Traffic is down and your rankings have not moved. What is going on?

When organic traffic falls while positions hold, in most cases you have not lost the ranking: you have lost the click. The page is where it was, but something has appeared above it — an AI Overview, a block that answers the question outright — and the searcher no longer has a reason to scroll.

It is the most common scenario we see right now, and it has a recognisable signature: impressions steady or rising, clicks falling, average position unchanged. If your numbers have that shape, the problem is not technical.

But it is not the only possibility, and the difference between hypotheses is settled with data rather than intuition. Here is the order that works.

A word on method, before the steps

We have been running these investigations since 2004 — as an Italian SEO agency, first for domestic brands, later on international programmes: search work for Amilon in Spain, in Germany, and now in Poland. Companies buying SEO services in Italy from abroad tend to arrive with exactly this question. Across that time the search results page has been rebuilt repeatedly — universal results, the mobile shift, featured snippets, and now generated answers — and each rebuild produced the same wave of panic, the same emergency rewrites, and the same discovery six months later that half the emergencies were seasonality.

The investigative discipline has not changed once in twenty years. Establish when, before deciding why. Everything below follows from that.

Step 1 — Establish when, precisely

Before asking why, you need the day.

Search Console, sixteen-month comparison, daily granularity. It distinguishes three very different shapes:

The step — the drop happens over one or two days. An event: an algorithm update, a change to the site, a technical fault.

The slope — the decline spreads across weeks. Something structural: competitors improving, content ageing, demand moving.

The season — the drop returns every year in the same months. Not a problem: your market. It must be compared against the same period last year, not against last month.

Half of all emergencies close right here. It was seasonality, and the wrong comparison was the previous month.

Step 2 — Separate “fewer people searching” from “fewer people clicking”

Two opposite diagnoses, distinguished by two numbers.

If impressions fall alongside clicks, demand dropped or you left the results. If impressions hold and only clicks fall, you are still there and nobody is clicking.

The second case is the one this era produces. Across the 29 queries we track in our own sector, 20 already return an AI Overview (measured July 2026): on those, a share of the click that used to arrive simply no longer exists.

This is not catastrophic news and it is unhelpful to present it as such. It is a change of conditions — and it moves the objective from being clicked to being cited.

Step 3 — Look at pages, not the site

Totals hide everything. A site down twenty per cent may have one page down eighty and everything else flat.

Export pages, compare two comparable periods, sort by clicks lost in absolute terms. In our experience the picture is almost always concentrated: a handful of URLs account for most of the loss.

From there the question becomes tractable. Not “why is the site down” but “why is this page down”.

Step 4 — Check what changed on the page

In this order, because the first takes five minutes and the last takes a week.

Is it still indexed? Direct URL inspection. It happens more often than people expect: a noindex left over from a migration, a canonical pointing elsewhere, one redirect too many.

Is it still the same page? A visually successful redesign may have removed the exact paragraph that answered the question. Compare against an archived version.

Did the title change? The most underrated intervention of all: a title rewritten for brand reasons can cost a third of the click. Nobody records it, because at the time it does not look like an SEO change: it looks like a matter of style.

Has it aged? 2021 content on a 2026 topic. No error — just time.

Step 5 — Look outside your own site

The first four checks concern you. If none produced an explanation, the change happened elsewhere.

Who is above you now who was not before? Look at the live results page, not at a report. Sometimes the new occupant is not a competitor: it is a marketplace, a portal, a generated answer.

Did the intent shift? Do the people typing that phrase still want what your page offers? Intent moves quietly, and an informational page loses ground when its query turns transactional.

Was there an algorithm update on that date? Compare against the update history. If the dates coincide, the diagnosis changes character: there is no fault to fix, there is a criterion that moved.

Step 6 — Decide what to do, and in what order

This is where an investigation separates itself from a list of problems.

The rule we use: pages that were already doing well come first. A page that sat on page one and slipped recovers faster and is worth more than a page that never worked. The cost of intervening is similar; the return is not.

And a rule that always holds: on an already-ranking page you do not rewrite, you add. Put the summary at the top, refresh the data, strengthen internal links. Rewriting a page with years of history is the fastest way to turn a dip into a loss.

The mistake we see most often

Acting before the date is established. Ten things get changed at once, traffic moves, and there is no longer any way to know which change worked. We learned this the usual way, by changing ten things at once.

A serious investigation produces an explanation, not a list of optimisations. If an audit hands you thirty generic recommendations without saying what happened and when, you did not buy an investigation. You bought a checklist.

Frequently asked questions

How long does an SEO audit take?
From a few days to two weeks, depending on site size and access. The long part is not the analysis: it is reconstructing what was changed on the site and when, because almost nobody records it.

What access is needed to run an SEO audit?
Search Console and Analytics with full history, read access to the CMS, and — very usefully — a change log for the site. Without history the work is still possible, but you lose the ability to establish precisely which day the decline began, and that date is what the rest of the investigation depends on.

Can a traffic drop be caused by AI Overviews alone?
It can, and it has a recognisable signature: impressions steady, clicks falling, average position unchanged. When the data has that shape there is nothing to repair — no penalty, no technical fault: there is work to do on being quotable, which is a different objective from ranking.

How often should an SEO audit be repeated?
A full investigation happens when something happens. Monitoring, by contrast, is continuous — precisely so that next time, “when did it start” has an immediate answer.

---

Our founder has been investigating search behaviour since 2004, the agency since 2006. In that time the results page has been rebuilt several times and the method has not needed rewriting once. That is the part worth buying: not a tool, but the order of the questions.

→ Request an audit