Traffic is down, rankings held: how to tell if you have a problem

If your clicks are falling while your positions hold, the loss is almost never spread evenly across your site. It concentrates on pages that answer questions, and it largely spares pages that sell things. So before rewriting anything, split your pages into those two groups and look at them separately. Most businesses that do this discover the decline is real, confined to content that was never producing enquiries anyway, and far less alarming than the aggregate number suggested.

This is the most common thing we are asked about at the moment, and it usually arrives as a single scary chart: sessions down twenty or thirty per cent year on year, rankings apparently unchanged. The aggregate chart is the problem. It averages two completely different things.

What the research actually says

The study everyone quotes on this is Seer Interactive's analysis of AI Overviews and click-through rate, published in November 2025. It covered 3,119 search terms across 42 organisations and some 25 million organic impressions between June 2024 and September 2025, which makes it one of the more substantial datasets available.

Its headline findings for the third quarter of 2025 are stark. On queries showing an AI Overview, organic click-through rate averaged 0.70% where the brand was cited in the answer, and 0.52% where it was not — year-on-year declines of roughly 49% and 65%. Queries with no AI Overview did better at 1.45%, but were still down around 46%.

Two things in that study get left out of most summaries of it, and both change what you should do about it.

It looked only at informational queries. The authors are explicit that the analysis covered "informational and educational queries most vulnerable to AIOs". It does not tell you what is happening to commercial and transactional searches — the ones where somebody is choosing a supplier. Quoting a 61% CTR collapse as though it applies to your product pages is quoting the study incorrectly.

It does not establish cause. The authors say plainly that they "cannot definitively prove that citation causes higher CTRs" and that brands with stronger authority may simply be more likely to be cited. Being cited correlates with more clicks. That is not the same as a promise that getting cited will produce them.

The split that tells you what to do

Open Search Console, set a twelve-month comparison, and divide your pages into two buckets.

Bucket one: pages that answer a question. Guides, definitions, "what is", "how to", comparison explainers. These are the pages AI answers replace most directly, because the user's need is satisfied by the answer itself.

Bucket two: pages that sell something. Service pages, product pages, pricing, location pages, anything someone lands on when they are choosing rather than learning.

Now compare the decline in each. In almost every account we look at, the picture is lopsided: bucket one down sharply, bucket two roughly flat or mildly down. If that is what you see, you have lost traffic that was never going to enquire, and your commercial funnel is intact.

If instead bucket two is falling, that is a different and more serious problem, and it is usually not about AI at all — it is a technical issue, a competitor who has got better, or something that happened during a site change.

The number that settles it

Sessions are the wrong unit for this question. Enquiries are the right one.

Put your qualified enquiry count beside the traffic chart for the same period. If enquiries held while sessions fell, you lost readers, not buyers, and the correct response is to stop reporting on sessions — which we have argued at more length elsewhere. If enquiries fell in step, the problem is real and worth money to fix.

This one comparison resolves most of the panic we see, in about ten minutes.

When it genuinely is a problem

Three situations where falling informational traffic matters a great deal rather than a little.

Your business model is the traffic. If you monetise attention — advertising, affiliate, sponsorship — then informational clicks are the product, and this is an existential change rather than an inconvenience.

Your funnel genuinely starts with education. In some considered purchases the guide is where the relationship begins, and the enquiry comes three visits later. If you can show that path in your data, losing the first visit does eventually cost you the enquiry — but you need to show it, not assume it.

You are absent from the answers replacing your clicks. If an AI answer now satisfies the question your page used to answer, the remaining prize is being the source it names. That is a different discipline from ranking, and we have written about why assistants name one business and not another.

What not to do about it

Do not rewrite pages that were already working. Rankings holding is evidence the pages are fine; the change happened in the results page around them, not on your site.

Do not buy AI-specific technical products on the strength of this. Google states there are "no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary" and that no special files or markup are needed — their guidance is worth reading directly. For Google specifically, this remains ordinary search work done well. For the assistants that are not Google, the work genuinely differs, and that distinction is worth keeping straight.

Do not conclude that content is finished. What has deteriorated is content written to capture a click on a question. What still works is content written for someone deciding whether to buy — comparisons, pricing explanations, honest accounts of who something suits. Those pages are harder to replace with a summary, because the reader wants a position, not a definition.

A reasonable response

Do the split. Put enquiries next to sessions. If the commercial bucket is intact, change your reporting rather than your website, and move new content towards decision-stage questions rather than definitional ones.

If the commercial bucket is falling, that is worth investigating properly, and it is usually fixable. Tell us what you are seeing and we will tell you which bucket your problem is in — including when the honest answer is that you do not have one.