Search digestAugust 2026

August 2026 gave site owners a difficult combination: a global spam rollout, unusually large ranking losses in an independent sample, and reporting errors close enough to the rollout to confuse diagnosis. At the same time, Google and Microsoft introduced more ways to measure AI visibility. The useful question is how these changes alter the next decision you make about your site.

zoom_in

Product change

The spam update was short; the ranking losses were not small

Google's official incident record puts the rollout between 18 August at 09:27 and 21 August at 01:49 Pacific time. It applied worldwide and to all languages. Duration tells us when to investigate; it does not tell us how severely a particular site was affected.

SE Ranking's measurements, published by Search Engine Land on 27 August, supply a more useful indication of scale. 16.71% of tracked top-10 URLs fell beyond position 100, against 9.2% in the comparison window. That is 7.51 percentage points, or approximately 82% more relative to baseline.

zoom_in

The study tracks positions 1–100. It cannot distinguish deindexing from a fall below its tracking range, establish a cause for each domain, or describe the impact across every country. The headline is substantial ranking disruption in a defined U.S. sample, not a finding that 82% of websites were penalised.

Glenn Gabe's four anonymised case studies, published on 31 August, offer a different kind of evidence. His examples include large programmatic sections with little original value, thin affiliate setups and misleading functionality. These are selected investigations by a consultant, not a representative sample or Google's official list of targets. They suggest questions to ask about affected pages; they do not establish that using AI or programmatic publishing is itself the cause.

Data

A reporting dip is not automatically a ranking loss

The Search Console anomaly record documents an undercount of Discover clicks and impressions on 13 August, including impressions for Generative AI in Discover where available. A separate error reduced reported AI-in-Search impressions from 13–17 August. Google describes both as logging issues; its 21 August note says the missing AI-in-Search data was restored.

This matters when the first version of a monthly report was exported before the correction. An old spreadsheet can preserve an apparent drop that the current report no longer shows. Re-export the affected dates and retain the date of extraction alongside the data. Do not extend this explanation to metrics or dates outside the documented incident.

Our reading of August is that diagnosis needs three comparisons: the corrected search report, actual website sessions and outcomes, and a consistent ranking sample where available. They measure different things. Agreement between them strengthens a hypothesis; disagreement tells you where to investigate next. None should be substituted for another without explaining the difference.

Product change

AI visibility becomes a measurable operating choice

On 31 August, Google said the AI controls and insights first announced in June had reached websites worldwide. The updated announcement describes a Search Console choice governing appearance in, and grounding of, generative AI Search features such as AI Overviews, AI Mode and AI Overviews in Discover. Opting out removes traffic and impressions from those features. Google says this control is not a ranking signal for results outside them.

The dedicated performance reports expose impressions, pages, countries and time-based views, plus device information for Search. These impressions remain included in the overall performance report. The announcement does not promise a full AI-specific clicks, CTR or conversion report.

For a site owner, the immediate benefit is a better baseline: which pages appear, in which markets, and how that visibility changes. The decision about participation should then reflect the business model. A publisher, a software vendor and an online retailer may value exposure and referral visits differently. Record the setting and its change date so later analysis has the necessary context.

Product change

Clarity adds the distinction between being found and getting visits

Two August releases make Microsoft's Clarity Citations dashboard more useful for analysis. On 3 August, it added labels and filters for branded and non-branded grounding queries, along with a corresponding Share of Authority breakdown. These are queries used by AI systems to find supporting material; they are not necessarily the full question a person typed.

That distinction matters because a brand can be well represented when people already name it, yet scarcely appear during broader category research. Reporting the two together can hide the difference. A practical starting point is to compare the non-brand view with the topics and pages your team actually wants new customers to discover.

On 13 August, Scrape-to-Referral insights added an AI-operator view, a ratio comparing scrape activity with referrals, coverage context and links into filtered session recordings. The coverage checks matter: a ratio assembled from mismatched domains or incomplete CDN and Clarity data is not a fair comparison.

Product change

Country-level analysis is now harder to avoid

Google's 28 August site reputation announcement took effect on 30 August. Outside the European Economic Area, a relevant manual action continues to affect the site section concerned. For EEA searchers, that manual-action impact does not apply; the section may instead be separated in Google's systems and rank independently over time. Search Console notifications and reconsideration remain available.

The distinction follows the searcher's location. For publishers hosting third-party sections, a worldwide traffic total can therefore hide the behavior that matters. Compare the same section across EEA and non-EEA markets, and inspect any actual notices before explaining a change. This is a change in enforcement, not a withdrawal of the policy or a promised recovery.

Country differences also appear in independent AI visibility research. SE Ranking's France study, published on 18 August, examined 100,000 keywords across 20 niches on 7–8 August. It found AI Overviews for 52.63% of sampled keywords. The launch it discusses occurred in July; the August news is the measurement.

zoom_in

The variation across niches is the actionable result. A single market-wide percentage is a poor forecast for a site's own exposure. Inspect a stable set of queries in the countries and categories that matter to the business. Treat the study as a reason to measure locally, not as a promise that half of every site's queries will show an AI Overview.

Product change

SPA teams can inspect the experience after the first page load

The August update to Google's SPA and Core Web Vitals guidance describes Chrome 151 APIs for measuring route transitions without full page loads. Chrome 152 DevTools then added soft-navigation support to live metrics. This helps teams investigate the journey inside an application, where a fast landing page does not necessarily imply fast subsequent transitions.

Chrome had not announced a timetable for incorporating those route measurements into CrUX. Other browser engines also did not yet support these APIs. The distinction is essential: new local and real-user measurement capabilities do not establish a change in Search Console's Core Web Vitals report or Google's ranking systems.

For a catalogue or application built around client-side navigation, test a representative sequence: initial entry, category change, product or detail view, and the next important interaction. Keep full-load and route-transition observations distinguishable. The practical gain is finding slow or unstable steps that a landing-page-only check would miss.

Product change

Two smaller changes deserve concrete implementation checks

Google's 20 August Preferred Sources button lets readers select a publication and return to the page. The publisher documentation describes standard and custom JavaScript implementations, plus a deeplink fallback. Eligibility is at domain or subdomain level, not for a subdirectory such as /blog.

For publishers with a returning audience, that is a testable distribution option. Check eligibility, put the invitation somewhere sensible in the reading experience, and measure usage. It should earn its place alongside existing subscription or newsletter prompts; merely installing it is not evidence of improved ranking.

Meanwhile, Microsoft's Webmaster API notice set 31 August as the SOAP and POX retirement date and directed users to REST. Teams using automated Bing reporting or submission should verify successful jobs, not just a working dashboard. Record the protocol used by each dependency and investigate any gaps around the deadline.

These items deserve less space than the spam rollout, but their actions are clearer. One is an audience experiment; the other is an integration check with a concrete deadline. Neither requires a wholesale SEO strategy change.

Action

What August changes about the work

Our conclusion is that the strongest response to August is more precise diagnosis. Broad rankings, AI visibility, referral traffic, regional enforcement and on-site experience now need to be read together, while retaining the distinctions between them. A single blended score would make that work harder.

For September planning, put work with a clear trigger ahead of broad speculative projects: correct stale reports, repair a confirmed integration failure, inspect the specific pages that lost visibility, and baseline the new measurement surfaces. Reserve larger content or architecture changes for findings that actually justify them.