Google AI Overviews may be opening more visible routes back to the web. New data from Peec AI, reported on September 23 by Search Engine Roundtable, suggests that the share of AI Overviews containing external links has risen from close to zero to more than 25%.
The datapoint was reported by Barry Schwartz in Search Engine Roundtable after Peec AI’s Malte Landwehr posted a chart showing the increase. If the measurement is representative, it would mean that more than one in four AI Overview responses in the tracked dataset now contains at least one outbound link embedded within the answer experience.
There is an important limitation: Peec AI has not published the methodology behind this specific chart. The sample size, query mix, countries, devices, measurement interval and precise definition of an “external link” were not disclosed alongside the reported result. Google has not confirmed the 25% figure either. The finding should therefore be treated as an interesting third-party signal, not a verified platform-wide measurement.
The reported jump is large enough to matter—but too opaque to generalize
Search Engine Roundtable describes the Peec AI chart as moving from almost 0% of AI Overviews containing external links to above 25%. That is a substantial change in prevalence if the underlying tracking conditions remained comparable.
But percentages in AI-search measurement are extremely sensitive to sampling. Commercial queries, informational searches, local questions and news-oriented prompts can trigger different AI Overview layouts. Desktop and mobile interfaces can also expose links differently, while product experiments may affect only subsets of users.
Without those details, the safest interpretation is narrow: Peec AI observed a sharp increase in external-link presence within whatever AI Overview population its chart measured. It is not yet possible to conclude that exactly one quarter of all AI Overviews globally contain these links.
That distinction matters because the SEO implications depend on whether the change is a broad product shift or a concentrated experiment affecting particular query classes.
Google has independently said it wants AI Search to expose more web links
Although Google has not validated Peec AI’s number, the direction of the observation is consistent with Google’s own recent product announcements. In May, Google published an official update titled “5 new ways to explore the web with generative AI in Search”, saying it was improving how links appear in AI Mode and AI Overviews and adding more direct routes to relevant websites.
Google framed generative search as a starting point for further exploration rather than a closed answer surface. The company described direct links within AI responses, website previews and other mechanisms intended to make original sources easier to discover.
In June, Google went further in an official update for publishers and site owners. It said it had increased the number of inline links directly within generative AI responses and was continuing to experiment with link designs intended to encourage visits to websites.
Those statements do not prove Peec AI’s reported 25% figure, but they make the broader direction plausible: Google has publicly acknowledged that link presentation inside AI Search is an active product-design priority.
An external link inside the answer is different from a source card
The placement of a link matters almost as much as whether a source is cited. AI Overviews have long included source references and supporting web results around the generated answer, but an inline external link changes the interaction model.
A user reading a sentence can encounter the destination at the point where the underlying topic, entity or claim becomes relevant. That is closer to the conventional web behavior of clicking contextual links than opening a separate citation panel or scanning a source carousel.
For publishers, this could create another traffic surface inside an experience frequently criticized for answering questions before users reach the organic results below.
But the existence of that surface does not tell us how often people actually use it.
More links do not automatically mean more publisher traffic
The strongest interpretation to avoid is that a rise in external-link prevalence proves AI Overviews are now sending substantially more traffic to publishers. Peec AI’s reported chart measures the presence of links, not clicks.
An answer could contain several outbound links and still satisfy most users without a click. Conversely, one well-placed contextual link could attract meaningful traffic if the user needs deeper information, a product page, an original source or a detailed guide.
CTR depends on visual prominence, wording, user intent, answer completeness, device layout and the destination itself. None of those variables can be inferred from the reported percentage alone.
The new datapoint is therefore best understood as an increase in traffic opportunity, not evidence that traffic has already increased by a corresponding amount.
Google’s AI Search strategy has been under pressure over the open-web question
Publishers have spent much of the generative-search transition worrying that AI answers could consume web content while reducing the need to visit the sites that produced it. That tension becomes especially important for Google because its traditional search ecosystem historically connected discovery with publisher traffic.
Google has repeatedly argued that its generative AI features are designed to help people discover and visit websites. Its June publisher update said AI Overviews and AI Mode include prominent links and that the company is continuing to improve those experiences as jumping-off points to the web.
The design problem is structural. The more complete the generated answer becomes, the less reason some users have to click. Google can counter that effect partly by making links more contextual, more visible and more useful at moments when a user wants to go deeper.
If Peec AI’s observation reflects a broader rollout, the increase in external links could be one expression of that strategy.
The timing makes the signal more interesting
The reported change arrives after several months in which Google has publicly emphasized web exploration inside generative search. The May product update introduced new ways to expose relevant websites, while the June publisher announcement explicitly said inline links had increased.
That makes the September chart potentially more meaningful than an isolated UI experiment with no prior context. A sustained rise would suggest that Google’s stated effort to make generative answers more connected to external websites is becoming measurable in third-party tracking.
However, only a transparent time series could establish that confidently. We would need to know when the increase began, whether it was gradual or abrupt and whether the same prompt set was tested consistently before and after the change.
Without that information, correlation with Google’s announced product direction remains suggestive rather than conclusive.
Peec AI has published large AI Overview datasets before, but that methodology cannot simply be assumed here
Peec AI is not new to measuring Google’s generative search surfaces. In May 2026, the company published an analysis based on 500,000 prompts and reported that AI Overviews appeared in 86.7% of its April sample.
Importantly, Peec disclosed that the dataset skewed toward business-oriented prompts with clear buying intent and excluded navigational queries. That methodological disclosure made the scope of the result easier to interpret.
The September external-link chart, as reported by Search Engine Roundtable, does not include comparable methodological detail. It would be inappropriate to assume that the same 500,000-prompt dataset, query mix or measurement process was used.
This is exactly why third-party AI visibility statistics need sample-level context. A percentage without the denominator and query distribution can be directionally interesting while remaining difficult to generalize.
Even the reporter flags uncertainty around the result
Search Engine Roundtable’s Barry Schwartz did not present the number as settled fact. He wrote that the result did not intuitively feel right to him, but considered it worth sharing because it points in the opposite direction from much of the recent discussion about Google reducing outbound traffic.
Schwartz also disclosed that he has a small investment in Peec AI and said that relationship has led him to cover the company less often.
Neither point invalidates the underlying chart, but both reinforce the need to treat the story as a reported datapoint awaiting stronger methodological support rather than a confirmed Google product metric.
The most useful next step would be publication of the underlying sample and a clear definition of what Peec counts as an external link inside an AI Overview.
For SEO, link eligibility may become a separate visibility metric
If inline links continue expanding, AI Overview optimization will need to distinguish several layers of visibility. A page can help ground an answer, appear in a source module, receive an inline link or actually generate a click. Those are related outcomes, but they are not the same outcome.
This creates a more detailed funnel for GEO measurement. Retrieval asks whether Google selected the page as useful evidence. Citation or source visibility asks whether the page was exposed to the user. Inline-link visibility asks whether Google placed a direct route to that page within the answer itself. Referral data then measures whether the user took that route.
Collapsing those stages into one “AI visibility” percentage can hide meaningful differences. A site could be heavily used as source material while receiving few visible links, or it could receive prominent links that generate unusually valuable visitors.
The Peec AI datapoint is interesting precisely because it concerns one of these intermediate layers rather than generic AI Overview prevalence.
Google is also giving publishers more visibility into generative-search appearances
Google’s June update introduced additional Search Console insights about how pages appear in generative AI Search features, including impressions and information about which pages surface in AI responses and in which countries.
The company also introduced controls allowing website owners to manage whether their content and links appear in generative AI Search. Google said sites opting out would not receive impressions or traffic from those generative experiences.
By August 31, Google said those controls and insights had rolled out globally. That gives publishers a more direct way to observe whether their pages are appearing inside AI Search, although the available reporting still does not resolve every question about citation placement and click behavior.
As external-link designs evolve, Search Console data could become essential for separating a visually noticeable product change from a change that materially affects publisher performance.
The commercial value of an AI Overview click may differ from ordinary search traffic
Not every click has equal value. A user who leaves an AI Overview after reading a synthesized answer may be further along in the information process than someone clicking a conventional blue link near the beginning of research.
The AI answer may have already narrowed options, explained terminology or eliminated irrelevant choices. The resulting click could therefore represent a more specific intent: verifying a source, examining a product, reading original reporting or completing an action the generated summary cannot perform.
That possibility makes referral quality as important as referral volume. Publishers and brands should monitor engagement, conversions and landing-page behavior from AI-influenced search rather than judging the change only by raw session counts.
Peec itself has recently emphasized this distinction in its analytics products, but the new external-link chart does not provide conversion data and should not be interpreted as doing so.
The real question is whether the link increase persists
Google constantly experiments with Search interfaces, and generative AI surfaces are particularly fluid. A sharp change observed over a short period can represent a test, a partial rollout or a temporary configuration rather than a stable product policy.
The next several weeks of tracking will therefore matter more than the first spike. If the share of AI Overviews containing external links remains above 25%, expands across query classes and appears across countries and devices, the case for a durable change becomes stronger.
If the number falls back quickly, the chart may instead have captured an experiment or a narrow slice of traffic.
Longitudinal measurement with a stable query set is the clearest way to distinguish those possibilities.
Publishers should treat this as an opportunity signal, not a traffic recovery announcement
The Peec AI observation is encouraging for anyone concerned that generative search is becoming a one-way extraction layer. A rise from almost no AI Overviews with external links to more than one in four would materially expand the number of generated answers that offer a direct route to the open web.
But the evidence does not yet justify declaring that Google has reversed the traffic impact of AI Overviews. The measurement is third-party, the methodology has not been published and no CTR or referral increase accompanies the chart.
What can be said with greater confidence is that Google itself has spent 2026 adding and emphasizing direct web links in its generative search experiences. The Peec AI data may be an early quantitative signal that those design changes are becoming more prevalent.
For SEO and GEO teams, that makes external-link presence worth tracking as its own metric. The traffic door may be opening wider, but we still need to measure how many users actually walk through it.