Why Google AI Overviews cut your clicks, and how to still appear

If your Google impressions held steady but your clicks slid, an AI Overview is the most likely reason. The summary now sits at the top of a large share of results, answers the question in place, and leaves the ten blue links below it for the minority who still scroll and click. This is a different problem from getting cited by ChatGPT and Perplexity: those are separate answer engines you court with their own crawlers. An AI Overview is Google answering with Google, built from the same index that already ranks you, which changes both why you lost the click and what you can do about it.

What the click drop actually looks like

The clearest read comes from a July 2025 Pew Research Center study of a panel of US adults who shared their browsing. When an AI summary appeared, users clicked a search result 8 percent of the time; when no summary appeared, they clicked 15 percent of the time, roughly double. Only 1 percent clicked one of the sources cited inside the summary itself. And this was not a fringe surface: 58 percent of the panel ran at least one search that produced an AI summary in a single month. So the pattern is not that you fell in the rankings, it is that the answer is being read where it is shown and the click is being absorbed above your link.

How an AI Overview is built, and why it matters

An AI Overview is grounded in Google’s ordinary Search index, not a separate corpus you opt into. Google’s own documentation describes it using a query fan-out technique: it takes the question, issues several related searches across subtopics, and pulls a supporting page for each, which is why the links beside an Overview are often broader than a single ranked result. Two things follow. First, the gate to appear is the classic one: you have to be indexed and relevant for the query in the first place. Second, within that, the pages that get cited are the ones that cleanly answer one of the fanned-out sub-questions, so the passage discipline from the sibling guide on getting cited by ChatGPT and Perplexity applies here too, on top of ordinary ranking rather than instead of it.

The control that surprises people

Because the Overview is built from the Search index, for a long time there was no way to leave it without leaving Search. The two crawl-level controls, a nosnippet tag or blocking Googlebot, both remove you from normal results as well, so opting out of the summary meant opting out of the ranking under it. And the move people reach for by reflex, disallowing Google-Extended in robots.txt, does nothing here: Google-Extended governs whether your content trains and grounds the Gemini models, while Google classifies AI Overviews as a Search feature. Blocking it feels like leaving the AI answers and does not touch them, the same trap as blocking an AI crawler and assuming you are out of the assistant.

The opt-out that finally arrived, for some

On 3 June 2026 Google added a Search generative AI performance report and an opt-out toggle in Search Console, under Settings then Search generative AI. A property can be set to Include (stay eligible), Exclude (opt out) or Inherit, and Exclude removes it from three surfaces, AI Overviews, AI Mode and AI Overviews in Discover, while leaving ordinary Google Search and the Discover feed untouched. It is the first control that separates the two. The catch is reach: it launched UK-first to a subset of site owners under a legally binding UK Competition and Markets Authority requirement (the Digital Markets, Competition and Consumers Act 2024, under which Google was designated as holding strategic market status in general search in October 2025), took effect on 17 June 2026, and a wider global rollout is planned but not dated. So for most sites the honest position is still that you cannot cleanly leave AI Overviews, which makes appearing well the more useful goal than opting out.

How to still be the page that gets cited

• Stay indexed and relevant: the Overview draws only from pages Google already ranks, so ordinary SEO is the entry ticket, not an afterthought.
• Answer the sub-questions: because the query fans out, a page that cleanly resolves one specific sub-topic is more liftable than a broad page that half-answers many.
• Write liftable passages: descriptive headings that name the question, short direct answers, and specific attributable figures that a model can quote and check.
• Add structured data: JSON-LD removes ambiguity about what the page is, which helps a model trust it is the right source to cite.
• Show freshness: a visible update date and the current year signal the page is not stale, which weighs on what gets pulled.

AI Mode is the same eligibility, more absorption

AI Mode is the separate, fully conversational version of this: it answers in prose, follows up, and links out even less than an Overview does. It draws on the same index and is covered by the same Search Console toggle, so there is no extra work to appear, but there is less click to win when it is the surface. Treat it as the direction of travel rather than a separate project.

Measure it, do not guess

Two trend lines tell you whether the work is landing. Where the new Search Console generative AI report has reached you, it now separates AI-surface impressions and clicks from ordinary Search, so you can see how much of your query set triggers an Overview and how the click split moves. Everywhere it has not, fall back to the manual grid: run a fixed set of buyer prompts each month and record, for each, whether an Overview appeared, whether you were cited, and whether the citation carried a link. The tool builds that prompt set, scores your page on every signal above across the answer surfaces, and returns the fixes in priority order, so you are tracking the trend rather than guessing at it.