC/Google Ads) bid strategy - ad-specific match types, Quality Scores,
and CPC optimization are a different domain
Brand naming or tagline development - that is copywriting, not search research
Formatting content to win snippets or PAA - that is aeo-optimization
Making content more citable by AI engines - that is geo-optimization
Key principles
Search intent is more important than volume - A keyword with 500 monthly searches
and clear transactional intent will drive more revenue than a 50,000-search keyword
that is purely informational. Always qualify intent before qualifying volume.
Cluster keywords by topic, not individual pages - One page should own a cluster
of semantically related terms. Building one page per keyword creates duplication,
splits authority, and fragments the user experience.
The SERP + AI Overview is the source of truth - No tool tells you more about what
Google wants to rank than the current top 10 results and whether an AI Overview fires.
Content type, length, format, featured snippet presence, and AI Overview citations
all reveal the implicit standard for a keyword.
Long-tail keywords convert better - Longer, more specific queries have lower
volume but higher purchase intent and lower competition. A content strategy built
on long-tail clusters outperforms chasing high-volume head terms in most niches.
Competitor gaps reveal the fastest wins - Finding keywords where competitors
rank in positions 4-15 (or not at all) is faster than trying to beat them on
keywords where they dominate. Gaps are the entry points.
Every keyword has three surfaces to evaluate - A keyword that looks mediocre
for organic ranking may have excellent snippet opportunity (AEO) or strong AI
citation potential (GEO). Evaluating all three surfaces prevents blind spots and
reveals non-obvious wins that single-surface research misses.
Research and optimization are separate phases - This skill identifies WHAT to
target and WHERE the opportunity is. Optimization (HOW to format for snippets,
HOW to boost AI citations) is a downstream activity handled by aeo-optimization
and geo-optimization. Do not mix phases - complete research before starting
optimization.
Core concepts
Search intent taxonomy classifies every keyword into one of four categories based
on what the searcher is trying to accomplish. Informational intent ("how does X work",
"what is Y") signals content and education needs. Navigational intent ("brand name",
"site login") signals the user knows where they want to go. Transactional intent
("buy X online", "X pricing", "X discount code") signals readiness to act.
Commercial investigation ("best X", "X vs Y", "X review") sits between informational
and transactional - the user is evaluating options before deciding. See
references/search-intent-mapping.md for detailed classification guidance including
intent-to-surface mapping.
Keyword difficulty (KD) is a 0-100 score estimating how hard it is to rank on
page one for a keyword, based primarily on the backlink authority of the current
top-ranking pages. High difficulty does not mean impossible - it means you need
more authority, better content, or a more specific angle to win. Treat KD as a
relative filter, not an absolute gate.
Search volume vs. traffic potential are related but different. Search volume is
the average monthly searches for one keyword. Traffic potential is the estimated
traffic the top-ranking page receives for the entire cluster of keywords it ranks
for. A keyword with 200 monthly searches may have traffic potential of 2,000 if
the ranking page captures dozens of related terms. Always evaluate traffic potential
over raw volume.
Keyword cannibalization occurs when two or more pages on the same site compete
for the same keyword, splitting ranking signals and confusing Google about which
page to surface. Symptoms include ranking oscillation, positions that drop when
publishing new content, and two pages from the same domain appearing for the same
query. Resolve by merging, redirecting, or clearly differentiating the pages.
Tri-surface keyword scoring evaluates every keyword across three surfaces:
organic opportunity (0-10), AEO opportunity (0-10), and GEO opportunity (0-10).
The composite score (0-30) reveals total opportunity, and the priority surface
tells you where to focus optimization efforts. See references/tri-surface-scoring.md
for the full scoring rubrics and worked examples.
SERP feature landscape - SERP features (featured snippets, PAA, video carousels,
image packs, shopping results, AI Overviews) are research signals, not just ranking
features. Their presence or absence tells you which surfaces are active for a keyword.
A keyword with a featured snippet and an AI Overview has tri-surface potential. A
keyword with only shopping results is organic-only. Map features during research to
inform scoring.
AI search query patterns - AI search engines (Google AI Overviews, ChatGPT Search,
Perplexity) do not answer every query. They tend to fire on informational, comparison,
and multi-step queries while skipping navigational, transactional, and simple factual
queries. Understanding these patterns helps predict GEO opportunity during research.
See references/search-intent-mapping.md for the full AI Overview intent pattern table.
Common tasks
1. Map search intent for a keyword list
For each keyword, classify it using the four-type taxonomy. Apply this decision order:
Check modifiers first - Words like "buy", "order", "coupon", "discount" signal
transactional. Words like "best", "top", "review", "vs", "alternative" signal
commercial investigation. Words like "how", "what", "why", "guide", "tutorial"
signal informational. Brand name only = navigational.
When modifiers are absent, check the SERP - Look at the top 3 results.
Are they product pages, comparison articles, definitions, or brand homepages?
The content type Google rewards reveals the intent.
Assign a primary intent and note a secondary if relevant - Many keywords blend
types. "Best project management software" is primarily commercial investigation
with transactional secondary (the user may click through to pricing).
Output format: a table with columns
keyword | intent | confidence | content_type | snippet_type | ai_overview_present.
The last two columns feed into AEO and GEO scoring. Record snippet type as
paragraph/list/table/none and AI Overview presence as yes/sometimes/no.
See references/search-intent-mapping.md for the full classification guide.
2. Build a keyword cluster from a seed topic
Start with one seed keyword and expand outward:
Generate variants - Use a keyword tool to pull: questions (People Also Ask),
autocomplete suggestions, related searches, and lexical variants. For the seed
"project management software", variants include "best project management tools",
"project management app for teams", "free project management software", etc.
Group by SERP overlap - Keywords that return the same top-ranking URLs belong
in the same cluster. If "project management software" and "task management tool"
return 6 of the same top-10 results, one page can rank for both.
Identify the primary keyword - The one with the highest traffic potential
becomes the primary term (used in title, H1, URL). All others are secondary terms
woven into subheadings and body copy.
Name the cluster - Give it a descriptive label: "project management software -
top-of-funnel commercial". This label drives content brief decisions.
Annotate with dominant surface - Score the cluster's keywords using tri-surface
scoring and assign a surface label: [ORG], [AEO], [GEO], or [AEO+GEO]. This tells
the content team which optimization approach to take after writing.
See references/keyword-clustering.md for semantic, SERP-based, modifier-based,
and surface-aware clustering methods.
3. Tri-surface scoring
This is the signature task of modern keyword research. For each keyword (or cluster),
produce scores across all three surfaces.
Process:
Gather raw data - For each keyword, collect: search volume, traffic potential,
KD, SERP features present, featured snippet format and holder, PAA count, and
AI Overview presence (check Google, optionally ChatGPT Search and Perplexity).
See references/tool-specific-workflows.md for tool-by-tool data collection steps.
Score organic (0-10) - Based on traffic potential, KD inversion, intent-business
alignment, and content gap existence.
Score AEO (0-10) - Based on snippet presence, format match, PAA count, voice
search likelihood, and current holder strength. Score 0 for navigational and
most transactional keywords where snippets don't fire.
Score GEO (0-10) - Based on AI Overview trigger, citation density, query type
match, entity relevance, and content uniqueness. Score 0 when no AI Overview fires.
Calculate composite - Sum the three scores (0-30). Apply business-goal weighting
if needed.
Assign priority surface - Highest-scoring surface becomes the priority. If two
are within 2 points and both above 5, mark as dual-surface opportunity.
See references/tri-surface-scoring.md for the complete rubrics, weighting tables,
and three worked examples.
A keyword gap is a keyword where a competitor ranks in the top 20 but you do not.
Framework for surface-aware gap analysis:
Pull keyword rankings for 3-5 competitors using a tool (Ahrefs, Semrush).
Export keywords where competitor is in positions 1-20.
Filter out keywords where your site already ranks positions 1-5 (already winning).
Filter for keywords matching your target intent.
Sort by traffic potential descending.
New: Surface gap analysis - For the top 50 gap keywords, check whether they
trigger featured snippets and AI Overviews. A keyword gap where the competitor
ranks organically but doesn't hold the snippet or isn't cited by AI is a
multi-surface gap - you can potentially win the snippet or AI citation even if
the organic position takes time to capture.
Cross-reference with your content inventory. Existing pages that can be optimized
are quick wins; missing pages are content creation opportunities.
See references/tool-specific-workflows.md for tool-specific gap analysis steps.
5. Find long-tail variations with surface annotations
Long-tail keywords (typically 3+ words, lower volume, higher specificity) are easier
to rank for and often signal stronger intent. To find them:
Question-format long-tails ("how to X for Y") score high on AEO (snippet targets)
Comparison long-tails ("X vs Y for Z") score high on GEO (AI engines love comparisons)
Feature/qualifier long-tails ("X with API for enterprise") are usually organic-only
Location long-tails are almost always organic-only (local pack, not snippets or AI)
Target long-tail keywords with dedicated FAQ sections, comparison pages, or use-case
landing pages. Annotate each with its likely priority surface.
6. Produce a keyword research report
The keyword research report is the primary deliverable of this skill. It synthesizes
all research into an actionable document.
Report structure:
Executive summary (1 paragraph) - Total keywords analyzed, top clusters,
dominant surface opportunity, and 3 biggest findings.
Intent distribution - Pie chart or table showing the breakdown of informational,
commercial, transactional, and navigational keywords in the dataset.
Surface opportunity map - Table showing how many keywords have their priority
surface as organic, AEO, GEO, dual, or tri. This reveals whether the overall
strategy should lean toward traditional SEO, snippet optimization, or AI citation.
Top keyword clusters - For each cluster: name, primary keyword, keyword count,
average composite score, dominant surface, and recommended content type.
Quick wins - Keywords where you have the best ratio of opportunity to effort:
striking-distance organic keywords (positions 5-15), snippet-eligible keywords with
no current holder, and AI Overview queries citing weak sources you can displace.
Content recommendations - Map clusters to content calendar items: new pages,
existing pages to optimize, FAQ additions, and comparison articles to create.
Full keyword data - Appendix table with all keywords and their scores. Use
the spreadsheet template from references/tool-specific-workflows.md.
7. Detect keyword cannibalization
Run a site search for the target keyword (site:yourdomain.com "keyword phrase") and
audit Google Search Console for pages sharing the same top query.
Diagnosis:
Two pages ranking for the same query: Check which page Google prefers (higher
avg. position). The preferred page keeps the keyword; the other page is re-optimized
for a different term or redirected.
Rankings oscillating week to week: Classic cannibalization signal. Consolidate
the weaker page's content into the stronger one via a 301 redirect.
New page tanked the ranking of an existing page: Re-differentiate the new page's
focus term or merge it back into the original.
Snippet cannibalization: If two of your pages alternate holding the featured
snippet for the same query, Google may drop both. Consolidate to ensure one
definitive page owns the snippet-eligible answer.
Anti-patterns
Mistake
Why it's wrong
What to do instead
Chasing volume over intent
A high-volume keyword that doesn't match your buyer's stage sends irrelevant traffic that bounces
Filter by intent first, then sort by volume within the right intent category
One page per keyword
Creates thin, near-duplicate pages that split link equity and rarely rank
Cluster semantically related keywords to one page; build depth
Ignoring the SERP
Targeting a keyword without checking what type of content currently ranks leads to mismatched format
Always check the top 10 before writing a brief; match dominant content type
Targeting KD 70+ with a new site
New domains lack the authority to rank on competitive terms
Start with KD < 30 to earn rankings, traffic, and links; build up to harder terms
Skipping competitor gap analysis
Building content only from brainstorming misses proven opportunities
Always run a gap report before finalizing your content calendar
Never updating keyword research
Search behavior evolves; queries from 2 years ago may have shifted in intent or volume
Audit top content annually; refresh keyword targets based on current SERP data
Ignoring snippet and AI Overview presence
Treating all keywords as organic-only misses AEO and GEO opportunities that may be easier to win than organic rankings
Record SERP features and AI Overview status for every keyword during research
Treating every keyword as organic-only
Defaulting to traditional SEO without checking if a keyword is better won through snippets or AI citations
Run tri-surface scoring for all priority keywords; assign a priority surface
Scoring GEO when no AI Overview fires
Assigning GEO opportunity to a keyword where AI engines don't generate an answer wastes effort
Always verify AI Overview presence manually; GEO = 0 if no AI answer fires
Mixing research and optimization
Trying to format content for snippets or AI citations before finishing keyword research leads to premature decisions
Complete the full research workflow (seed > expand > classify > score > cluster) before any optimization work
Gotchas
Traffic potential and search volume diverge most on commercial keywords - A keyword with 300 monthly searches may be the primary term for a page that also ranks for 40 related variants, giving it 8,000 monthly visits. Conversely, a 10,000-volume head term may have traffic potential of only 6,000 because ranking page 1 earns only a small CTR share. Always pull traffic potential from your tool (Ahrefs "TP", Semrush "Traffic"), not raw volume.
SERP feature presence changes by location, device, and login state - An AI Overview you see in a logged-in US Chrome session may not fire in an incognito session or in another country. Always verify SERP features in incognito mode and, where relevant, from the target country using a VPN or tool like SERP API, before assigning AEO or GEO scores.
Keyword cannibalization diagnosis is wrong if you use position average - Google Search Console averages positions across all queries and dates. Two pages fighting for the same query may each show position 8 in the average, but the reality is one shows at 3 and the other at 15, alternating week by week. Filter GSC by specific queries and look for multiple pages appearing or for high impression/low click patterns on the same query across different pages.
Tri-surface scoring becomes meaningless if you score GEO for navigational queries - Navigational queries ("brand login", "product dashboard") almost never trigger AI Overviews. Assigning any GEO score above zero to navigational keywords inflates composite scores and misdirects content effort. GEO score must be zero unless you have verified an AI Overview fires for that query.
Keyword clusters built from tool "related keywords" lists ignore SERP overlap - Tool-suggested related keywords group by semantic similarity, not by whether Google actually returns the same URLs for both queries. Two semantically similar keywords may trigger completely different SERPs (different content types, different competition). Validate cluster membership by checking that 5+ of the top 10 results overlap between the keywords.
References
For detailed content on specific topics, read the relevant file from references/:
references/search-intent-mapping.md - Deep dive into the four intent types,
classification signals, intent-to-content-type matrix, intent-to-surface mapping
for AEO and GEO, AI Overview intent patterns, and how to validate intent
assumptions from SERP data. Load when classifying a keyword list or writing a
content brief.
references/keyword-clustering.md - Methods for clustering keywords (semantic,
SERP-based, modifier-based, surface-aware), building pillar-and-spoke topic
clusters, surface-aware cluster annotation, and tooling options. Load when building
a cluster or planning a content architecture.
references/tri-surface-scoring.md - Complete scoring rubrics for organic (0-10),
AEO (0-10), and GEO (0-10) opportunity scores, composite scoring, business-goal
weighting, three worked examples, and limitations. Load when scoring keywords
or producing a tri-surface keyword research report.
references/tool-specific-workflows.md - Step-by-step workflows for Ahrefs,
Semrush, Google Search Console + free tools, and ChatGPT/Perplexity manual audits.
Includes spreadsheet template with columns, formulas, and conditional formatting.
Load when performing hands-on keyword research with specific tools.
Only load a references file if the current task requires deep detail on that topic.
Companion check
On first activation of this skill in a conversation: check which companion skills are installed by running ls ~/.claude/skills/ ~/.agent/skills/ ~/.agents/skills/ .claude/skills/ .agent/skills/ .agents/skills/ 2>/dev/null. Compare the results against the recommended_skills field in this file's frontmatter. For any that are missing, mention them once and offer to install:
