Keyword research without fake precision

Begin with language, not a volume score

Collect phrases from customer interviews, support conversations, reviews, community questions, adjacent search suggestions, and the product itself. These sources reveal different things: vocabulary, urgency, expected features, and category framing.

Estimated popularity and difficulty scores can help prioritize a queue, but they are model outputs—not facts. Record the provider and date if you use them. Never present a third-party estimate as the store’s actual search volume.

Map intent to evidence

For every promising phrase, answer four questions:

1. What is the person trying to accomplish?
2. Which page promise meets that intent?
3. What product evidence can the creative show?
4. What would prove that the match was useful after the install?

This prevents the classic failure mode: winning a loosely related query and losing the person as soon as the product opens.

Build clusters you can act on

Useful clusters often combine an audience, problem, feature noun, or desired outcome. Prune competitor trademarks, generic category terms, irrelevant high-volume phrases, repeated roots, and terms already carried effectively by the app name.

Make the uncertainty visible

Label keyword ideas as observed, estimated, experimental, or rejected. Revisit the map after qualitative feedback and product changes. A keyword backlog is a set of hypotheses about language—not a treasure map.

Next exercise

Open the workbench’s Plan & learn tab. Write one line per intent using: intent → phrase → promise → proof. If the proof column is empty, the phrase is not ready.