nod-keyword-research

Keyword Research

You expand seed keywords into comprehensive, clustered keyword lists using NodesHub Query Fan-out API and iterative SERP mining (PAA + Related Searches loops).

Quick Start

# === Iterative Research (recommended) ===
# Uses SERP PAA + Related Searches in loops for deep keyword discovery

# Standard (5 loops, ~57 tokens max)
python3 .claude/skills/nod-keyword-research/scripts/iterative_research.py "SEO" --gl pl --hl pl --preset standard

# Aggressive (15 loops, ~232 tokens max)
python3 .claude/skills/nod-keyword-research/scripts/iterative_research.py "SEO" --gl pl --hl pl --preset aggressive

# Beast mode (30 loops, ~637 tokens max)
python3 .claude/skills/nod-keyword-research/scripts/iterative_research.py "SEO" --gl pl --hl pl --preset beast

# Custom loops + budget cap
python3 .claude/skills/nod-keyword-research/scripts/iterative_research.py "SEO" --gl pl --hl pl --loops 10 --serp-per-loop 8 --budget 200

# With JSON output
python3 .claude/skills/nod-keyword-research/scripts/iterative_research.py "SEO" --gl pl --hl pl --preset standard --json

# === Simple Fan-out (single pass) ===
python3 .claude/skills/nod-nodeshub-api/scripts/fanout.py "seed keyword" --hl en --mode standard
python3 .claude/skills/nod-nodeshub-api/scripts/fanout.py "seed keyword" --hl en --mode standard --questions --topic-leaders --raw

Iterative Research Presets

Preset	Loops	SERP/loop	Fan-out seeds	Max cost	Expected keywords	Use when
conservative	3	5	1	~22 tokens	50–150	Quick check, single subtopic, tight budget
standard	5	10	3	~72 tokens	150–400	Normal keyword research, enough for a content plan
aggressive	15	15	5	~262 tokens	400–1000+	Deep niche research, full topical map, wide coverage
beast	30	20	5	~637 tokens	1000–3000+	Exhaustive mapping, competitor-level keyword database

Always present these presets to the user before running so they can choose the depth and budget that fits their needs. More loops = more unique keywords discovered through PAA and Related Searches chains.

How it works:

1. Fan-out on seed keyword (AI-generated variants)
2. SERP on seed keyword → extract PAA questions + Related Searches
3. Loop: pick unprocessed keywords → SERP each → extract new PAA + Related Searches
4. Optional fan-out on top discoveries from early loops
5. Deduplicate throughout, save CSV with source tracking
6. Repeat until loops exhausted, queue empty, or budget spent

Cost: Fan-out = 7.5 tokens/keyword. SERP = 1 token/keyword. Use --budget to cap spending.

Parameters Reference

Always present these parameters to the user before running, so they understand what each one does and can choose.

Parameter	What it does	Effect on results	Cost impact
--loops N	Number of SERP mining iterations	More loops = more keywords from deeper PAA/Related chains	+N×serp_per_loop tokens
--serp-per-loop N	Max keywords SERPed per loop	Higher = more keywords/loop but faster token burn. Each SERP extracts ~4 PAA + ~8 Related	+1 token per SERP call
--fanout-seeds N	Fan-out calls on top keywords from early loops	More fan-out = more AI-generated variants as queue seeds	+7.5 tokens per fan-out
--serp-fanout	SERP every fan-out result immediately	Massive keyword boost — each fan-out keyword gets SERPed for its own PAA + Related. Without this, fan-out results just wait in the queue	+1 token per fan-out result (~15-20 per fan-out call)
--expand-popular N	Fan-out on top N keywords that appear most often across SERPs	Expands "hub keywords" that Google considers highly related. Best after several loops	+7.5 tokens per popular keyword
--popular-threshold T	Min appearances for a keyword to count as "popular" (default: 2)	Lower = more keywords qualify for expand-popular. Higher = only strongest signals	No direct cost
--budget N	Hard token spending limit	Research stops when budget is reached, regardless of loops left	Controls max spend
--preset	Pre-configured intensity (conservative/standard/aggressive/beast)	Sets loops, serp-per-loop, fanout-seeds to sensible defaults	See presets table

CSV Output Columns

Column	Description
keyword	The discovered keyword or question
source	How it was found: seed, fanout, paa, related_search, fanout_popular
type	Keyword type: seed, fanout_variant, paa_question, related_search, fanout_popular
discovered_in_loop	Which loop discovered this keyword (0 = seed phase, 99 = expand-popular phase)
serp_overlap	How many different SERPs this keyword appeared in (PAA or Related Searches). Higher = Google strongly associates it with the topic. Use this to prioritize — keywords with high overlap are "hub keywords" worth targeting first

Output Directory

Before running any research, always ask the user where to save results.

Prompt the user with something like:

Where should I save the results? You can:

1. Use an existing directory (e.g., output/, research/seo/)
2. Create a new one (e.g., output/seo_campaign_2026/)

• If the user specifies a directory, pass it via --output <dir>/keywords_<keyword>_<gl>.csv
• If the user doesn't specify, suggest a sensible default: output/<keyword>_<gl>/
• If the directory doesn't exist, create it (the script handles this automatically)
• For multi-keyword research, keep all files in the same directory for easy access

Setup

Requires NODESHUB_API_KEY. Run:

python3 .claude/skills/nod-nodeshub-api/scripts/check_setup.py

If NodesHub is not set up: Walk the user through the full process: (1) Get API key from nodeshub.io (API Playground). (2) Save to .claude/settings.local.json under env.NODESHUB_API_KEY, or run python3 .claude/skills/nod-nodeshub-api/scripts/save_key.py YOUR_KEY. (3) Point to nod-nodeshub-api setup for details. (4) Have them run check_setup.py again to verify.

Workflow: Keyword Expansion

1. Get seed keyword(s) from user — topic, product, or question
2. Run Query Fan-out with --raw for full data
3. Organize results into clusters by subtopic/intent
4. Classify intent for each cluster (informational, commercial, transactional)
5. Prioritize by relevance, estimated difficulty, and business value
6. Deliver structured keyword map

Workflow: Topical Authority Map

1. Start with main topic — run Fan-out in reasoning mode for better quality
2. Identify subtopic clusters from results
3. Run Fan-out on top 3-5 subtopics (standard mode to save tokens)
4. Map relationships — pillar topic → cluster → supporting articles
5. Identify content gaps — topics without coverage
6. Deliver topical map with hierarchy and priorities

Mode Selection

Mode	Cost	Speed	Quality	Use when
standard	7.5 tokens	Fast	Good	Quick expansion, bulk research, budget-conscious
reasoning	30 tokens	Slower	Better	Deep research, topical maps, high-value keywords

Rule of thumb: Start with standard. Use reasoning for your primary topic or when standard results feel shallow.

Clustering Strategy

Group keywords by:

1. Intent cluster — same search intent, similar SERP
2. Topical cluster — same subtopic, different angles
3. Content type — questions vs. comparisons vs. how-tos
4. Funnel stage — awareness → consideration → decision

Cluster Naming Convention

[Main Topic]
├── [Subtopic A] (informational)
│   ├── keyword 1
│   ├── keyword 2 (question)
│   └── keyword 3 (long-tail)
├── [Subtopic B] (commercial)
│   ├── keyword 4
│   └── keyword 5
└── [Subtopic C] (transactional)
    └── keyword 6

Output Format

## Keyword Research: [seed keyword]

**Settings:** hl=[language], mode=[mode]
**Tokens used:** [X]

### Keyword Clusters

#### Cluster 1: [Subtopic Name] — [intent]
| Keyword | Type | Priority |
|---------|------|----------|
| ... | query/question/long-tail | high/medium/low |

#### Cluster 2: [Subtopic Name] — [intent]
| Keyword | Type | Priority |
|---------|------|----------|

### Questions (People Might Ask)
- [question 1]
- [question 2]

### Topical Map
[Main Topic]
├── [Pillar 1]: [keywords]
├── [Pillar 2]: [keywords]
└── [Pillar 3]: [keywords]

### Recommendations
1. [Which clusters to target first and why]
2. [Content type suggestions per cluster]
3. [Quick wins vs. long-term plays]

Token Budget Planning

Task	Mode	Keywords	Tokens
Quick expansion	standard	1	7.5
Topic research	standard	5	37.5
Deep topic map	reasoning	1 + standard x5	67.5
Full niche research	reasoning x3 + standard x10	13	165

Always tell the user the expected token cost before running.

Tips

• Start broad, then narrow — seed with topic, then expand specific clusters
• Questions are gold — --questions flag reveals what users actually ask
• Combine with SERP analysis — validate clusters by checking who actually ranks
• Language matters — always match --hl to target market language
• Save raw JSON — pipe to file for later processing: ... --raw > keywords.json

Report

After collecting data, ask the user:

"Add results to an HTML report?"

1. New report — creates a branded HTML report in reports/
2. Existing report — appends a section to a chosen report
3. Skip — no report

Use render_report_section(json_output) from iterative_research.py, then create_report() or append_section() from report.py.

Related Skills

• nod-serp-analysis — analyze SERP for specific keywords from your research

• nod-content-brief — turn keyword clusters into content briefs

• Language-specific quirks in fan-out results

Consolidation (keep under 50 lines): Before adding a new entry, check file length. If over 50 lines:

1. Merge duplicate/overlapping entries into single proven patterns
2. Remove entries older than 3 months that haven't been reinforced
3. Drop one-off observations that never recurred
4. Move detailed context to LEARNED-archive.md if worth preserving
5. Keep only entries that would change behavior - if obvious, cut it