Harvest Seed Terms From Outside The Tools

A keyword tool can only expand the seeds you give it, so a plan seeded with your own product vocabulary returns the vocabulary of every competitor who did the same. Whole pockets of demand stay invisible because nobody in the market has ever typed the customer word for the thing into a tool. The fix is to collect raw language first from places where people write unprompted, then run that through the tool for expansion rather than the other way round.

The skill file

---
name: keyword-seeds-from-outside-tools
description: Use when your keyword tool keeps returning the same terms every competitor already targets and the list feels closed.
---

# Harvest Seed Terms From Outside The Tools

A keyword tool can only expand the seeds you give it, so a plan seeded with your own product vocabulary returns the vocabulary of every competitor who did the same. Whole pockets of demand stay invisible because nobody in the market has ever typed the customer word for the thing into a tool. The fix is to collect raw language first from places where people write unprompted, then run that through the tool for expansion rather than the other way round.

## What you need first

- unprompted customer language: support tickets, sales call notes, your own reviews and competitor reviews
- forum and subreddit threads where the category is discussed by buyers rather than vendors
- a logged-out browser set to the target country for autocomplete and People Also Ask harvesting
- competitor ad copy, from an ad transparency library or a paid tool

## Method

1. Collect at least 200 verbatim phrases from sources where nobody was asked a leading question. Review text and support tickets are the strongest; survey answers are the weakest, because the question shapes the wording.
2. Harvest autocomplete for each core term with the alphabet and question words appended, logged out and in the target country, and keep the order returned. Order is a rough ranking of relative demand that the volume column will not give you for the tail.
3. Expand People Also Ask two levels deep on the main terms. It is generated from a live model of related questions rather than a clickstream database, so it surfaces phrasings tools report as zero.
4. Read competitor ad copy and third-party comparison pages for decision-stage wording. Ad copy has been tested against real click behaviour, and the phrasing that survived that test is usually the customer phrasing.
5. Strip your own jargon. Where an internal product or category name differs from the noun customers use, keep the customer noun as the seed and record the internal name only as a synonym.
6. Deduplicate by meaning rather than by string, then feed the survivors into the tool for expansion and volume. Treat a zero-volume return as unmeasured, not as absent demand.
7. Tag every seed with the source it came from, so terms harvested from support tickets can later be checked against conversions and the weak sources dropped next round.

## What this produces

A sourced seed list of roughly 100 to 300 customer-language phrases, each tagged with its origin, ready for expansion and clustering.

## Where this goes wrong

- Harvesting autocomplete while logged in, so your own search history comes back to you as market demand.
- Treating every forum sentence as a query - people write in sentences and search in fragments, so the seed is the noun phrase, not the post title.
- Letting the tool silently drop low-volume seeds during expansion, which discards exactly the terms this work existed to find.

---

From the QuQi skill library - https://www.quqi.io/skills/keyword-seeds-from-outside-tools

What you need first

• unprompted customer language: support tickets, sales call notes, your own reviews and competitor reviews
• forum and subreddit threads where the category is discussed by buyers rather than vendors
• a logged-out browser set to the target country for autocomplete and People Also Ask harvesting
• competitor ad copy, from an ad transparency library or a paid tool

Method

1. 01 Collect at least 200 verbatim phrases from sources where nobody was asked a leading question. Review text and support tickets are the strongest; survey answers are the weakest, because the question shapes the wording.
2. 02 Harvest autocomplete for each core term with the alphabet and question words appended, logged out and in the target country, and keep the order returned. Order is a rough ranking of relative demand that the volume column will not give you for the tail.
3. 03 Expand People Also Ask two levels deep on the main terms. It is generated from a live model of related questions rather than a clickstream database, so it surfaces phrasings tools report as zero.
4. 04 Read competitor ad copy and third-party comparison pages for decision-stage wording. Ad copy has been tested against real click behaviour, and the phrasing that survived that test is usually the customer phrasing.
5. 05 Strip your own jargon. Where an internal product or category name differs from the noun customers use, keep the customer noun as the seed and record the internal name only as a synonym.
6. 06 Deduplicate by meaning rather than by string, then feed the survivors into the tool for expansion and volume. Treat a zero-volume return as unmeasured, not as absent demand.
7. 07 Tag every seed with the source it came from, so terms harvested from support tickets can later be checked against conversions and the weak sources dropped next round.

What this produces

A sourced seed list of roughly 100 to 300 customer-language phrases, each tagged with its origin, ready for expansion and clustering.

Where this goes wrong

• Harvesting autocomplete while logged in, so your own search history comes back to you as market demand.
• Treating every forum sentence as a query - people write in sentences and search in fragments, so the seed is the noun phrase, not the post title.
• Letting the tool silently drop low-volume seeds during expansion, which discards exactly the terms this work existed to find.

Use this skill in your own AI

The download is a plain markdown file with the name and trigger in its frontmatter. Where an assistant supports skills it can load itself, that frontmatter is what it reads to decide this one applies.

Questions about this skill

More in Keyword research