On 15 October 2026 Google announced five changes to how AI Overviews and AI Mode link out to the web. The headline is about placement, with links moving higher and appearing more often, but the decision for a publisher is about evidence. Google has shown that links are becoming more prominent. It has not shown that more people click them, and the only publisher-facing report it offers measures impressions rather than visits.
What exactly did Google announce on 15 October 2026?
The announcement came in a Google blog post titled "More ways to connect with the web from Search", written by Hema Budaraju, Google's vice president of product management for Search. As reported by PPC Land on 18 October and BetaNews on 16 October, it contains five changes, and they do not share the same status.
The first is that inline links, the links attached to words or phrases inside an AI answer, will be placed higher and shown more often. The second is a test of a dedicated section at the top of some answers that points to official sites, with context on why each is useful. The third is a wider rollout of news carousels, the cards shown under AI Overviews, which highlight a reader's Preferred Sources, extending to additional locations and languages that Google did not name. The fourth is that links at the end of some AI responses get image thumbnails and a more clickable design. The fifth is a test, on AI Mode video results only, in which creator and publisher names link through to Google Search profiles.
Two of the five are tests with no stated start date, market or size. The others are described as launching or rolling out without a schedule.
How does this differ from the May 2026 link changes?
Google made a similar announcement on 6 May 2026, also signed by Budaraju, covering subscription labels, a "Further Exploration" section after AI responses, creator attribution for social and discussion sources, more inline links, and desktop hover previews. TechWyse reported that Google said early testing showed users were significantly more likely to click labelled links, but no figures were published.
The pattern across both announcements is the same. Google describes interface changes and attaches qualitative claims about engagement, but publishes no click-through rates, no share of answers affected and no before-and-after traffic. The October changes extend the direction of travel rather than introducing a new mechanism.
What does the data say about clicks from AI answers?
The independent evidence points the other way from Google's framing. The PPC Land report on the announcement cites a randomised study of 1,065 US participants which found that AI Overviews cut outbound organic clicks by 39.8 per cent, and an Ahrefs measurement of a 58 per cent click-through reduction for top-ranking pages in February 2026. Both are secondary citations in that article, so check them against the original studies before relying on them in a client document.
On the publisher side, reporting on the announcement cites Chartbeat data showing that news publisher traffic from Google fell 40 per cent between July 2025 and July 2026. Google's own line, attributed to Nick Fox, Google's senior vice president of Knowledge and Information, is that it sends "billions of clicks" a week through AI features. That figure has no denominator, so it cannot be compared with either study.
The honest reading is that the two sets of numbers are not in conflict. A larger absolute number of links can coexist with a lower click rate per search, because AI answers resolve more queries without a visit.
Can publishers measure whether the new links work?
Not from Google's tools, at least not fully. Google launched Generative AI performance reports in Search Console on 3 June 2026, rolled out to a subset of site owners while it tests them. According to Search Engine Journal's coverage by Carolyn Shelby, the report shows impressions from AI Overviews and AI Mode, the URLs that appeared, and breakdowns by page, country, device and date.
It does not show clicks, queries, click-through rate, average position, the passage used in an answer, or any conversion or revenue data. An impression is counted when a link is shown within a generative feature, and in AI Overviews the link must be scrolled or expanded into view. The property-level chart counts several URLs from one site in the same response as one impression, while the page table can credit each URL separately, so the totals will not always reconcile. Google also warns that AI impressions are not equivalent to traditional organic impressions, so a blended click-through rate would mislead.
That leaves a gap. If links move higher and appear more often, impressions in this report should rise. Whether visits rise is something a publisher has to establish from its own analytics and logs.
Which of the five changes matter most for publishers?
Three deserve attention, in this order.
The Preferred Sources carousel expansion matters most, because it is the one lever a publisher can act on directly. Google says use of Preferred Sources has grown more than sevenfold since the start of 2026, but it gave no baseline and did not define "use". Earlier milestones counted unique sources selected rather than users, at about 90,000 in December 2025, over 345,000 in May 2026 and over 600,000 in August 2026. That growth is a reason to pursue reader opt-ins, covered in our piece on Google Preferred Sources, but it is not a traffic figure.
The Search profile test for creators and publishers matters second. It adds a Google-hosted page between the AI answer and the publisher's own site, so whether it adds or removes a step to the visit is an open question that Google has not addressed.
The official-site section matters third and is the largest unknown. Google has not said how it decides which sites count as official, or whether informational publishers will appear beside brands and institutions.
What should publishers do now?
Treat this as a measurement problem before it is a strategy problem. A practical sequence follows.
First, record a baseline. Capture Search Console AI impressions where the report is available, and pair them with referral sessions from Google in your own analytics, so any change after the rollout can be attributed with some care.
Second, separate what Google reports from what actually reaches your servers. AI answers are assembled from fetches that often never register as sessions, and a large share of that activity is invisible to client-side analytics, as we explain in why AI bots do not show up in Google Analytics. Server logs and edge data show the retrieval side that Search Console impressions cannot.
Third, avoid planning revenue around a promised recovery. The announcement contains no click data, no scale and no schedule for several of the changes. A forecast built on it would rest on the claim rather than the evidence.
Fourth, keep options open on the value exchange. Whether or not links improve, the commercial question is how a publisher is paid when its content is used to answer a reader, which is the territory of AI licensing, pay-per-crawl and monetisation at the edge. At blankspace we think the useful response to uncertain referral traffic is visibility into AI retrieval at the CDN layer, so a publisher can see and price that demand directly instead of waiting for Google's reporting to catch up.
What is still not known?
Several things remain unclear from the announcement as reported. Google has not said which countries and languages the carousel expansion covers, or where the creator-profile test runs. It has not said what it will measure to decide which tests to keep. It has not said whether any change increases traffic to publishers. And it has not said when Search Console will add clicks to the AI report, beyond saying it is consulting site owners on which metrics to include.
Until those answers arrive, the defensible position is that Google is making links more visible while leaving publishers unable to verify the effect.
Frequently asked questions
Did Google say the new links will increase traffic to publishers?
No. The reporting on the 15 October announcement states that Google published no click data and made no explicit traffic claim. It said inline links have become far more common and that it will place more of them higher in responses, without giving figures for either.
Does Search Console show clicks from AI Overviews and AI Mode?
Not currently. The Generative AI performance reports launched on 3 June 2026 show impressions, URLs, and breakdowns by page, country, device and date. They do not include clicks, queries, click-through rate or average position, and Google says it is still deciding which metrics to add.
What is the difference between an inline link and an end-of-response link?
An inline link is attached to a word or phrase inside the AI answer, next to the claim it supports. An end-of-response link sits in a list of sources after the answer. Google says it will place inline links higher and more often, and give end-of-response links thumbnails and a more clickable design.
Should publishers encourage readers to add them as a Preferred Source?
It is a low-cost action that Google is clearly investing in, and the news carousel expansion makes it more relevant. Google's sevenfold growth claim has no baseline, though, so treat it as a promising channel to test rather than a proven source of traffic.
How can publishers see AI retrieval that Search Console does not report?
By looking at server logs and edge data, where AI fetches appear as requests with identifiable user agents and behaviour. This shows how often AI systems retrieve your pages, which complements the impression counts in Search Console but is a different measure from reader visits.