Why Your Average Position Can Collapse While Your Rankings Do Not Move

An Impression Does Not Mean Anyone Looked

Google states that an impression is counted whenever an item appears in the current page of results, whether or not the item is scrolled into view, as long as the user need not click to see more results (Google, 2026). Appearing on the page is the bar. Being seen is not.

Everything downstream follows from how the metric is defined, and the definition is more permissive than most people assume.

Google says an impression means a user has seen, or potentially seen, a link to your site (Google, 2026). The operative words are potentially seen, and the mechanics are spelled out: an impression is counted whenever an item appears in the current page of results, whether or not the item is scrolled into view, as long as the user need not click to see more results (Google, 2026).

So a result sitting at position 87 on a page that returned a hundred results counts as an impression, identically to a result at position 2 that filled the screen. No eye needs to reach it. No scroll needs to happen. It simply has to be on the page that was returned.

For ordinary human searching this is a sensible rule, because people do scroll and a strict visibility test would be unmeasurable. It becomes consequential when something requests very deep pages of results at volume, because every one of those requests mints real impressions at depths no person would ever reach.

Why the Average Moves When Nothing Moved

Google defines the position value as the topmost position occupied by a link to your property, averaged across all queries (Google, 2026). It is an average, so its value depends on the mix of impressions in the pool, not only on where you rank.

The second definition is the one that does the damage, and it is equally plain.

Google states that the position value shown in the Performance report is the topmost position occupied by a link to your property or page in search results, averaged across all queries (Google, 2026).

Averaged is the word to sit with. If a site ranks in the low teens for the handful of terms that actually produce its business, and then a large volume of impressions arrives at positions between sixty and ninety, the average is dragged toward the new material regardless of what the original rankings did. The pages that make money can hold their exact positions while the headline number halves.

This is why average position is close to useless as a health metric for a site that is not already ranking well across the board. It is a mix, and the mix can be changed by things that have nothing to do with you. Treating it as a rank is the error, and it is an easy one to make because the field is called position.

What Polluted Data Actually Looks Like

The signature is consistent: a step change rather than a ramp, the overwhelming majority of impressions sitting at position 41 and deeper, near zero clicks against a very large denominator, a skew toward one device and one country, and head terms the site never ranked for.

You can identify this from an export in about ten minutes, and it is worth doing before drawing any conclusion.

Look at the daily series first. Real ranking change is a slope, because pages move gradually and unevenly. Machine-generated volume tends to arrive as a step: a normal day, then a slightly elevated day, then a flat plateau several times higher that holds steady. Schedules produce flat lines. Algorithms usually do not.

Then bucket your impressions by position. This is the decisive test. If the large majority sit at position 41 and deeper, with a heavy concentration beyond 60, you are looking at pages of results no ordinary searcher requested. In a healthy profile most impressions cluster in the first two or three pages, because that is where people stop.

Check clicks against that volume. Automated deep querying produces essentially no clicks, so the click rate collapses toward zero as the denominator grows. Then look at the device and country splits, which usually skew hard toward one of each, and finally at the queries themselves. A systematic sweep of broad head terms with no local modifier, on a site that has never ranked for them, is not an audience arriving.

Any one of these could have an innocent explanation. Together they are a pattern, and the pattern is measurement rather than performance.

There Is Nothing to Block

These impressions are created inside Google, by something querying Google. No request reaches your website, so no firewall, filtering layer or server configuration can prevent them. It is not a security event and hardening in response accomplishes nothing.

This is the part that surprises people, and it is worth being precise about because the instinct is to defend.

An impression is recorded when your listing appears on a results page that Google served to somebody. The transaction happens between the requester and Google. Your website is not involved and receives nothing. There is no visit to block, no address to ban and no rule to add, because from your server’s point of view nothing happened at all.

That distinction matters for anyone who does have filtering in front of their site. Traffic protection deals with requests arriving at your infrastructure, which is a genuine and separate problem. It has no reach into what a third party asks a search engine, and any tool that claims otherwise is describing something it cannot do.

So the honest answer to what should we do about it is usually nothing, other than measuring around it. This is uncomfortable advice to give and to receive, and it is still the correct advice.

How to Report Honestly While It Is Happening

Stop quoting the sitewide average for the duration, and report the tracked query set instead. Position on the specific terms that produce enquiries is unaffected by pollution elsewhere, so it keeps telling the truth while the headline number does not.

The practical fix is not technical, it is a reporting decision, and it should be made before somebody panics rather than after.

Report the query set, not the site. Pick the terms that actually produce your business, ten or twenty of them, and track position on those individually. Those figures are unaffected by whatever is happening at position 70 on unrelated terms, so they continue to answer the only question that matters. This is better practice permanently, not only during an episode.

Say plainly what the sitewide figure is doing and why. A client who is told in advance that the average will look alarming for a while, and why, is reassured. The same client discovering it themselves a month later has every reason to doubt the next thing you tell them.

Resist the urge to act on it. The strongest temptation is to start rewriting pages that appear to have collapsed, and that is how a working page gets damaged in response to a measurement artefact. If the tracked terms are steady, the pages are fine.

Then wait, and keep watching the shape. Sweeps of this kind frequently stop as abruptly as they started, and the average returns without anybody doing anything. If instead your tracked terms genuinely decline, that is a different problem, and the way to tell an algorithmic move from an artefact is set out in penalties versus algorithm updates. That is a real problem and it will be visible in the place you are now correctly looking.

If you would rather have someone measure your own picture properly, our free audit reports what is actually reachable and readable on your domain, and it costs nothing.