Welcome to Pulse of the week: Google’s definition of lost clicks tested; a case study showed why the Core Web Vitals fix was targeting the wrong thing; the report measured how agents handled pricing pages; John Mueller weighed in on agent access; and Bing’s longtime leader announced his retirement.
Here is what is important to you and your work.
AI Overview Don’t Just Catch Low-Value Clicks
The researchers shared new data from a randomized field test in Google’s AI Overview. They found no measurable difference in bounce rates, return to search, or time on site between clicks that occurred with snippets and those that occurred without them.
Important facts: Research has measured a 39.8% drop in organic clicks when AI Overviews appear. Losses are concentrated in informational questions, while navigational and functional questions show no measurable change in small samples. Google’s VP of Search Liz Reid said AI Overview has cut “bounce clicks,” the lower number of visits users leave quickly, but she hasn’t released the data to back that up.
Why This Matters
If AI Overview was driving primarily low-value visits, the extra clicks websites get when the snippets are removed should look pretty bad, but it doesn’t. Additional clicks had the same bounce rates, dwell time, and return-to-search behavior as others.
That leaves Google’s click quality defense unsupported by test data. The drop in clicks on AI Overview queries can’t be chalked up as a loss of visitors who wouldn’t otherwise convert. We put the real test in April, when we started measuring the loss of clicks. Reid repeated the definition of clicks in several public settings, and Sundar Photosi answered the same traffic question in May, but Google still hasn’t released segmented click data that can solve it.
Read our full story: Overview of Google AI Research Finds Lost Clicks Were Not Low Quality
LCP Repair Often Directs the Wrong Thing
John Mueller of Google pointed to a study that explains why many fixes for Largest Contentful Paint (LCP) fail to improve the score. In store architectures that vary by vendor, the browser can lock into the wrong component, so every configuration after that targets something that was never LCP.
Important facts: The case study, published on web.dev, traces a year of Core Web Vitals work. The team tracked down LCP’s weak points in the store in the browser locking the wrong feature on the page, a side effect of how its templates are loaded, and adjusted the pages so the browser could match the original main content. The retailer reports that a high percentage of its online stores passed LCP later.
Why This Matters
The useful part here is the order of operations, in addition to some adjustments. Before compressing another hero image, confirm which element the browser recognizes as LCP, because in template-driven or carousel-heavy structures it may not be what you think it is.
Our look at Core Web Vitals across CMS platforms found a similar pattern in HTTP Archive data, where real-world LCP often breaks when the platform is slow to let the browser get the main image, not just when the page is heavy.
What People Say
The post drew a busy thread on Mueller’s LinkedInwith over 60 comments, and most of them focus on whether fast pages are driving conversion.
Skepticism came from two camps. For a reason, Manhal Abou Zaki, SEO manager at Omnicom Media Group, called this link indirectly:
No matter how many lessons are presented, a page that loads quickly can support and help conversions, but it’s less likely that someone will believe they’re converting because the page loads so fast.
Francisco Antonio Fuentes Figueroa, an SEO consultant, went after the self-promotion of the reported change, noting that it was a year full of pre/post comparisons with other changes being sent around it and “there is no visible control group,” and asking if there is any A/B or random release data available to differentiate the LCP fix.
David Swinstead, who writes as a CRO Standup Comic, said he supports the speed change link but remains adamant about its size: “But 9% increase?” The counterweight comes from Georgi Petrov, the founder of Uxify, who said that his 50/50 test runs the other way, that “LCP changes move the conversion more than people expect.”
The second camp went in order. Vijay V., head of SEO, questioned whether Core Web Vitals is a feature at all, saying he “has never seen evidence that it isn’t.” Mueller’s response narrowed the debate down to a line: “Sure I could make more money, but would I be better off?” Vishal Parmar supported the skepticism from the platform’s perspective: sites with poor LCP, CLS, and INP rank well in terms of competition.
Others have taken a broader view of the case study. Waqar Abdullah said the real lesson is to treat the LCP feature as a business priority, not just to make the site faster. Amarachi Kalu put it as “checkbox revenue”—that conversion is “what businesses remember.” Matt Bennett kept it light, noting the only place he still encountered structural change in “Google products.”
Read our full story: Google’s Mueller Flags Case for Why LCP Fixes Targeted Error
AI Agents Fallback to Third-Party Resources When Pages Won’t Load
A report from Siteline ran a simulated Claude agent against 100 top B2B software products to see if it could match its values. When an agent hit access errors or a price they couldn’t read, they often pulled the numbers from third-party sites instead.
Important facts: Siteline founder David Kaufman made the agent try to report prices and features on a set of top B2B software sites. In notable attempts, the agent hit an access error or illegible pricing, and in those cases it was more likely that he would leave the product page and pull numbers from outside sources that might be out of date. Many failures are traced back to fees loaded with JavaScript that agents do not provide, or to prices hidden behind a sales contact.
Why This Matters
The AI agent is now a guest that you can accidentally lock out. A page that looks perfect in the browser may read as empty to the agent if critical components load on the client side after the initial download. If that happens, the agent does not stop. It gets the answer from somewhere else, usually a third-party or competitor’s site that lists its prices in plain text.
The fix is to make the price and key features readable in the original HTML rather than loading them later, and display them prominently on the page.
Read our full coverage: AI Agents Struggle for B2B Price Learning, Report Findings
Mueller Says Don’t Blindly Block Agent Browsers
John Mueller was asked if Google’s quality standards will change as AI agents browse user sites. His answer was that most of the principles hold, with a new trend that you should know.
Important facts: The question, put to Mueller in Bluesky, was whether a guideline such as “images provide a satisfactory experience” still applies if the visitor is an information agent and not a person. Mueller said he expects most policies to remain the same, because a site that’s useful to people is often useful to agency browsers. He added that more information will emerge, and he named non-blocking agent browsers as the best new methods that will work.
Why This Matters
As my colleague Roger Montti points out, Mueller’s response draws a line between two different things. Content quality standards are not rewritten in agent time, because those standards are written for people and agents work for those people. What’s new is the accessibility of the technology. A site can meet every quality bar and still harm itself by blocking agents that actively work on behalf of the user.
Roger compares it to the early days of nofollow, when some sites indexed their internal links to accumulate PageRank and ended up starving important pages. Blocking agent browsers may follow a similar pattern, a technical decision made for one reason at an unintended visibility cost later. The takeaway here is to check your bot and access rules before they silently terminate the visitor category that Mueller says we don’t block blindly.
Read our full story: Google Answers the Question About SEO for AI Agents
Bing’s Fabrice Canel Announces Retirement
Fabrice Canel, the Chief Product Manager who led Bing’s crawling and development team and championed IndexNow, has announced his retirement from Microsoft after nearly 30 years. He shared the news in a farewell LinkedIn post, writing that he has accepted Microsoft’s Voluntary Retirement Program, which takes effect July 1.
His exit removes one of Bing’s most visible contacts in the SEO and webmaster community, at a time when Bing’s index is quietly powering AI search products like Copilot and ChatGPT web results. That role why the Bing index is still important, even if its share of consumer searches is still there small.
Read our full story: Fabrice Canel, Longtime Bing Search Leader, Retires from Microsoft
Theme of the Week: Plumbing Under Search
AI overview testing is about measuring, whether the traffic that Google is getting is really important. An LCP case study is also a matter of benchmarking, to see if the browser is hitting the feature you think it is. A Siteline report is about reach, whether or not an agent can read your page well enough to send a buyer to your site. Mueller’s answer is about the rules of that access, and when the blocking agents turn into a problem of visibility. Canel’s retirement is a human version of the same story, one of the people who created and defined that layer from the beginning.
Taken together, the week is less about the new feature and more about the underlying mechanics, how sites are measured, read, and accessed as more audiences come through the AI layer. The work of staying seems to go down one level, to see if the systems in front of your readers can use what you publish.
Top Stories of the Week:
Additional resources:
Featured Image: fizkes/Shutterstock
