Average Position Is an Average of What, Exactly?
It’s the mean position of your best-ranking result across every impression that actually happened. Every location, every device, every layout variant, weighted by how often each occurred. That makes it representative and almost impossible to act on directly, because a single number has flattened a distribution you needed to see.
The habit worth building is to treat average position as a pointer to a question rather than an answer.
Why the mean is the wrong summary here
Position data is bounded at one end and long-tailed at the other. A page can be first, but it can also be fortieth, and the arithmetic mean of those two is twenty — a position the page never occupied.
Concretely, a page that sits third for most impressions but occasionally appears around thirty will report an average somewhere in the high single digits. Nothing about that number describes the page’s behaviour. You would read it as “middle of page one, some room to improve” when the reality is “already near the top for the audience that matters, with a separate long-tail problem attached.”
The reverse case is worse. A page steadily around eleven and a page that alternates between four and eighteen can report the same average, and they need completely different work.
What gets blended in
Four things are mixed into every average position figure, and they blend in proportions you can’t see from the top-level number.
Geography. If you appear third domestically and twenty-fifth abroad, your average is a weighted blend of two unrelated situations. Filtering by country usually splits it into two numbers that each make sense.
Device. Mobile and desktop pages differ in layout and ordering, so the same URL genuinely holds two positions and the average is a blend of both.
Query. At page or site level, the average blends every query that produced an impression. A page ranking first for its exact-match term and fiftieth for a dozen tangential terms will report a mediocre average, driven entirely by queries you never targeted.
Time. An average over three months absorbs whatever happened in week two. Short windows are noisier but interpretable; long windows are stable and uninformative.
The rule for splitting it
Every time an average position looks strange, split it along one axis and look again. Country, then device, then query, then a shorter date range. One of those four splits almost always turns the odd number into two ordinary ones.
If splitting doesn’t resolve it, the page probably really is unstable, which is a finding rather than a measurement artefact — and the shape to look for is described in keyword cannibalisation looks like flapping.
Impressions decide what the number means
A subtlety that trips up a lot of reporting: average position is weighted by impressions, so it tells you about the queries where you already appear. Improve a page enough that it starts surfacing for a whole new set of broader queries and you will typically appear lower down for those — which pulls the average worse while the page is unambiguously doing better.
Rising impressions with a worsening average position is one of the most common healthy patterns there is. It means you gained reach at the edges. Reading it as a decline, and “fixing” it, is how teams undo their own progress.
The corollary is that average position should never be read without impressions next to it. Alone it’s ambiguous in both directions.
Where a tracker helps
A tracker’s job in this pairing is to give you the unblended number: one location, one device, one query, one point in time, reproducible. That’s what lets you say “the average moved because the abroad segment moved” rather than guessing.
The seed comparison for that split of roles is in rank tracker vs. Search Console, and the reason the tracker’s number is bounded and ordinal in the first place is covered in position is an ordinal, not a quantity.
What to actually do
- Never report average position alone. Pair it with impressions and clicks.
- Split before you interpret — country, device, query, then a tighter date window.
- Prefer the median or a distribution when your tool will give you one. If it won’t, look at the position bands instead of the mean.
- Treat a worsening average with rising impressions as good news until proven otherwise.
- Use averages for direction over months, and a tracker for anything you need to explain this week.