Choosing Which Keywords to Track

Track keywords you could plausibly answer a question with. A tracked set is a sample, and its job is to let you detect and explain change. Terms you rank nowhere for and terms you’ll always rank first for both fail that job — one is a flat line at the bottom, the other a flat line at the top, and neither will ever tell you anything.

Most tracked sets grow by accumulation instead of design, which is why most of them are mostly noise.

The four things a tracked keyword should be

Reachable. Somewhere in the top hundred already, or with a page genuinely aimed at it. A term you don’t rank for at all can be worth watching deliberately as a target, but it belongs in a separate list, because it contributes nothing to your main series except a permanent zero that wrecks any average.

Attached to one URL. You should be able to name the page that’s supposed to rank. If you can’t, you have a content gap, not a tracking need. Recording the target URL is also what makes drop diagnosis fast — sitewide drops versus page-level drops.

Commercially legible. Somebody should be able to say why the business cares. Volume alone isn’t a reason; a high-volume term that never converts is an expensive flat line.

Stable enough to interpret. Some queries are permanently unsettled because the intent behind them is ambiguous. They’re worth tracking, but in their own group, or their movement will drown everything else.

What to leave out

Your brand name. You rank first, you will keep ranking first, and it adds a guaranteed top position to every average you compute. Track brand queries in Search Console where the interesting metric is volume, not position.

Near-identical variants. Singular and plural, minor word order, common misspellings. These usually return the same or near-identical result pages, so they inflate your keyword count and your bill while adding no independent information. Track one and note the variants.

Terms with no realistic page. Aspiration is not a sample.

Anything added “to see what happens” that nobody has looked at since. Every tracked keyword is a small ongoing cost in money and a larger one in attention.

Sizing the set

Smaller than instinct suggests. A set of thirty to eighty keywords, chosen well, grouped, and actually reviewed, is more useful than eight hundred that get skimmed.

The reason is attention, not cost. A large set guarantees that something in it moved dramatically this week, purely by chance, and whatever moved most will be the thing you look at. That’s a random-number generator driving your priorities — and the effect is worse if you’re averaging raw positions, for the reasons in position is an ordinal, not a quantity.

If you have a large site and genuinely need breadth, get it from Search Console, which covers every query that produced an impression at no cost, and reserve the tracker for the terms you need reproducible, comparable numbers on.

Grouping is most of the value

A flat list of keywords supports one operation: sorting. Grouped keywords support diagnosis.

Useful groupings, roughly in order of how often they earn their keep:

• By target URL — makes page-level problems visible immediately.
• By section or template — catches template-level causes.
• By intent — informational and transactional terms behave differently enough that averaging them hides both.
• By layout type — feature-heavy result pages and clean ten-link pages are different animals, per what position one means on a crowded result page.
• By volatility class — one group for the permanently unsettled queries, kept out of the headline number.

Reviewing the set

A tracked set goes stale in both directions: pages get retired and terms stop mattering, while new pages ship without anyone adding their terms.

Once a quarter is enough. Three questions per keyword: does this still have a target URL, has anyone looked at it, and would we notice if it moved. Anything that fails all three comes out. Anything shipped since the last review and not tracked goes in.

Deleting keywords feels like losing history. It isn’t — export first, and keep the export. What you’re actually losing is a line in a chart nobody reads.

What to actually do

1. Give every tracked keyword a target URL at the moment you add it. No URL, no tracking.
2. Cut brand terms and near-duplicate variants out of the main set.
3. Keep not-yet-ranking targets in a separate list so they can’t distort aggregate numbers.
4. Group by URL and section before anything else.
5. Review quarterly and actually delete. A tracked set you trust is small.