There is a specific kind of marketing report that has become dangerous. Every number on it is up. Impressions, rankings, reach, engagement. Nothing is up in the business.
This used to be a sign that someone was reporting badly. It is now frequently a sign that someone is reporting accurately on metrics that stopped meaning what they used to mean.
Here is what broke, how to tell whether it broke for you, and what to measure instead.
The Metrics Were Always Proxies
Impressions, rankings, and reach were never valuable in themselves. They were stand-ins for something harder to observe: whether people who could buy from you were encountering you.
That substitution worked because the relationship held. If you ranked, you were seen. If you were seen, some proportion clicked. If they clicked, some proportion converted. Each step had a stable enough conversion rate that measuring the top told you something reliable about the bottom.
A proxy is only as good as the relationship underneath it. When the relationship changes, the proxy keeps producing numbers and stops carrying information, and there is nothing in the number itself that tells you which state you are in.
What Actually Changed
Three things, and they compound.
The results page got a layer above the results. Google confirmed in February 2026 that AI Overviews appear on roughly half of United States search queries. On those queries, a summary sits above the organic results and frequently answers the question. Your ranking is still your ranking. It is now below the answer.
Clicks fell on exactly those queries. Seer Interactive measured organic click-through on queries showing an AI Overview falling from 1.76 percent to 0.61 percent between the periods it studied, published in September 2025. That is roughly a 61 percent decline in the proportion of impressions that become visits, on the queries where summaries appear.
Ranking stopped predicting citation. BrightEdge reported in February 2026 that only about 17 percent of AI Overview citations came from pages ranking in the organic top ten, down from roughly 76 percent in mid-2024. Being in the summary and being in the top ten have substantially decoupled.
Put those together and impressions can rise while visits fall, rankings can hold while visibility declines, and none of it appears on a report built on the old proxies.
The Diagnostic: Does Your Funnel Still Hold?
Ten minutes in your search console will tell you which situation you are in.
Pull the last twelve months. Compare impressions and clicks as two lines, not as a table.
Both rising together: your proxies still hold. The relationship between being seen and being visited is intact for your query mix. This is common for businesses whose queries are navigational, transactional, or in categories where summaries appear less often.
Impressions rising, clicks flat or falling: the relationship has broken for you. You are being seen more and visited less, which means impressions have stopped carrying information about visits. Any report using impressions as a success measure is now reporting on something that does not connect to your business.
Both falling: a different problem, and worth diagnosing separately before assuming it is this one.
Now do the second half. Take your average click-through rate across the period and check whether it declined. If impressions rose while click-through rate fell by a substantial margin, you have the pattern directly.
This does not require a tool and it is not ambiguous. The two lines either diverge or they do not.
What Replaces Them
The replacement is not one metric. It is a short set that reconnects to the business.
Citation presence across a fixed question set. Whether you are named when your buyers ask what they actually ask. This is the direct measure of the thing rankings used to approximate.
Branded search volume. People searching your company name specifically. It is free in your search console, it is hard to fake, and it rises when recognition improves. A buyer who reads a summary that names you and then searches your name produces exactly this.
Inquiries by channel, including a self-reported field. Attribution cannot see a buyer who read a summary and then typed your name into a browser; that arrives as direct traffic with no referrer. A single question at the point of inquiry, how did you hear about us, produces messier and more informative data than any model guessing.
Click-through rate, watched as a diagnostic rather than a goal. Not something to maximize, but the number that tells you whether the relationship between impressions and visits is still holding.
The Trade-Off
Switching to these measures means accepting a report that is less flattering and harder to explain.
Impressions always go up. That is most of why they endured as a headline number: they produce a chart that slopes upward regardless of whether anything happened. Citation presence across a fixed question set does not do that. It moves slowly, sometimes sideways, and occasionally down when a competitor does something well.
There is also a real cost in interpretability. "Impressions up 40 percent" needs no explanation. "We are named in eleven of our twenty-two tracked questions, up from eight, and appearing alongside stronger competitors than last quarter" requires someone to understand the measurement before the number means anything.
The honest position is that the new measures are worse as communication and better as information. If the person reading your report will not sit through the explanation once, you will end up back on impressions, and you will be flying on an instrument that has quietly stopped working.
Where This Fits
Knowing what to measure is separate from knowing how. Our guide to measuring citation presence covers building the question set and running the check.
See how measurement is built into the work at axiaatlas.com/services, or book a demo and we will run the funnel diagnostic on your own data.
Frequently Asked Questions
Are rankings completely useless now?
No. They still matter for queries where no summary appears, which is roughly half of them, and they remain a reasonable diagnostic for technical and relevance problems. What has changed is that they are no longer a sufficient proxy for visibility. Track them as one input rather than as the headline.
My impressions and clicks are still rising together. Do I need to change anything?
Not urgently. That pattern means the relationship is holding for your query mix, which is genuinely the case in some categories. Re-run the diagnostic quarterly, because summary coverage has expanded steadily and category by category. The pattern holding today is not a guarantee it holds next year.
Is this just Google, or does it affect other channels?
The specific mechanism described here is search. The general principle, that a proxy metric can keep producing numbers after the relationship underneath it changes, applies everywhere. Reach on social platforms has undergone a similar decoupling from actual attention over a longer period.
How do I explain a flat metric to someone expecting growth?
By explaining the measurement before showing the number, and by showing it alongside branded search volume and inquiries, which are the things that should be moving. A single flat number with no context invites the wrong conclusion. The same number next to two rising ones tells a coherent story.
Should I stop reporting impressions entirely?
Keep them, and demote them. They are useful as context and as a diagnostic when compared against clicks. What they should not be is the headline, because a headline number that always rises is not measuring anything.
Related reading
- Answer Engine Optimization: What It Is, and Why It Is Not Just SEO With a New Name
- How to Measure Whether Answer Engines Are Recommending You
Sources
- AI Overview prevalence on roughly half of United States queries: Google, official disclosure, February 2026.
- Organic click-through rate on AI Overview queries, 1.76 percent falling to 0.61 percent: Seer Interactive, September 2025.
- Citation overlap with the organic top ten, about 17 percent in early 2026 against roughly 76 percent in mid-2024: BrightEdge, February 2026.