In this article
I spend a lot of time explaining something to my clients that initially sounds contradictory: that falling clicks on a page that still ranks well often isn't bad news.
In part 2 of this series, I went through how Google's AI Overviews have turned click-through rates upside down. Position 1 drops from a click-through rate of around 30 per cent to somewhere between 8 and 12 per cent when an AI Overview appears for the same search. This article takes it a step further: if clicks no longer tell the whole story, what should you measure instead?
Why clicks are lying right now
Most of the clicks that are disappearing were tied to purely informational searches. The user really just wanted a quick answer, not a visit to your website. That type of search is now increasingly being answered directly, either in an AI Overview or in a conversation with ChatGPT.
This means that a falling click count today can mean two completely different things:
- Something has actually gotten worse. You've lost your position, competitors have taken your place, or something technical is wrong.
- Nothing has gotten worse at all. You're still being found just as often, but the user no longer needs to click through to get their answer.
These two situations require completely different responses, and clicks alone can't distinguish between them.
The four metrics I stick to
Average position (Google Search Console)
This is, in practice, a measure of whether Google thinks your content deserves a good spot. If your position remains stable or improves, it's a sign that Google still considers your content to be what they refer to as "high quality".
It's worth mentioning why this is also relevant beyond Google Search itself, but with a caveat I want to be clear about. Some AI models, especially Google's own Gemini, still fetch information directly from Google's search index when they need to answer something topical. This is often called retrieval—the process of fetching relevant information before a response is generated. For Gemini specifically, this means that a signal that performs well with Google often has a positive spillover effect on AI visibility there.
But as I've shown throughout the rest of this series, especially in part 1, that connection is far weaker for ChatGPT, Claude, and Perplexity. The proportion of AI-cited pages that were actually in Google's top ten has dropped sharply over the past year, and several models choose sources in ways that no longer closely follow classic Google rankings. So, a good ranking on Google is still a sensible thing to aim for, but it's no longer a reliable shortcut to general AI visibility across all models.
But I have to add a major caveat here, because average position is actually one of the most misinterpreted metrics in Search Console. The number is an average across all queries that generated an impression in the selected period, and it can be misleading in several ways at once:
- A page might show an average position of 14, while it actually ranks at position 3 for its main keyword and position 60 for a dozen less relevant searches.
- If you remove content you rank poorly for from a page, the aggregate number improves, even though you've actually shrunk your reach, not strengthened it.
One commentator has called average position “survivorship bias turned into a line graph”, and that's an apt description.
My conclusion is that average position is most useful when filtered by a single page or single query over time, not as one aggregate number for the entire site. Look for the trend over several months, not the exact number in a given week.
Impressions (Google Search Console)
How often do you actually appear in the search results, regardless of whether anyone clicks? This is perhaps the most underrated of the four metrics.
If you see a clear increase in impressions, you know your SEO strategy is having an effect. You're becoming visible more often, even if not all those impressions are converted into clicks anymore. And since a page that appears often for a given topic is also one that Google clearly considers relevant for that topic, an increase in impressions can be read as an early signal of increased visibility in Gemini specifically, given the link to the Google index I mentioned above, but not necessarily in the other models.
The same caveat applies here as for position: an impression doesn't necessarily mean someone actually saw the link. For most result types, an impression is counted the moment the result is present in the search results, even if it's far down the page and the user never scrolled to it.
Conversions
These should remain stable, or ideally increase, regardless of what happens with clicks and impressions. If conversions are falling while your impressions are going up, it's a sign of a completely different problem than an AI takeover of the search landscape, and it deserves its own, separate investigation—not an explanation like "it's just because of AI Overviews".
I go into more detail about how I think about conversion tracking in a separate article, and about why good UX is the entire CRO job.
AI visibility specifically
Finally, we try to measure some of the new type of visibility to which we've effectively "lost" the clicks—that is, whether you actually appear inside the AI answers themselves, not just in the classic search results.
This is neither easy nor precise yet. I've dedicated part 4 of this series to this very topic: what tools are available, what they can actually measure, and why I still treat their numbers as a rough guide rather than the gospel truth.
What I actually tell clients when clicks are falling
The short answer is: look at the other three first, before you conclude anything.
If your position has remained stable, your impressions have gone up, and your conversions are unchanged or better, while clicks have fallen, the most likely explanation is that search behaviour has changed around you, not that your work has stopped being effective.
The users who used to click through for a simple answer no longer do so, but that doesn't mean Google or the AI models have stopped finding and valuing your content.
If, on the other hand, your position has dropped, impressions have gone down, and conversions have failed along with the clicks, there's something real to investigate, and AI Overviews aren't the excuse.
Sources
- Support.google.com, What are impressions, position, and clicks? (Search Console Help), the primary source for how GSC actually defines and calculates these metrics
- MBADV Agency, Google Search Console Performance Report & Search Analytics, with further reference to Ahrefs' analysis of 300,000 keywords (February 2026)
- SEMOptimiser, Google Search Console: the complete 2026 guide
- Yositeup, Google Search Console Data Guide 2026: Read Your Metrics Right
- SoRank, Average Position: How to Read Google's Most Misunderstood Metric in 2026
- Ritner Digital, Impressions Down but Average Position Up in Google Search Console
- Incremys, Google Search Console Impressions: Analysis Guide (2026)
- Incremys, Google Search Console Position: Analysis Guide (2026)
The figures on how average position and impressions are calculated have been verified directly against Google's own documentation. The specific figure on the click-through rate drop, from 7.3 to 1.6 per cent, comes from a secondary source that itself refers to an Ahrefs analysis I have not read in its original form.
