Your Rankings Held and Your Traffic Fell. Both Things Can Be True.

An impression does not mean anybody looked at you

Google counts an impression whenever your result appears in the current page of results, whether or not it was scrolled into view. Impressions therefore measure appearances rather than attention, which is why they can rise in the same month clicks fall.

Start with the definition, because it is less demanding than people assume. Google states that an impression means a user has seen, or potentially seen, a link to your site in Search, Discover or News¹. The parenthetical is not decoration. It is the whole mechanism.

The rule that follows is more specific. 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 results¹. If your listing sat below the fold, below an answer block, below a map and below four other results, and the person never scrolled that far, that is still an impression.

There are narrower rules for some formats, and they cut the other way. Inside independently scrolling or expanding widgets, such as a results carousel or an expanding section, the item typically has to be scrolled into view or expanded by a click before an impression registers¹. So the same word means slightly different things depending on where you appeared.

The practical consequence for anybody reading a report is that impressions going up is not automatically good news, and impressions holding steady while clicks fall is not a contradiction. It usually means you kept appearing and stopped being chosen, which is a different problem with a different fix.

Position records your best placement, then averages it

Search Console reports the topmost position your link occupied, averaged across queries. Google’s own example counts an appearance at positions 2, 4 and 6 as position 2, so the figure is a best case smoothed twice before you ever see it.

The position number is doing more processing than most people realize. Google states that the value shown is the topmost position occupied by a link to your property or page, averaged across all queries in which your property appeared¹. Two separate simplifications are happening: best case first, then an average across everything.

Google gives the worked example itself. If a query returns your property at positions 2, 4 and 6, the position is counted as 2, because 2 is the topmost. If another query returns you at 3, 5 and 9, that is counted as 3. The reported average across those two queries is 2.5¹. Nothing about positions 4, 5, 6 or 9 survives into the number you read.

Google is also unusually direct about the limits of its own metric. Its documentation states that the position value is a complex metric that can be misleading if you do not understand the subtleties¹, and illustrates the point with a knowledge panel occupying position 6, noting that the number looks bad while the element is in fact in a very prominent place on the page¹.

So a stable average position is genuinely compatible with a worse experience for the person searching. The mix of queries can shift underneath the average, your best placement can hold while your typical one slides, and the visual reality of the page can change without the number moving at all.

What an answer block does to the numbering

Google states that an AI Overview occupies a single position and that every link inside it shares that position. One element can therefore take a large share of the screen while consuming one slot in the count, which moves everything below it down the page without changing much in the report.

Here is the structural detail that explains a lot of otherwise confusing reports. Each element in search results occupies a single position, whether it contains one link or many links and child elements¹. Position is calculated from top to bottom down the primary side of the page and then down the secondary side¹.

Applied to AI features, that produces a specific effect. Google states that an AI Overview occupies a single position in search results and all links in the AI Overview are assigned that same position¹. A block that may fill a large part of the first screen counts once in the ordering, so a result reported at position 2 can now sit considerably further down the page than position 2 used to mean.

There is a second wrinkle worth knowing if you appear inside these features. Clicking a link to an external page in an AI Overview counts as a click¹, so citations are measured. But in AI Mode a follow-up question is treated as an entirely new query, with all impression, position and click data in the new response counted as coming from that new query¹. One person working through a problem appears as several searches.

We are deliberately not attaching a number to how much traffic these features remove. Figures circulate widely, none of the ones we could find traces to a primary source, and this site does not publish numbers it cannot stand behind. The mechanism is documented. The magnitude, for your specific site, is something only your own data can tell you. We covered the surrounding shift in what AI Mode means for a Florida local business.

The order to check things in

Rule out the boring explanations before reaching for the interesting ones. Compare against the same period a year earlier, check whether the query mix has changed, separate branded queries from unbranded, and confirm there is no technical cause before concluding that AI features are what took your clicks.

Almost every diagnosis of this problem starts in the wrong place, with the most dramatic possible explanation. The disciplined order starts with the causes that are common, boring and checkable, which is also the order Google’s own troubleshooting guidance uses: technical issues first, then manual actions, then seasonality and shifts in what people are searching for².

Seasonality deserves particular weight in a market like ours, because Southwest Florida has a demand curve that repeats every year. Google suggests comparing a drop against the same period previously rather than against last month². A decline in May that also happened last May is a season, not an algorithm, and we set that out in how to check whether an update actually hit you.

Then look at composition rather than totals. Split branded queries from unbranded ones, because a fall in people searching your company name is a completely different problem from a fall in people searching what you sell. Look at whether the queries producing impressions have changed, since gaining a lot of impressions on loosely related terms will lower your click-through rate without anything having gone wrong.

And check whether the impressions are even human. Automated traffic distorts these reports more than most owners expect, which is its own topic in what bot traffic does to your Search Console data. If the answer to all of the above is that nothing changed, then a change in what the results page looks like becomes the reasonable remaining explanation.

1. Compare the same period last year. Before comparing to last month. A repeating shape is a season rather than a penalty.
2. Separate branded from unbranded queries. These are different problems and averaging them together hides both.
3. Check whether the query mix changed. New impressions on loosely related terms lower CTR without anything breaking.
4. Rule out technical causes. A stray noindex produces a gradual decline that reads exactly like an algorithm.
5. Look at the actual results page. Search your own terms and look at what now sits above you. The report cannot show you that.
6. Only then conclude it is the SERP. It often is. It should still be the last answer rather than the first.

What to do when the diagnosis is that nothing broke

If position held, no technical cause exists and the results page simply changed shape, the work moves from ranking to being chosen: the title and description, the kind of query worth pursuing, and whether the page answers something an overview cannot.

This is the outcome people find hardest, because there is nothing to fix and therefore nothing to blame. If your placement held, your site is healthy and the page above you now answers the question, the honest conclusion is that the ranking is no longer the thing that was producing the click. That reframes the work rather than ending it.

The first lever is the part of the listing you actually control. Your title and description are what turn an appearance into a visit, and on most sites they were written once and never revisited. If clicks fell while impressions held, the listing is appearing on the page¹ and being passed over, which is a copy problem before it is anything else.

The second is which queries you pursue. Questions with short factual answers are exactly the queries an overview handles well, and competing for them is increasingly expensive. Questions that require judgement, local specificity or something only your business knows are harder to summarise away, which is why the local pages and the genuinely specific material tend to hold up better.

And the honest limit on what anybody can offer you here. Nobody can restore clicks that are being absorbed higher up the page, and a firm claiming they can is describing something outside their control. What is genuinely offerable is a correct diagnosis, work on the listing itself, and a shift toward queries that still reward a visit. If you want a factual starting point, our free SEO audit covers what a machine currently sees of your site.

• Impressions holding while clicks fall is usually a listing problem, not a ranking one.
• Average position can hold while your typical placement gets worse.
• An answer block takes one position and a lot of screen.
• Check seasonality against last year before anything else.
• Nobody can win back clicks absorbed above the results.