AI Overviews: Why Your Traffic Drops While Rankings Hold

The symptom: rankings climb, clicks fall

You open Search Console. Average position is improving, sometimes you have reached the top five on queries you have targeted for months. Yet the number of clicks is flat or falling. This is not a tracking glitch, and it is not a penalty. It is a pattern now well documented, usually called the decoupling between ranking and traffic.

Ranking-traffic decoupling describes a situation where a page's position in search results stays stable or improves, while the number of clicks it receives declines, because an AI-generated answer satisfies the searcher directly, without requiring a visit to the source page.

This pattern hits closed informational queries hardest: "what is," "how to calculate," "why does." These are exactly the formats a generative engine can summarize in one paragraph. Pages that cover a specific case, a price, a location, or a personal situation stay less exposed, because no generic summary can fully replace them.

Why this happens: zero-click search takes over

Since the wide rollout of Google's AI Overviews and the rise of answer engines such as ChatGPT, Perplexity, or Copilot, a growing share of searches end without a single link being opened. This is called zero-click search: the user gets an answer directly on the results page.

Several studies published in 2026 converge on three findings:

• Informational queries are hit hardest, with organic session drops in the 20 to 40% range on the pages that used to answer them.
• Organic click-through rate for top-ranking pages falls between 34 and 58% on queries that trigger an AI Overview.
• On affected queries, the share of zero-click searches now reaches 65 to 70%.

What this does not mean

It does not mean classic SEO is pointless. A page that is not well indexed or well ranked has no chance of being pulled into an AI Overview or cited by an answer engine in the first place: clean indexing and technical relevance remain the foundation. What changes is what counts as success. Raw organic traffic becomes a partial metric. It needs to be complemented with citation and presence metrics, not replaced.

Not every answer engine behaves the same way

Google's AI Overviews, ChatGPT's answers, and Perplexity's answers do not draw on the same sources or follow the same mechanics. A page absent from an AI Overview can still be cited by Perplexity, and the reverse is just as common. Diagnosis should happen engine by engine rather than as one blanket verdict along the lines of "AI is killing my traffic." Reality is rarely that uniform, and treating three different systems as one blurs the very data you need to act on.

This is also why a single Search Console screenshot rarely tells the whole story. Search Console only reports on Google. A page can look untouched there while quietly losing ground in ChatGPT or Perplexity answers, or the other way around. Any diagnosis that stops at Google alone is, at best, half the picture.

Diagnose it in Search Console in ten minutes

Before blaming AI Overviews for a traffic drop, check the mechanics in your own data. The method:

1. Open the Performance report in Search Console and display average position, impressions, and clicks over the last twelve months.
2. Filter by page, then identify the queries generating the most impressions for that specific page.
3. Compare how average position and clicks moved over the same period: a stable or improving position combined with falling clicks is the characteristic signal of decoupling.
4. Manually search the queries in question on Google and note whether an AI Overview or answer panel appears above the classic results.
5. Sort your most affected pages by intent type: a page answering a simple, closed question is more exposed than one that compares, calculates, or personalizes.
6. Segment by device: AI Overviews do not render identically on mobile and desktop, and the click gap can concentrate on just one of the two.

Classic SEO metrics vs AI-era metrics

Tracking classic organic traffic alone hides part of what is really happening. Here are the metrics to add to your dashboard.

None of these metrics stands on its own. A business can lose classic traffic while gaining citations and brand recognition at the same time. In that case, the traffic drop is not a failure, it is value shifting toward a channel that does not yet get measured in clicks.

Building this second column of metrics takes discipline rather than a big budget. A fixed set of ten to fifteen questions your prospects would plausibly ask an AI assistant, checked monthly against the same engines, is enough to start. What matters is consistency over time, not the size of the question set on day one.

What no longer works

Publishing three-hundred-word pages that answer a closed question, hoping to capture informational traffic, makes less and less sense: it is exactly the format an AI Overview summarizes best. Writing a "better" version of the same generic answer does not change the underlying mechanic. The same applies to glossary pages and generic definitions with no distinct point of view: they feed the models without ever earning the visit.

Multiplying near-duplicate articles to cover every possible phrasing of the same question does not work any better: a generative engine already merges those variations into a single answer on its own side. The volume of generic content produced no longer correlates directly with the traffic it earns, while it still carries a real cost in writing, maintenance, and review.

What works: earn the click, earn the citation

Two levers, and they reinforce each other.

The first is giving people a reason to click that goes beyond raw information: a personalized calculation, a simulator, a comparison that depends on the reader's own context, a concrete case that cannot be generalized into a summary. An AI Overview can explain what a non-compete clause is; it cannot evaluate yours. An interactive tool, a personalized diagnostic, or a direct point of contact remain reasons to click that no summary replaces.

The second, more structural lever is working to become the cited source rather than only the clicked page. That is the whole point of GEO, visibility inside AI answer engines: a brand cited in an AI Overview or a ChatGPT answer earns recognition and, according to several analyses published in 2026, more clicks on what remains clickable, including paid clicks. To find out where you stand on that front, the method for tracking AI search visibility details how to build a fixed question set and a tracking table over time.

These two approaches, classic SEO and GEO, are not opposed: they complement each other. What actually changes between the two is covered in our comparison, GEO vs SEO: what actually changes.

A sector example: Swiss trust and advisory firms

A Geneva-based trust company publishing a generic article on "how VAT works in Switzerland" gets absorbed by answer engines almost every time: it is a closed, factual question, already covered hundreds of times elsewhere. The same firm is better served publishing content that does not exist anywhere else in that form: a scenario tied to a specific industry, a reading grid specific to a cantonal regulation, or a simulator that requires real data input. This type of content stays hard to absorb into a generated answer, because its value only exists when applied to a particular situation, not when averaged into a summary.

What this means for publishing frequency

This shift pushes toward fewer generic articles and more content that commits to a point of view, a proprietary method, or a tool. A slower publishing pace with pages that have a real chance of being cited or read generally produces more value than a high volume of interchangeable content.

When this is NOT an AI Overviews problem

Before blaming every traffic drop on AI Overviews, rule out the more mundane causes, which are often easier to fix:

• A seasonal drop in search volume for your target queries, visible by comparing impressions year over year, not just clicks.
• Internal cannibalization: two pages on your site compete for the same query and neutralize each other in the rankings.
• A recent technical issue, slow load times, server errors, content accidentally blocked for crawlers, that hurts indexing before it ever affects AI Overview eligibility.
• A data lag in Search Console: the most recent data stays partial for several days.

If average position drops at the same time as traffic, the problem is most likely a classic ranking issue, not a zero-click search issue. The two diagnoses call for different fixes, and mixing them up wastes time on the wrong fix.