Average position is a story, not a number

The figure Search Console shows for your position is not the position of any one thing; it is an impression-weighted average of where your links appeared across every query you showed up for, over whatever window you are looking at. Each time a link is seen it has a position, those positions are averaged, and queries that generated more impressions pull the average toward themselves more strongly than rare ones do. That construction is reasonable for what it is, but it means the headline number is a blend — a single value standing in for a distribution that might range from the top of page one to the bottom of page five. The moment you treat that blend as if it described a place you actually rank, you have started to misread it, because there is no single place. There is a spread, and the average is only its centre of gravity, weighted toward wherever you happened to be seen most. Two sites with the identical average of 8.0 can be living completely different lives: one sitting steadily around the eighth result on almost every query, the other split between a clutch of first-page winners and a long tail languishing on page five. The single figure flattens those two situations into the same number, and from the number alone you cannot tell the comfortable case from the precarious one.

Because it is one number standing in for a whole distribution, the average is structurally incapable of showing you the things that usually matter most. It cannot tell you that a single high-volume query is carrying the figure while a dozen secondary queries quietly deteriorate beneath it, because their decline is averaged away under the weight of the one term’s impressions. This is the masking effect that does the most quiet damage: one strong, heavily-seen query can hold the whole average aloft like a single tall person raising a group’s mean height, while underneath it a spread of important but lower-volume terms slides without ever showing up in the headline. You feel fine looking at the number and you are bleeding where the number cannot see. It cannot distinguish a commercial query worth real money from an informational one worth almost nothing; both are just positions to be averaged. To the formula, the term that brings buyers ready to purchase and the term that brings idle browsers who will never convert are mathematically identical inputs, each contributing its position weighted only by how many times it was seen. A site can therefore watch its most valuable handful of queries degrade while its average holds perfectly steady, propped up by a mass of low-value terms that no one in the business would trade a single commercial ranking for. The averaging does not merely fail to flag the loss; it actively conceals that loss behind the reassuring outward stability of a single number that has every appearance of health and almost none of its underlying substance. The average has no concept of worth, only of frequency, and frequency and worth are very often inversely related. And it cannot tell you whether a movement reflects your own pages getting better or worse, or simply the mix of queries you appear for shifting underneath you. All of this lives in the breakdown — by query, by page, by intent — and all of it is invisible in the aggregate. The information has not been destroyed; it is sitting right there in Search Console, one click down, in the per-query rows the average was computed from. That is the small irony of the headline number: the detail that would correct it is immediately available, in the same tool, and yet the convenience of the single figure tempts people to stop at it. The fix is not a different tool or a fancier metric; it is the discipline of clicking down to the rows before forming a view. The number is not lying; it is simply summarising, and summary is exactly the operation that discards the detail a decision needs. A summary is a compression, and compression is lossy by design; it keeps the gist and throws away the specifics, which is fine when you want a feel for the trend and fatal when you want to know what to do next. The mistake is not in Search Console for computing the average — it is a perfectly sensible thing to compute — but in the reader who asks a compressed number to answer questions only the uncompressed data can address.

The sharpest way to feel the problem is the case where the average and the outcome point in opposite directions. Your average position improves, and you are pleased — until you notice that the improvement came from newly ranking for a batch of low-intent long-tail terms sitting in middling positions, while the few commercial queries that actually convert slipped several places and cost you real traffic. The metric rose; the revenue fell. The reverse is just as common and just as misleading: the average sags because you have started appearing, far down, for many queries you never ranked for before, which is expansion dressed as decline. A site that broadens its footprint — publishing more, ranking for more, reaching into adjacent topics — will frequently see its average position drift downward simply because it now appears for many more terms, most of them new and therefore low, and the new low positions drag the mean down even as total reach grows. Punishing yourself for that, or worse, reversing the expansion to protect the average, would be a textbook case of optimising the metric at the expense of the thing the metric was supposed to track. The number fell; the business grew. Only the breakdown can tell you which of those is the real event. In both cases the aggregate tells a story the business flatly contradicts, and if the aggregate is the only thing you watch, you will act on the story and not the business. That is not a rare edge case; it is the ordinary behaviour of a value that averages the incomparable. It happens quietly, all the time, on ordinary sites with ordinary query mixes, precisely because the things being averaged — commercial and informational, high-volume and rare, valuable and worthless — are genuinely not the same kind of thing. Any time you average across categories that differ in the dimension you actually care about, the average can move against the dimension you care about. That is not a flaw peculiar to search; it is a property of averages, and it shows up here because position is being averaged across queries that are anything but equivalent.

There is no worse moment to trust the single average than during the rollout of a core update, and we happen to be in one — the November update has been turbulent across most industries, with more violent per-query movement than the previous one — a larger share of pages that reached the top ten were nowhere near the top twenty before it began, and most verticals saw greater swings than they did in the summer’s update produced. During a rollout, rankings lurch, partially settle, and lurch again over a span of weeks, so any reading taken mid-storm is a photograph of turbulence rather than a result. The aggregate is especially deceptive here because it smooths all that violent movement into one comparatively calm line, tempting you to conclude that little happened when a great deal did, or that you were hit when you were merely caught in the churn. The only sound move is patience: let the update finish settling, then ask whether qualified traffic and conversions to the pages you care about held or recovered. The storm is not the time to read the instruments. Pilots are trained not to chase every reading in turbulence, because the instruments swing wildly when the air is rough and a panicked correction to a momentary value can be worse than holding steady; the analogy is exact here. The mid-rollout average swings for reasons that have nothing to do with any decision you should make, and reacting to each swing manufactures work and anxiety without producing insight. The discipline is to note that the air is rough, keep your hands steady, and wait for smooth conditions to read the dials that actually matter.