# A handful of seeds into a deduped keyword plan.

A five-step agent that expands seed keywords into long-tail variants, dedupes them in code, clusters them into named groups, and writes the plan to a Google Doc.

_Source: https://mentionova.com/agents/keyword-cluster-planner_

It expands your seeds, dedupes them in code, and clusters what is left into named groups in a Google Doc.

Category: Research & analysis.

- **5** chained steps
- **On demand** run mode
- **1** click to add

## Pre-built, then yours to customize.

Add it to your workspace in one click. Sensible defaults and example inputs are already filled in.

- Deduped in code, so the result is repeatable
- Named clusters, not numbered ones
- One recommended page per cluster — no self-competition
- Written straight to Google Docs

Integrations: Google Docs.

Steps in the template: Trigger → LLM call → Code → LLM call → Google Doc push.

## What happens on every run.

### Expand the long tail

Every seed becomes the realistic variants people actually type.

### Dedupe in code

A Filter step removes near-duplicates deterministically, not on a model’s judgement.

### Cluster into named groups

Groups with names you can brief against, not cluster IDs.

### Recommend a page per cluster

One page per cluster, so you stop writing three that compete.

### Write the plan to a Doc

Delivered to Google Docs, ready to divide up.

## The time this gives back.

| Doing it by hand | Time |
| --- | --- |
| Expanding seeds by hand | ~1.5h |
| Spotting duplicates in a sheet | ~1h |
| Grouping and naming clusters | ~1.5h |
| Per plan, by hand | ~4 hrs |

| Keyword cluster planner | Time |
| --- | --- |
| Agent expands and dedupes | auto |
| Agent clusters and names | auto |
| You review the plan | ~25m |
| One agent run | ~25 min |

The templates that publish put a Human review step before the CMS step, and nothing ever posts to Reddit for you. Review is a step, not a lock: a workflow you build or edit can publish or message a channel without one, so add it wherever you want the sign-off.

## Questions, answered.

### Why dedupe in code instead of with the model?

Because deduping is exact work. A Filter step gives the same answer every run; asking a model to spot duplicates does not, and a keyword plan you cannot reproduce is hard to trust.

### How is this different from clustering with SERP analysis?

This starts from seeds you already have and organises them. The SERP version starts from one seed and discovers the cluster from live results. Use this when you have a list, that one when you have a hunch.

### Can I brief from the clusters?

Yes. Each cluster comes with a recommended page, which is the input the content brief agent expects.

## Other agents in Research & analysis

- [Daily citation digest](https://mentionova.com/agents/daily-citation-digest)
- [AEO visibility refresh suggestions](https://mentionova.com/agents/aeo-visibility-refresh-suggestions)
- [Cluster keywords & analyze SERP](https://mentionova.com/agents/cluster-keywords-and-analyze-serp)
- [Analyze SERP for target keyword](https://mentionova.com/agents/analyze-serp-for-target-keyword)
- [Reddit pain-point miner](https://mentionova.com/agents/reddit-pain-point-miner)

---

_Mentionova, https://mentionova.com. Markdown rendering of the page above for AI readers._
