Keyword Clustering

Keyword clustering is the process of grouping related keywords so that a single page can rank for all of them, rather than creating separate thin pages for each term. It is the bridge between raw keyword research and a content architecture. Without clustering, a site accumulates redundant, competing pages. With clustering, every page is built to own a topic - ranking for dozens or hundreds of related terms while building concentrated topical authority.

---

Why clustering matters

Search engines rank pages, not keywords. A well-optimized page targeting a primary keyword will naturally rank for dozens of semantically related variants. Building separate pages for "project management software" and "project management tools" and "project management app" splits the backlinks, internal links, and topical signals that should all feed into one authoritative page.

Benefits of proper clustering:

• Reduces the total number of pages needed (quality over quantity)
• Consolidates link equity to fewer, stronger pages
• Prevents cannibalization before it starts
• Maps clearly to a content calendar (one cluster = one content item)
• Creates a hierarchy that supports pillar-and-spoke internal linking

---

Clustering methods

Method 1: SERP-based clustering (most accurate)

Group keywords based on which URLs appear in their top search results. If two keywords return the same ranking URLs, Google considers them the same topic.

How to do it:

1. For each keyword in your list, record the top 5-10 ranking URLs.
2. Compare URL overlap across keywords. Use a clustering tool (Keyword Insights, Surfer SEO, KeyClusters) or do it manually for smaller lists.
3. Keywords sharing 3+ of the same top-5 URLs belong in the same cluster.
4. Keywords with 0-1 URL overlap belong in different clusters (or need their own page).

Threshold guidance:

• Same 3+ of top 5 URLs: same cluster (high confidence)
• Same 2 of top 5 URLs: likely same cluster (verify manually)
• Same 1 of top 5 URLs: borderline (check if topically related)
• 0 shared URLs: separate clusters

Advantages: Most accurate method because it reflects actual Google behavior, not just keyword similarity. Catches non-obvious clusters where synonyms rank together.

Disadvantages: Time-intensive at scale without tooling; requires live SERP data.

---

Method 2: Semantic/lexical clustering

Group keywords by shared words, stems, and synonyms using linguistic similarity. Faster than SERP-based clustering, appropriate for early-stage research or lists where you can not pull SERP data.

Approaches:

Exact match grouping - Keywords containing the same root phrase go together:

• "email marketing" + "email marketing tools" + "email marketing tips" = one cluster

Modifier stripping - Remove modifiers and group by the core concept:

• "best CRM for small business" + "top CRM tools" + "CRM software for startups" all strip to "CRM" -> same cluster

Synonym mapping - Identify synonyms and near-synonyms that mean the same thing:

• "project management software" / "task management tool" / "work management platform" are often synonymous; verify with SERP overlap

Advantages: Fast, works without tool access, good for brainstorming.

Disadvantages: Less accurate than SERP-based. Can over-cluster (merging keywords that Google treats as different) or under-cluster (splitting keywords that Google treats the same).

---

Method 3: Modifier-based clustering

Group keywords by the type of modifier that qualifies the seed keyword. Useful for mapping the content types needed across a topic.

Modifier categories:

Modifier type	Examples	Content type
Question modifiers	"how to X", "what is X", "why X"	How-to guide, explainer
Comparison modifiers	"X vs Y", "X alternative", "X compared to Y"	Comparison/versus page
Audience modifiers	"X for small business", "X for developers"	Use-case landing page
Feature modifiers	"X with API", "X free plan", "X enterprise"	Feature/plan pages
Location modifiers	"X in NYC", "X near me"	Local landing page
Stage modifiers	"X tutorial", "X examples", "X best practices"	Educational content

Apply this method after semantic clustering to identify which clusters need sub-pages (a large cluster with many modifier types may split into a pillar + spoke pages).

---

Building pillar-and-spoke topic clusters

The pillar-and-spoke model is the standard architecture for topical authority.

Pillar page:

• Covers the broadest, highest-volume keyword in the cluster
• Provides a comprehensive overview of the topic
• Links out to each spoke page for deep dives
• Typically 2,000-5,000 words; comprehensive but not exhaustive
• Targets commercial investigation or informational intent at the category level
• Example: "Email Marketing: The Complete Guide"

Spoke pages:

• Each covers one sub-topic of the pillar in depth
• Targets a more specific, often long-tail keyword cluster
• Links back to the pillar page
• Typically 1,000-3,000 words; thorough on the specific sub-topic
• Examples: "Email Marketing Automation", "Email Marketing A/B Testing", "Email Marketing for E-commerce"

Hub page (optional):

• A topical hub is a navigation-oriented page listing pillar pages in a domain
• Useful for large sites covering multiple pillars under one umbrella topic
• Example: a "Marketing Resources" hub linking to pillars on Email, SEO, Social

Internal linking rules:

• Every spoke links to its pillar (reinforces pillar authority)
• Pillar links to all spokes (signals topical depth)
• Spokes can link to related spokes within the same pillar
• Never link from a pillar to a page on a different pillar without purpose

---

Cluster size guidelines

Cluster size	What it means	Action
1-3 keywords	Small, specific cluster	One targeted page; may be a spoke
4-15 keywords	Standard cluster	One well-optimized page targeting all terms
16-40 keywords	Large cluster	Consider splitting into pillar + 2-3 spokes
40+ keywords	Very large cluster	Definitely needs pillar + spoke architecture

Signs you are over-clustering (merged too much):

• Your page would need to cover 5+ distinct sub-topics to address all keywords
• The keywords span multiple intent types (informational + transactional in one cluster)
• The top SERP results for different keywords in the cluster are completely different pages

Signs you are under-clustering (split too much):

• You have two planned pages where the top SERP results are 70%+ the same URLs
• Your pages would be shorter than 600 words to cover the "separate" topics
• Your planned pages are semantic synonyms of each other

---

Manual clustering process (no tooling)

For lists under 100 keywords, manual clustering is practical:

1. Export to a spreadsheet - One keyword per row, with search volume and intent.
2. Sort by root word - Alphabetical sort often groups related keywords together.
3. Create a "cluster" column - Assign a cluster name to each keyword.
4. Merge by SERP spot-check - For any two clusters you are unsure about, search both keywords and compare the top 3 results. Same URLs? Merge the clusters.
5. Name each cluster - Use the highest-volume keyword as the cluster name.
6. Count keywords per cluster - Any cluster with 15+ keywords may need a pillar/spoke split.
7. Assign content type - Based on intent and cluster size, assign: new page, existing page to optimize, FAQ addition, or pillar + spokes.

---

Tooling options

Tool	Method	Best for
Keyword Insights	SERP-based + NLP	Large lists (1,000+); automated clustering
Surfer SEO	SERP-based	Clusters tied to content editor workflow
KeyClusters	SERP-based	Standalone clustering at low cost
Ahrefs / Semrush	Semantic (built-in grouping)	Quick grouping during research; less precise
Screaming Frog + custom script	Custom SERP scrape	Technical teams building their own workflows
Manual spreadsheet	Semantic + modifier	Lists under 100 keywords

No-tool fallback: For any cluster where you are unsure, search both keywords in the same incognito browser and compare the top 5 results. If 3 or more results match, they are the same cluster.

---

Cluster validation checklist

Before finalizing a cluster and briefing a writer:

• Primary keyword identified (highest traffic potential in the cluster)
• All secondary keywords can be naturally addressed in one piece of content
• SERP-verified: top results for primary keyword match the content type you plan
• No existing page on your site already owns this cluster (check for cannibalization)
• Intent is consistent across all keywords in the cluster
• Cluster size is appropriate (4-15 keywords for a standard page; larger = pillar)
• Internal links planned: this page links to its pillar; pillar links back

A validated cluster becomes a one-to-one mapping to a content brief. One cluster = one content item = one calendar slot. This is the link between keyword research and content execution.

---

Surface-aware clustering

In tri-surface keyword research, clusters should be annotated with their dominant surface opportunity. This annotation tells content creators not just what to write about but which surface to optimize for first.

How to annotate clusters with a dominant surface

After building your clusters using any of the methods above, add a surface annotation step:

1. Score each keyword in the cluster using the tri-surface scoring rubric from references/tri-surface-scoring.md. You need organic, AEO, and GEO scores per keyword.

2. Average the scores across the cluster to get cluster-level surface scores:

   • Cluster organic score = average of all keywords' organic scores
   • Cluster AEO score = average of all keywords' AEO scores
   • Cluster GEO score = average of all keywords' GEO scores

3. Assign the dominant surface based on the highest cluster-level score. If two scores are within 1 point, mark it as a dual-surface cluster.

4. Add the surface label to the cluster name:

   • "email marketing automation [ORG]" - organic-dominant
   • "how to improve email deliverability [AEO]" - AEO-dominant (snippet opportunity)
   • "best email marketing tools for startups [GEO]" - GEO-dominant (AI citation opportunity)
   • "email marketing vs SMS marketing [AEO+GEO]" - dual-surface

Surface patterns by cluster type

Certain cluster shapes naturally align with specific surfaces:

Cluster type	Dominant surface	Why
Question clusters ("how to X", "what is X", "why X")	AEO	Question queries trigger featured snippets and PAA boxes at high rates. Voice search also favors question-format content.
Comparison clusters ("X vs Y", "best X for Y", "X alternatives")	GEO	AI engines produce detailed comparison answers with multiple citations. These are the highest-value GEO clusters.
Data-rich clusters ("X statistics", "X benchmarks", "X report")	GEO	AI engines cite sources with original data and statistics. If your content has proprietary data, GEO opportunity is high.
Transactional clusters ("buy X", "X pricing", "X free trial")	Organic	Neither snippets nor AI Overviews fire for purchase-intent queries. Pure organic play.
Tutorial clusters ("X tutorial", "X step by step", "X for beginners")	AEO + Organic	Step-by-step content wins list snippets and ranks well organically. Some GEO opportunity if the tutorial is comprehensive.
Definition clusters ("what is X", "X definition", "X explained")	AEO	Paragraph snippets fire for definition queries. AI engines answer these from training data with fewer citations, so GEO is lower.

Impact on content brief

The dominant surface annotation changes the content brief requirements:

• Organic-dominant clusters: standard SEO brief - focus on depth, internal linking, keyword placement, and matching the SERP content type
• AEO-dominant clusters: brief must include snippet-formatted answer blocks, PAA subheadings, concise answer paragraphs (40-60 words), and structured lists/tables. Reference the aeo-optimization skill for formatting details.
• GEO-dominant clusters: brief must include citable claims with supporting data, clear expert positioning, unique analysis or original research, and structured content that AI engines can extract. Reference the geo-optimization skill for citation optimization details.
• Dual/tri-surface clusters: layer the requirements - lead with the primary surface format, then address secondary surface needs in supporting sections

Search Intent Mapping

Search intent (also called user intent or query intent) is the underlying goal a person has when they type a query into a search engine. Google's core mission is to match results to intent, which means intent is the single most important dimension of any keyword. A technically well-optimized page targeting the wrong intent will not rank because it does not satisfy what the searcher needs.

---

The four intent types

Informational

The searcher wants to learn, understand, or find an answer. No immediate action is implied. These are the highest volume queries on the web.

Signals:

• Question words: "how", "what", "why", "when", "who", "where"
• Explanatory phrases: "guide to", "tutorial", "explained", "definition of", "examples of"
• Research phrases: "statistics on", "history of", "overview of"

Examples:

• "how does compound interest work"
• "what is a product-led growth strategy"
• "why is my React app re-rendering"
• "JavaScript closures explained"

Content types Google rewards:

• Long-form guides and tutorials
• How-to articles with step-by-step structure
• Definitions and explainers
• FAQ pages
• Wikipedia-style reference articles

Conversion potential: Low to medium. Informational content builds awareness and trust, feeds remarketing audiences, and earns backlinks. It rarely converts directly but is essential for top-of-funnel and authority building.

---

Navigational

The searcher already knows where they want to go and is using the search engine as a shortcut to get there. They have a specific destination in mind.

Signals:

• Brand names: "Stripe dashboard", "Notion login", "Figma"
• Site-specific phrases: "GitHub repo for X", "Ahrefs keyword explorer"
• Destination phrases: "sign in", "login", "account", "download"

Examples:

• "Notion templates"
• "Vercel dashboard login"
• "Tailwind CSS docs"
• "react-query GitHub"

Content types Google rewards:

• Official brand homepages
• Login and account pages
• Official documentation sites
• Official download pages

Conversion potential: Very high for your own brand (the searcher is already a user). Very low for competitors' brand terms - do not waste resources trying to rank for "Salesforce login" unless you are Salesforce.

Key rule: Do not build content targeting competitors' pure brand navigational terms. Comparison pages ("X vs Salesforce") work; pure navigational terms do not.

---

Transactional

The searcher is ready to take a specific action, often a purchase, signup, or download. Intent to convert is explicit.

Signals:

• Purchase intent: "buy", "order", "purchase", "shop for"
• Conversion intent: "sign up", "get started", "free trial", "download"
• Deal signals: "coupon", "discount", "promo code", "deal", "cheap"
• Subscription signals: "pricing", "plans", "subscription"

Examples:

• "buy mechanical keyboard online"
• "Notion pricing"
• "free project management software"
• "HubSpot free trial"
• "Figma discount code"

Content types Google rewards:

• Product pages with clear CTAs
• Pricing pages
• Free trial / signup landing pages
• Category/collection pages (e-commerce)
• Service pages with contact forms

Conversion potential: Highest of all four types. These searchers have money or action intent. Optimize for clarity, trust signals, and minimal friction.

---

Commercial Investigation

The searcher is evaluating options before committing. They know the category or problem but are researching which solution to choose. This is the "consideration" stage of the funnel.

Signals:

• Comparison: "vs", "versus", "or", "compared to", "difference between"
• Evaluation: "best", "top", "review", "reviews", "rated", "recommended"
• Alternatives: "alternative to", "instead of", "like X but"
• Use-case qualifying: "for small business", "for developers", "for teams"

Examples:

• "best CRM for startups"
• "Ahrefs vs Semrush"
• "Notion alternatives for project management"
• "Stripe reviews"
• "top email marketing tools 2024"

Content types Google rewards:

• "Best X" listicles and comparison guides
• Side-by-side comparison tables
• Review articles with pros/cons
• Use-case landing pages
• Buyer's guides

Conversion potential: High. The searcher is actively deciding. Content here is some of the most valuable in the funnel - it captures buyers at decision time and can include affiliate links, CTAs to trials, or soft conversion offers.

---

How to classify intent from SERP features

When modifier words are absent or ambiguous, the SERP itself is the ground truth. Analyze the top 3-5 results:

SERP signal	Likely intent
All results are product/service pages	Transactional
All results are "best X" or comparison articles	Commercial investigation
All results are how-to guides, tutorials, or explainers	Informational
Top result is the brand's homepage or login page	Navigational
Mix of product pages and review articles	Transactional + commercial investigation
Mix of guides and product pages	Informational + transactional (blended)
Featured snippet with a definition	Informational
Featured snippet with a step list	Informational (task-oriented)
Shopping carousel at the top	Transactional (strong product signal)
Local pack (map + 3 businesses)	Transactional with local modifier

SERP features as secondary signals:

• People Also Ask (PAA) - PAA questions are almost always informational sub-intents within the main query. They are subheading opportunities regardless of the primary intent.
• Video carousel - Strong signal that the audience prefers video. Consider a companion video or a YouTube-embedded section in your article.
• Image pack - Visual intent. Include strong imagery and optimize alt text.
• Knowledge panel - Entity/brand query. Content strategy here is brand PR, not SEO.
• Sitelinks - Navigational signal; the brand is dominant for this query.

---

Intent-to-content-type mapping matrix

Intent	Primary format	Page type	CTA type
Informational	Long-form guide, how-to, explainer	Blog post, wiki, resource hub	Newsletter signup, content upgrade, internal links
Navigational	Brand page, login, docs	Homepage, login page, docs site	None (user already committed)
Transactional	Product/service page, category page	Product page, pricing page, signup landing	Buy now, start free trial, add to cart
Commercial investigation	Listicle, comparison, review	"Best X" article, versus page, buyer's guide	Soft CTA: "see pricing", "compare plans", affiliate link

Blended intent pages - Some keywords require satisfying two intents in one page. "Email marketing pricing" is both transactional (show me plans) and commercial investigation (how does it compare?). In these cases, lead with the primary intent in the hero section and address the secondary intent in a supporting section below.

---

Validating intent assumptions

Assumptions made from keyword modifiers are sometimes wrong. Always validate against the live SERP before committing to a content format.

Common false assumptions:

• "Email marketing" sounds informational but the SERP is dominated by tool homepages - it has a strong navigational/commercial intent overlay.
• "Notion" sounds navigational (brand name) but also returns templates, tutorials, and feature comparison content - there is informational intent in the mix.
• "Best practices for X" sounds informational but for some categories Google favors product comparison pages because searchers want tool recommendations.

Validation checklist:

1. Search the keyword in an incognito window (or use a rank tracker with SERP preview).
2. Note the content type of results 1, 2, and 3 (article, product page, video, etc.).
3. Note the average content length (use a word count tool on 3-5 top results).
4. Check whether a featured snippet is present and what format it uses (paragraph, list, table).
5. Note any special SERP features (PAA, video, shopping, local pack).
6. Update your intent classification and content format before writing the brief.

Intent mismatch consequences:

• Writing a how-to guide for a keyword where Google ranks product pages = won't rank
• Writing a product page for a keyword where Google ranks comparison articles = won't rank
• Writing a 500-word definition for a keyword where the top results average 3,000 words = won't rank

Always let the SERP tell you the format. Your job is to produce the best version of what is already working, not to invent a new format.

---

Assigning intent to a full keyword list

When you have 50-500 keywords to classify, use this workflow:

1. Auto-classify by modifier scan - Write a simple rule: if the keyword contains any of [buy, order, purchase, pricing, discount, coupon, free trial] -> transactional. If it contains [best, vs, review, alternative, compared] -> commercial investigation. If it contains [how, what, why, when, guide, tutorial, explained] -> informational. If it contains [brand name, login, sign in, download] -> navigational.

2. Flag ambiguous keywords - Any keyword that fires zero rules or fires multiple conflicting rules needs manual SERP review. These are typically 10-20% of a list.

3. Batch-verify the auto-classified results - Spot-check 5-10 keywords from each intent category against the live SERP. If your auto-classification is wrong more than 20% of the time, refine your modifier lists.

4. Output a classification table with columns: keyword | primary_intent | secondary_intent | content_type | priority

This classification table becomes the input to your content calendar and content brief creation process.

---

Intent signals for AEO and GEO surfaces

Traditional intent classification maps keywords to content types for organic search. In tri-surface keyword research, intent also determines which non-organic surfaces are viable for a keyword. Not every intent type works on every surface.

Intent-to-surface mapping

Intent type	Organic	AEO (snippets/PAA/voice)	GEO (AI citations)	Notes
Informational	High	High	High	Best tri-surface opportunity. AI engines answer informational queries extensively and cite sources. Snippets fire frequently.
Commercial investigation	High	Medium	High	AI engines produce detailed comparison answers with citations. Snippets are less common (Google prefers showing multiple results) but PAA is active.
Transactional	High	Low	Low	Snippets rarely fire for "buy X" queries. AI Overviews usually don't appear for direct purchase intent - Google shows shopping results instead.
Navigational	Low (unless your brand)	None	None	AI engines redirect to the brand. No snippet or AI citation opportunity for third parties.

How intent affects AEO scoring

High AEO intent signals:

• Question-format queries ("how to", "what is", "why does") - these directly trigger featured snippets and are natural voice search patterns
• Definition queries ("X definition", "X meaning", "what is X") - paragraph snippets
• Process queries ("how to X step by step") - list snippets
• Comparison queries with clear structure ("X vs Y") - table snippets
• "Best" queries with listicle format - list snippets

Low/zero AEO intent signals:

• Brand navigational queries - no snippet opportunity
• Direct purchase queries ("buy X", "order X") - shopping results, not snippets
• Login/account queries - Google shows the brand's page directly
• Price-only queries ("X pricing") - sometimes a snippet, but usually product page

How intent affects GEO scoring

High GEO intent signals:

• Multi-factor evaluation queries ("best X for Y") - AI engines love producing structured comparisons and cite multiple sources
• How-to queries with multiple steps - AI engines generate step-by-step answers and cite authoritative guides
• "Explain" queries ("how does X work") - AI engines synthesize explanations from multiple cited sources
• Emerging topic queries - AI engines pull from recent sources when training data is insufficient, increasing citation opportunity

Low/zero GEO intent signals:

• Simple factual queries ("population of France") - AI answers from training data without citations
• Navigational queries - AI engines redirect, don't generate answers
• Very recent event queries - AI engines may not have indexed recent sources
• Highly subjective queries ("is X worth it") - AI engines are cautious and may not generate an answer at all

---

AI Overview intent patterns

Google AI Overviews and other AI search engines do not fire equally across all query types. Understanding which patterns trigger AI answers helps you predict GEO opportunity during keyword research.

Query types that consistently trigger AI Overviews

Query pattern	Example	AI Overview behavior
"How to" process queries	"how to set up a home network"	Detailed step-by-step answer with 3-6 citations
"What is" explainer queries	"what is retrieval augmented generation"	Definition + context with 2-4 citations
"Best X for Y" comparisons	"best project management tool for remote teams"	Structured comparison with 4-8 citations
"X vs Y" comparisons	"PostgreSQL vs MySQL for web apps"	Side-by-side analysis with 3-5 citations
Multi-factor decision queries	"should I use TypeScript or JavaScript"	Pros/cons analysis with 3-6 citations
Complex informational queries	"how does mRNA vaccine technology work"	Detailed explanation with 4-7 citations

Query types that rarely trigger AI Overviews

Query pattern	Example	Why AI Overview doesn't fire
Direct purchase intent	"buy iPhone 16 Pro"	Google serves shopping results instead
Brand navigational	"Netflix login"	Direct link is more useful
Simple factual	"height of Mount Everest"	Knowledge panel handles this
Very local queries	"pizza near me"	Local pack is more appropriate
Current news/events	"election results today"	Top stories/news box serves this
YMYL health/finance (sometimes)	"should I take aspirin daily"	Google is cautious with AI answers on health/finance topics

Using AI Overview patterns in keyword research

During the research phase, use these patterns to quickly estimate GEO potential:

1. If a keyword matches a "consistently triggers" pattern, start with GEO score 5+
2. If a keyword matches a "rarely triggers" pattern, start with GEO score 0-2
3. Always verify with a manual search - patterns have exceptions
4. For borderline keywords, test on both Google and Perplexity - if either fires a detailed cited answer, there is GEO opportunity

Tool-Specific Keyword Research Workflows

This reference provides step-by-step workflows for performing tri-surface keyword research using popular SEO tools and free alternatives. Each workflow produces the data needed to score keywords across organic, AEO (answer engine), and GEO (AI search) surfaces.

---

Ahrefs workflow

Ahrefs is best for traffic potential estimates, keyword difficulty, and SERP feature analysis.

Step 1: Seed keyword expansion (Keywords Explorer)

1. Enter your seed keyword(s) in Keywords Explorer
2. Go to "Matching terms" for direct variants and "Related terms" for semantic expansions
3. Apply filters:
   • Volume: set minimum based on your niche (usually 50+ for B2B, 200+ for B2C)
   • KD: filter to your site's realistic range (new sites: 0-30; established: 0-60)
   • SERP features: check the "SERP features" column - note which keywords trigger featured snippets, PAA, and AI Overviews
4. Export the filtered list (include columns: keyword, volume, KD, traffic potential, SERP features)

Step 2: Competitor gap analysis (Content Gap)

1. Go to Content Gap tool
2. Enter 3-5 competitor domains in "Show keywords that the following rank for"
3. Enter your domain in "But the following target doesn't rank for"
4. Filter results:
   • Positions: 1-20 for competitors
   • Intersect: at least 2 of the competitors rank (validates the keyword)
5. Export and merge with your seed expansion list
6. In the export, note the SERP features column for each keyword - this feeds AEO scoring

Step 3: SERP feature audit for AEO scoring

1. For your top 50-100 priority keywords, click into each keyword's SERP Overview
2. Record in your spreadsheet:
   • Featured snippet present? (yes/no) and format (paragraph/list/table)
   • Who holds the snippet? (domain + DR)
   • PAA count (number of People Also Ask boxes)
   • Video carousel present?
3. These data points feed directly into the AEO Opportunity Score rubric

Step 4: AI Overview check for GEO scoring

Ahrefs has begun tracking AI Overview presence in SERP features (as of 2025). Check the "AI Overview" filter in Keywords Explorer. For keywords where Ahrefs shows AI Overview data:

1. Note whether AI Overview is present
2. For critical keywords, manually verify on Google (Ahrefs data may lag)
3. Record in your spreadsheet for GEO scoring

---

Semrush workflow

Semrush excels at keyword clustering, intent classification, and SERP feature filtering.

Step 1: Seed expansion (Keyword Magic Tool)

1. Enter seed keyword in Keyword Magic Tool
2. Use the topic groups on the left sidebar to explore sub-topics
3. Apply filters:
   • Volume: minimum threshold for your niche
   • KD%: realistic range for your domain
   • Intent: Semrush auto-classifies intent (I, N, T, C) - use this as a starting point
   • SERP features: filter by "Featured snippet", "People Also Ask", "AI Overview"
4. Export with all columns including intent and SERP features

Step 2: Competitor gap (Keyword Gap)

1. Go to Keyword Gap tool
2. Enter your domain vs 3-4 competitors
3. Select "Missing" tab (keywords competitors rank for but you don't)
4. Apply intent filter to focus on your target intent types
5. Sort by traffic potential or volume
6. Export and merge with seed expansion data

Step 3: SERP feature analysis

1. In the exported data, Semrush includes SERP feature indicators
2. Filter for keywords with featured snippets - these are your AEO candidates
3. Filter for keywords with "AI Overview" - these are your GEO candidates
4. For keywords with both, you have dual-surface opportunity
5. Cross-reference with the tri-surface scoring rubric

Step 4: Intent validation

Semrush's auto-intent is a good starting point but not always correct:

1. Spot-check 10-15% of your list against live SERPs
2. For any keyword where Semrush says "informational" but the SERP shows product pages, override the classification
3. Pay special attention to keywords Semrush marks as "commercial investigation" - these often have the highest AEO and GEO opportunity

---

Google Search Console + free tools workflow

For teams without paid tool access, this workflow uses GSC and free resources.

Step 1: Mine existing rankings (Google Search Console)

1. Go to GSC > Performance > Search results
2. Set date range to last 3 months
3. Export all queries with impressions > 10
4. Sort by impressions descending - these are keywords Google already associates with your site
5. Identify "striking distance" keywords: position 5-20 with decent impressions
   • These are your fastest organic wins
6. Add columns for intent classification (manual, using modifier rules from search-intent-mapping.md)

Step 2: Expand with free tools

Google Autocomplete:

1. Type your seed keyword into Google search
2. Note all autocomplete suggestions
3. Add a letter after the seed ("seed keyword a", "seed keyword b"...) for more suggestions
4. Record all unique suggestions

People Also Ask mining:

1. Search your seed keyword on Google
2. Click on PAA questions to expand them (this loads more questions)
3. Record all PAA questions - each is a potential keyword with high AEO value
4. PAA questions are inherently snippet-eligible, so they score high on AEO

AlsoAsked.com:

1. Enter your seed keyword
2. Export the question tree (shows how PAA questions branch from each other)
3. Use the question tree to build your keyword cluster hierarchy

Google Trends:

1. Compare seed keyword variants to see which has higher/growing interest
2. Check "Related queries" for rising terms (potential emerging opportunities)
3. Use trend data to break ties between similar keywords

Step 3: Manual SERP audit for AEO/GEO scoring

Without paid tools, you must manually check SERPs for AEO and GEO signals:

1. For your top 30-50 keywords, search each one in an incognito browser
2. For each keyword, record:
   • Featured snippet? Format? Holder domain?
   • PAA count?
   • AI Overview present? How many sources cited?
3. Also test on ChatGPT Search (chat.openai.com with search enabled) and Perplexity (perplexity.ai):
   • Does the query trigger a detailed answer?
   • How many sources are cited?
   • What types of sources are cited? (blogs, official docs, news, forums)
4. Record all data in your scoring spreadsheet

---

ChatGPT Search and Perplexity manual audit

This audit is essential for GEO scoring and should be performed monthly for priority keywords.

ChatGPT Search audit

1. Open ChatGPT with search enabled
2. Enter your keyword as a search query
3. Record:
   • Does it generate a search-augmented response? (some queries don't trigger search)
   • How many source citations appear?
   • What domains are cited? (note domain authority/type)
   • What content format do cited sources use? (list, how-to, data-rich, comparison)
   • Is your site cited? If not, which competitor is?

Perplexity audit

1. Open Perplexity.ai
2. Enter the same keyword
3. Record the same data points as ChatGPT Search
4. Perplexity tends to cite more sources per answer - note the full citation list
5. Compare which sources appear in both ChatGPT Search and Perplexity - consistent citations indicate strong GEO positioning

Monthly cadence

• Audit your top 20-30 priority keywords monthly
• Track citation changes over time (are you gaining or losing citations?)
• New keywords entering your pipeline should get a one-time audit before scoring
• Quarterly: re-audit your full keyword list (top 100) for GEO score updates

---

Spreadsheet scoring template

Use this column layout for your tri-surface keyword research spreadsheet:

Required columns

A: keyword
B: intent (informational / navigational / transactional / commercial)
C: search_volume (monthly)
D: traffic_potential (estimated total traffic for ranking page)
E: keyword_difficulty (0-100)
F: cluster_name
G: snippet_present (yes / no / n/a)
H: snippet_format (paragraph / list / table / n/a)
I: snippet_holder_domain
J: paa_count (0, 1-3, 4+)
K: ai_overview_present (yes / sometimes / no)
L: ai_overview_citation_count (number)
M: organic_score (0-10, calculated)
N: aeo_score (0-10, calculated)
O: geo_score (0-10, calculated)
P: composite_score (M + N + O)
Q: weighted_composite (with business goal weights applied)
R: priority_surface (organic / aeo / geo / dual / tri)
S: action (new page / optimize existing / FAQ / deprioritize)

Formulas (Google Sheets / Excel)

Organic Score (simplified):

=ROUND((MIN(D2/700,3)*3 + (3-MIN(ROUND(E2/23),3))*3 + IF(B2="transactional",2,IF(B2="commercial",2,IF(B2="informational",1,0.5)))*2) / 3, 0)

AEO Score (simplified):

=IF(B2="navigational",0, ROUND((IF(G2="yes",IF(I2="",2,1),IF(B2="informational",2,0))*2 + IF(G2="n/a",0,2)*2 + IF(J2="4+",2,IF(J2="1-3",1,0))*2) / 2, 0))

GEO Score (simplified):

=IF(K2="no",0, ROUND((IF(K2="yes",3,IF(K2="sometimes",1,0))*3 + MIN(L2,2)*2 + IF(B2="informational",2,IF(B2="commercial",2,1))*2) / 2, 0))

Note: these formulas are approximations of the full rubric. For the most accurate scores, evaluate each factor individually using the rubric in references/tri-surface-scoring.md.

Conditional formatting

• Green fill (score >= 7): high opportunity on this surface
• Yellow fill (score 4-6): moderate opportunity
• Red fill (score 0-3): low or no opportunity
• Bold text in priority_surface column
• Sort by weighted_composite descending to see your best opportunities first

Pivot table suggestions

• Pivot by cluster_name to see aggregate opportunity per topic cluster
• Pivot by priority_surface to see the distribution of organic vs AEO vs GEO opportunities
• Pivot by intent to see which intent type has the most tri-surface opportunity

Tri-Surface Keyword Scoring

In 2026, every keyword can drive traffic from three distinct surfaces: traditional organic blue links (SEO), answer engine features like featured snippets, People Also Ask, and voice results (AEO), and AI-generated search results from Google AI Overviews, ChatGPT Search, and Perplexity (GEO). Scoring a keyword on all three surfaces reveals where the real opportunity lies - some keywords are best won organically, others through snippet capture, and others by earning AI citations. This reference defines the scoring rubrics for each surface and a composite scoring method.

---

Why tri-surface scoring matters

Traditional keyword research only evaluates organic ranking opportunity. You look at search volume, keyword difficulty, and intent alignment, then decide whether to pursue the keyword based on whether you can realistically rank on page one. This single-surface view was sufficient when organic blue links captured the vast majority of clicks on the results page. That is no longer the case.

Answer engines now serve direct answers for roughly 30-40% of informational queries. Featured snippets, People Also Ask boxes, and voice search results pull users away from clicking through to organic listings. If you rank #3 organically but someone else holds the featured snippet, your effective click-through rate drops significantly. Meanwhile, AI search engines - Google AI Overviews, ChatGPT Search, and Perplexity - are creating an entirely new traffic channel by citing sources in their generated answers. Being cited in an AI Overview is functionally a new kind of "ranking" that exists outside the traditional ten blue links.

A keyword might have low organic opportunity (KD 80+) but high AEO opportunity (no featured snippet holder, strong PAA presence) or high GEO opportunity (AI Overview fires consistently but cites weak, outdated sources you could displace). Evaluating all three surfaces prevents blind spots and reveals non-obvious wins. Without tri-surface scoring, you might skip a keyword because organic difficulty is too high, completely missing that it represents a wide-open opportunity on answer engine or AI-generated surfaces.

---

Surface 1: Organic Opportunity Score (0-10)

Score each keyword on its traditional organic ranking potential.

Rubric

Factor	Weight	Scoring
Traffic potential	3	0 = <100/mo, 1 = 100-500, 2 = 500-2K, 3 = 2K+
KD inversion	3	0 = KD 70+, 1 = KD 50-69, 2 = KD 30-49, 3 = KD <30
Intent-business alignment	2	0 = no match, 1 = awareness only, 2 = consideration/commercial
Content gap exists	2	0 = top 3 results are strong, 1 = gaps in top 10, 2 = clear content gap

Score = sum of (factor score * weight) / max possible * 10, rounded to nearest integer

Simplified: add up the weighted scores. Max raw = 30. Divide by 3 to get 0-10.

How to evaluate

• Traffic potential: use Ahrefs/Semrush traffic potential metric (not raw volume)
• KD inversion: lower difficulty = higher score (inverted because easy = more opportunity)
• Intent alignment: does this keyword's intent match a page type you offer or plan to create?
• Content gap: are the current top results thin, outdated, or mismatched to intent?

---

Surface 2: AEO Opportunity Score (0-10)

Score each keyword on its potential for winning answer engine features (featured snippets, PAA, voice results).

Rubric

Factor	Weight	Scoring
Snippet presence	2	0 = no snippet possible (navigational/transactional), 1 = snippet exists with strong holder, 2 = snippet exists with weak/no holder OR no snippet yet on eligible query
Format match	2	0 = your content can't match the snippet format, 1 = partial match, 2 = you can produce the exact format (paragraph, list, table)
PAA count	2	0 = no PAA, 1 = 1-3 PAA questions, 2 = 4+ PAA questions (more PAA = more entry points)
Voice search likelihood	2	0 = unlikely voice query, 1 = possible, 2 = natural spoken question pattern
Current holder strength	2	0 = held by a top-3 authority domain (Wikipedia, gov), 1 = held by a mid-authority site, 2 = no holder or held by low-authority site

Score = sum of (factor score * weight) / max possible * 10, rounded to nearest integer

Max raw = 20. Divide by 2 to get 0-10.

How to evaluate

• Snippet presence: search the keyword and check if a featured snippet appears; note format
• Format match: can you create a concise paragraph answer (40-60 words), a numbered list, or a table that matches?
• PAA count: count People Also Ask boxes on the SERP
• Voice likelihood: question-format queries ("how do I...", "what is...") are voice-likely
• Holder strength: who currently holds the snippet? A DR 90 site is hard to displace; a DR 30 blog is vulnerable

Key insight

An AEO score of 0 is correct for navigational and most transactional keywords - snippets don't fire for "buy X" or "brand login". Do not force an AEO score onto keywords where answer features are irrelevant.

---

Surface 3: GEO Opportunity Score (0-10)

Score each keyword on its potential for earning citations in AI-generated search results.

Rubric

Factor	Weight	Scoring
AI Overview trigger	3	0 = no AI Overview fires for this query, 1 = AI Overview fires sometimes, 2 = AI Overview fires consistently, 3 = fires on Google + ChatGPT Search/Perplexity
Citation density	2	0 = AI answer cites 0-1 sources, 1 = cites 2-3 sources, 2 = cites 4+ sources (more citations = more opportunity for you to be one)
Query type match	2	0 = query type AI engines skip (navigational, very simple), 1 = AI engines summarize but rarely cite deeply, 2 = AI engines provide detailed cited answers (comparisons, how-tos, multi-step)
Entity/topical relevance	1	0 = your site has no topical authority here, 1 = some, 2 = strong entity match
Content uniqueness	2	0 = your content would be generic/duplicative, 1 = somewhat differentiated, 2 = unique data, original research, or proprietary insight

Score = sum of (factor score * weight) / max possible * 10, rounded to nearest integer

Max raw = 20. Divide by 2 to get 0-10.

How to evaluate

• AI Overview trigger: search the keyword on Google (logged out) and check if AI Overview appears; also test on ChatGPT Search or Perplexity
• Citation density: count how many source links the AI answer includes
• Query type match: comparison queries, multi-factor evaluations, and "how to" queries tend to trigger detailed AI answers with citations; simple factual queries get one-line answers with fewer citation opportunities
• Entity relevance: does your site have established authority on this topic? AI engines prefer citing known entities
• Content uniqueness: AI engines cite sources that add information beyond what's already in their training data - original research, proprietary data, unique frameworks, and first-party case studies get cited more

Critical rule: GEO = 0 when no AI Overview fires

If you search a keyword and no AI Overview, ChatGPT Search answer, or Perplexity answer appears, the GEO score is 0. Do not speculate about future AI coverage - score based on current observable behavior. Re-evaluate quarterly as AI search coverage expands.

---

Composite Score and Priority Surface

Calculating the composite score

The composite score is the simple sum of the three surface scores:

Composite Score = Organic Score + AEO Score + GEO Score

Range: 0-30. Higher = more total opportunity across all surfaces.

Applying business-goal weighting

Not all surfaces matter equally to every business. Apply a multiplier based on your primary goal:

Business goal	Organic weight	AEO weight	GEO weight
Maximum traffic volume	1.5	1.0	0.8
Brand authority / thought leadership	1.0	0.8	1.5
Lead generation / conversions	1.2	1.2	0.8
Awareness in AI-first audiences	0.8	1.0	1.5
Balanced / default	1.0	1.0	1.0

Weighted composite = (Organic * Ow) + (AEO * Aw) + (GEO * Gw)

Determining priority surface

After scoring, assign a priority surface to each keyword:

• If one surface score is 3+ points higher than both others: that surface is priority
• If two surfaces are within 2 points and both above 5: dual-surface opportunity
• If all three are within 2 points and above 4: full tri-surface opportunity
• If all scores are below 3: low opportunity keyword - deprioritize

The priority surface informs which optimization skill to invoke next:

• Organic priority -> standard SEO content optimization
• AEO priority -> invoke aeo-optimization skill for snippet/PAA formatting
• GEO priority -> invoke geo-optimization skill for citation optimization
• Dual/tri-surface -> optimize content for primary surface first, then layer secondary

---

Worked examples

Example 1: Informational keyword - "how to improve email deliverability"

Surface	Factor scores	Total
Organic	Traffic: 2, KD inv: 2, Intent: 1, Gap: 2 = 21/30	7
AEO	Snippet: 2, Format: 2, PAA: 2, Voice: 2, Holder: 1 = 18/20	9
GEO	Trigger: 2, Citations: 2, Query: 2, Entity: 1, Unique: 1 = 16/20	8

Composite: 24/30. Priority surface: AEO (highest score, clear snippet opportunity). Action: create a step-by-step guide formatted to win the featured snippet, include PAA subheadings. The strong GEO score means also structuring for AI citation (clear claims, data points, authoritative tone).

Example 2: Commercial keyword - "best project management software for remote teams"

Surface	Factor scores	Total
Organic	Traffic: 3, KD inv: 1, Intent: 2, Gap: 1 = 20/30	7
AEO	Snippet: 1, Format: 1, PAA: 2, Voice: 0, Holder: 1 = 10/20	5
GEO	Trigger: 3, Citations: 2, Query: 2, Entity: 1, Unique: 1 = 18/20	9

Composite: 21/30. Priority surface: GEO (AI engines love comparison queries and cite multiple sources). Action: create a comparison article with unique evaluation criteria and original analysis. Optimize for AI citation by including clear verdict statements, structured data, and original scoring methodology. Also target organic since score is decent.

Example 3: Transactional keyword - "buy standing desk adjustable"

Surface	Factor scores	Total
Organic	Traffic: 2, KD inv: 1, Intent: 2, Gap: 1 = 17/30	6
AEO	Snippet: 0, Format: 0, PAA: 0, Voice: 0, Holder: 0 = 0/20	0
GEO	Trigger: 0, Citations: 0, Query: 0, Entity: 0, Unique: 0 = 0/20	0

Composite: 6/30. Priority surface: Organic (only viable surface). Action: pure product/category page optimization. AEO=0 because transactional queries don't trigger snippets. GEO=0 because AI Overviews don't fire for direct purchase queries. This is a traditional SEO play.

---

Limitations and caveats

AI Overview volatility - Google's AI Overviews are still evolving rapidly. A keyword that triggers an AI Overview today may not trigger one next month, and vice versa. GEO scores should be re-evaluated quarterly. Do not over-invest in GEO optimization for keywords with inconsistent AI Overview behavior.

Tool immaturity - As of 2026, no single tool provides reliable automated tri-surface scoring. The rubrics above require manual evaluation for the AEO and GEO components. Tools like Ahrefs and Semrush are beginning to add SERP feature and AI Overview tracking, but the data is not yet comprehensive enough to fully automate scoring.

Scores are relative, not absolute - A keyword scoring 8/10 on organic is not "objectively" easy to rank for. It means it scores high relative to the rubric's criteria. Always cross-reference with domain-specific factors (your site's authority, content production capacity, competitive landscape).

Research, not optimization - This scoring framework tells you WHERE the opportunity is. It does not tell you HOW to capture it. For execution:

• Organic optimization: standard content SEO best practices
• AEO capture: use the aeo-optimization skill
• GEO citation: use the geo-optimization skill

Sample size - When checking AI Overview triggers, test at least 3 times across different days and browsers. AI Overview appearance can vary by session, location, and Google's ongoing experiments.

---

Spreadsheet template columns

When building a tri-surface scoring spreadsheet, use these columns:

keyword | intent | volume | traffic_potential | kd | organic_score | aeo_score | geo_score | composite | weighted_composite | priority_surface | cluster | action

Conditional formatting suggestions:

• Green highlight: any surface score >= 7
• Yellow highlight: surface score 4-6
• Red highlight: surface score 0-3
• Bold: priority_surface column
• Sort by: weighted_composite descending