Keyword research - the road map to the top ten

Why a roadmap is needed at all

A ninety-day plan for setting up rival tracking from nothing. At the end of each month you hold more measurement and less guesswork than you did at the start of it.

Rival tracking is not something set up in one sitting. Teams that add ten rivals and start tracking twenty phrases in a day stop looking at the table by the third week. The cause is not disinterest; it is the wrong order of setup.

The right order is measure, then compare, then infer. Comparison is impossible before measurement is running; inference is impossible before comparison has accumulated. Set up in that order, each month ends with something usable.

The wrong order works like this: inference is wanted first. "What are our rivals doing?" is asked in week one, the only data available is one day old, so the answer is a guess - and a move is made on the guess.

Ninety days is the minimum for a first meaningful inference. In less time what you hold is a handful of measurements that cannot yet be separated from noise.

Days 31-60: use the control group

In the second month rival data starts doing a job: interpreting your own moves.

The rule this month: before every move, record the control group's average movement for that week. After the move, compute the same average again and subtract it. What is left is the move's measurable effect.

Every result interpreted without that subtraction is wrong. In a week when every phrase fell two places, your phrase falling two places is not the move failing - it is the market. The reverse holds too: in a week when everything rose, your rise is not an achievement.

Metadata tracking also starts this month. Record your rivals' app names, subtitles and descriptions weekly. If a rival rose noticeably on a single phrase, you can check whether their metadata changed that week - if it did, your hypothesis is confirmed.

What is still not done this month: copying a rival's metadata. Liking a rival's subtitle and writing something similar means taking their positioning, and when that positioning does not fit your product the arriving user does not find what they came for.

• Record the control group's average before and after every move.
• Record rival metadata weekly; look for overlap with their movement.
• A rival moving across every phrase at once is strength, not metadata.
• Do not copy; positioning cannot be borrowed, it depends on the product.

What you hold at the end of ninety days

If the plan ran properly, five things exist at the end of the third month and none of them existed before.

A control group: the daily rank history of five to eight apps. Every move you make from now on will be interpreted with that group's movement removed.

A metadata log: when each rival changed what. That log carries concrete examples of how phrases are being targeted.

A tempo measure: the market's average weekly movement. You know your moves have to be larger than it.

A wall list: the phrases metadata cannot shift. You have stopped spending time on them, and stopping is the largest gain.

A source decision: whether review mining works in your category. If it does, the fourth month's work is settled.

After the ninety days

At the end of three months setup is finished and the work turns into maintenance. Maintenance takes far less time than setup but it has to be regular.

Weekly maintenance, fifteen minutes: the control group's total movement, is any rival diverging, has a new name appeared, is your own defended band intact. Those four questions produce a one-sentence output.

Monthly maintenance, half an hour: update the rival list, drop the ones that have gone, add the new entrants to the top 20. A list is right on the day it is built; six months untouched it no longer shows your search rivals.

Quarterly maintenance, an hour: run review mining again, revisit the wall list, check whether the strength threshold has fallen. If your rating count has grown, phrases that looked unreachable may now be reachable.

Keep those three rhythms and rival tracking never becomes a burden. Drop them and six months later you hold a stale list and a measurement nobody can interpret.

• Weekly fifteen minutes: four questions, one sentence of output.
• Monthly half an hour: update the list, drop the gone, add the new.
• Quarterly an hour: review mining and the wall list.
• Without the rhythm the list ages in six months and the measurement stops being readable.

Frequently asked

Is ninety days not too long?

There is no part that can be shortened: a control group needs a few weeks of data to mean anything, and an inference needs a few months to separate from noise. Conclusions drawn earlier are guesses.

How many rivals should I start with?

Five to eight. More costs nothing in rank measurement but it costs readability, and a list nobody looks at is not a control group.

When should I add the second country?

After the loop has settled in the first - at the earliest the end of month two. Starting both at once ends with half a measurement in each.

My rival is above me on every phrase - is the plan still worth running?

Yes, but it produces a different output. If the strength threshold is far above you, what emerges in month three is that these phrases are not taken with metadata and the work has moved to the product. Knowing that is a gain too.

Why is copying a rival's metadata wrong?

Positioning depends on the product. Their subtitle was built for their product; when it does not fit yours the arriving user does not find what they came for, and your rating falls - which weakens your strength across every phrase.