Keyword Research

---
name: keyword-research
description: Research keywords with Semrush or the free stack — volume, difficulty, questions people ask, and what you already half-rank for — feeding topics, posts, and FAQ blocks. Use when asked to do keyword research, find keywords for a topic, see what people search, mine People Also Ask questions, work out what to write about, research without paying for Semrush, or someone says 'what keywords should this post target'. For the go/no-go verdict on a topic, see topical-authority-check. For the whole site's plan, see site-architecture.
metadata:
  version: 1.0.0
---

# Keyword research

The system runs on any data source — Semrush when you have it, the free stack
when you don't. The course is blunt about the economics: Semrush is roughly
₹10,000 a month, *"which is not where a student should start"*. **You lose
competitor depth on free data, not correctness.**

## Before you start

| Input | Why |
| --- | --- |
| The topic or pillar being researched | Research serves a decision, not curiosity |
| Which tools you actually have | The stack determines the workflow, not the quality bar |
| Your own domain | Your existing rankings are the best free data you own |

## Step 1 — Pick the stack

**With Semrush**: connect it as an MCP server so the agent queries directly
instead of you exporting CSVs. Workflow inside every skill: *discovery tool →
report schema → execute report.* Ask it for: keyword volume and difficulty, the
domains currently ranking, the keyword gap against competitors, and your own
rankings.

**Without Semrush — the free stack that actually works:**

| Tool | Replaces | The catch |
| --- | --- | --- |
| Google Search Console | Your rankings, impressions, CTR, the queries you already half-rank for | Only your site — "still the single best content-refresh input you will ever get" |
| Google Keyword Planner | Volume and competition | Broad ranges unless actively spending |
| Google Trends | Seasonality, rising queries | Relative interest, not absolute volume |
| AlsoAsked / AnswerThePublic | People Also Ask mining for FAQ blocks | Few searches per day on free |
| Bing Webmaster Tools | Keyword research with real volume numbers | Bing data — directionally close for most topics |
| Tavily / Firecrawl | Scraping the live SERP and PAA yourself | You write the parsing |
| Semrush free account | Everything, in small doses | ~10 requests/day — enough to validate one topic, not plan a cluster |

## Step 2 — Run the four research passes

1. **What you already half-rank for** (Search Console): queries at positions
   5–20 are the cheapest wins on the board — feed them to content-refresher.
2. **Volume and difficulty** for the candidate terms (Semrush or Planner).
3. **The questions** (AlsoAsked / PAA): these become the FAQ block and the H2s
   that each answer one question on their own.
4. **Who ranks now** — the top-5 pattern for winnability, and (Semrush only)
   the gap against competitors.

## Step 3 — Cluster by intent, not by string

Group terms by what the searcher wants, not by shared words. One post targets
one intent cluster: a primary keyword, its variants, and the questions that
belong to the same visit. Terms with different intents split into different
posts — that split feeds cluster-planner.

## Step 4 — Hand off with sources attached

Every number carries its source and its shape: Planner ranges stay ranges;
Trends stays relative; quota-limited results say so. Downstream skills (the
topical check, the writer's FAQ block) inherit honest data or they inherit
guesses.

## Output

```
# Keyword research: <topic>

Stack used: <semrush / free — which tools>

## Primary candidates
| Keyword | Volume (shape) | Difficulty | Top-5 pattern | Intent |

## Already half-ranking (positions 5-20, Search Console)
| Query | Position | → route |

## Questions (PAA/AlsoAsked) — the FAQ/H2 feed
- <question> ...

## Intent clusters
| Cluster (one post each) | Primary | Supporting terms |

Caveats: <ranges, quotas, relative-only data — stated>
```

## When it breaks

| What you see | What it means | The fix |
| --- | --- | --- |
| Volumes look absurdly precise | Planner ranges presented as exact numbers | Ranges stay ranges; precision is the tool's, not yours |
| Research "done", no decisions possible | Data collected without the four passes | Each pass answers a question; run the ones the decision needs |
| One post targeting six intents | Clustered by shared words | Re-cluster by what the searcher wants |
| Free stack "not good enough" | Competitor-depth envy | You lose depth, not correctness — the system runs |
| Semrush free account exhausted instantly | ~10 requests/day spent on exploration | Spend them validating one topic, not browsing |
| FAQ block feels invented | Questions written, not mined | PAA/AlsoAsked are the source; quote real questions |
| Same research repeated monthly | Results never recorded | Findings land in the registry/brief, not the chat |

Never present tool output as more certain than the tool itself claims. A range
is a range, Trends is relative, and a quota-limited sample says so — downstream
skills build on what you hand them.

## Rules

- **Search Console first, always**, because what you already half-rank for is
  the highest-ROI data you own, and it is free.
- **Swap the data source, keep the system**, because the workflow is identical
  on paid and free stacks — only depth changes.
- **Cluster by intent, not string**, because posts target visits, and a visit
  has one intent.
- **Numbers keep their shape** — ranges, relatives, samples — because false
  precision compounds through every downstream decision.
- **Research serves a named decision**, because keyword data without a
  destination is a hobby with a dashboard.

## Related skills

- **topical-authority-check** — consumes this data for the go/no-go verdict.
- **cluster-planner** — intent clusters become the cluster roadmap.
- **content-refresher** — positions 5–20 route there.
- **blog-content-writer** — the questions feed its H2s and FAQ block.