Position Is an Ordinal, Not a Quantity

A rank number tells you the order of things, not how much of anything you have. Three places is not a unit. Losing three places near the top of the first page can cost most of a keyword’s clicks; losing three places in the forties costs approximately nothing, because nothing was arriving from there anyway.

Almost every misleading ranking report is built on treating position as a quantity you can add, subtract and average.

Why the scale isn’t linear

Click distribution across a result page is steep and front-loaded. The exact shape varies enormously by query type, layout and intent, so nobody should quote you fixed percentages — but the qualitative fact is not in dispute: attention concentrates at the top and decays fast, and by the time you’re on the second page it has decayed to nearly nothing.

That gives you a scale where the gap between adjacent numbers shrinks as the numbers grow. First to second is a large real change. Fourth to fifth is a smaller one. Forty-fourth to forty-fifth is not a change at all in any sense a business cares about.

Subtracting positions pretends all those gaps are the same size. They aren’t, so the difference is not a meaningful measurement.

Three things this breaks

Averaging across keywords. “Our average position improved from 18 to 15” can be produced by a genuinely valuable move from 4 to 3 on one term, or by two dead terms shuffling in the forties. The average can also worsen while every commercially relevant term improves, if you happen to start surfacing for new broad queries at the bottom of the page.

Ranking movement as a leaderboard. Sorting a keyword report by “places gained” puts the deep-page shuffle at the top, because that’s where the numbers move most freely. The genuinely important movements — a two-place change inside the top five — sit near the bottom of the sort.

Percentage change. “Up 40%” on a rank number is meaningless. It’s a percentage of an ordinal. There’s no such thing.

What to do instead

Band it. Group positions into bands and count how many keywords sit in each: top three, four to ten, eleven to twenty, twenty-one to fifty, beyond. Report movement between bands. This throws away the false precision and keeps the part that matters — a keyword leaving the top three is an event, and a keyword moving from 34 to 29 is not.

Weight by opportunity. If you must produce a single number, weight each keyword by something real: its search volume, its conversion value, or simply whether the business cares about it. An unweighted average across a tracked set treats your most valuable term and a term you added out of curiosity as equals.

Report clicks and impressions as the outcome. Position is a diagnostic. It explains why an outcome moved; it isn’t the outcome. This is the same discipline argued in how often should you check keyword rankings, and it makes the ordinal problem mostly disappear, because clicks genuinely are a quantity.

Look at distributions, not means. The set of positions your tracked keywords occupy is more informative than their average. A histogram takes the same space in a report and can’t be read wrong in the way an average can — see average position is an average of what.

The one place arithmetic is fine

Counting is fine. “We hold eleven top-three positions this month, up from eight” is a legitimate quantity: you’re counting keywords, not averaging ranks. Any metric built by counting keywords that satisfy a condition behaves properly, which is why band counts are the workhorse of decent ranking reports.

Visibility and share-of-voice scores are an attempt to do this more precisely by weighting each position by expected click share. They’re a real improvement over a raw average, with their own caveats — what those scores actually compute.

What to actually do

1. Stop averaging raw positions across keywords. Replace it with band counts.
2. Never quote a percentage change on a rank number.
3. Set alert thresholds by band crossing, not by places moved, so a deep-page shuffle can’t page you.
4. Sort keyword reports by value at stake, not by places gained.
5. Lead with clicks; use position to explain them.