Confirmed
Google puts Gemini 3.8 Flash into AI Mode on release day, and testers report fewer citations
Google's September 2, 2026 launch post for Gemini 3.8 Flash names AI Mode in Google Search as one of the products getting the new model immediately, alongside the Gemini app and Sheets, for AI Pro and Ultra subscribers. The next day, two independent SEO testers reported that AI Mode responses running on 3.8 Flash showed markedly fewer links and citations than the model it replaced.
In plain termsGoogle swapped in a faster, cheaper AI model for its AI Mode search feature, and testers noticed the new model links out to fewer of the websites it is summarizing than the one it replaced.
Why it mattersIf the citation drop holds up, it is a live case of a model swap quietly cutting the traffic AI Mode sends to the sites it draws on, with no policy change or announcement attached to it.
Our takeI've applied a checkable-methodology test to Google's own AI Mode numbers before, and it cuts the same way here: two testers' screenshots are a real lead, not a controlled before-and-after, so treat the citation drop as worth watching rather than confirmed until a larger sample or Google's own statement settles it.
Confirmed
Google expands Merchant Center's AI Performance Insights to four more countries
Google's Merchant Center help documentation now lists Australia, Canada, India, and New Zealand alongside the United States as eligible for AI Performance Insights, the reporting tool launched in July 2026 that shows how a brand's products surface in AI Mode and AI Overviews shopping queries. The expansion, reflected in the documentation as of September 3, 2026, stays limited to English-language queries in each market.
In plain termsThink of it as Google finally showing merchants a version of Search Console, but for the conversational shopping answers AI Mode and AI Overviews generate instead of traditional search results.
Why it mattersThis is Google's own visibility report for the gap between an approved feed and an actual AI-surfaced impression, but it covers only five countries and English-language queries so far, so most merchants still have no equivalent view into AI Mode shopping visibility.
Our takeI've written about the eligible-not-distributed gap, where a Merchant Center green checkmark confirms a feed passed policy checks but says nothing about whether a shopper ever saw the product. This insights tool is Google's own attempt to close that gap for AI-driven queries, though five countries and English-only is a narrow start against how global most catalogs are.
Confirmed
Databricks adds file reasoning and multi-step research planning to Genie Agents
Databricks announced on September 2, 2026 that Genie Agents can now analyze unstructured files, PDFs, documents, and images stored in Unity Catalog volumes, alongside structured data in the same conversation. A new agent mode plans and refines multi-step research queries on its own, and new developer APIs let teams embed Genie agents into custom applications and chatbots rather than using the built-in interface only.
In plain termsUntil now a Genie agent could query a company's database tables but not read a PDF sitting next to them, and now it can do both in the same conversation.
Why it mattersFor anyone building retrieval or agent tooling on top of a data warehouse, this closes a real gap by letting one agent reason across PDFs and spreadsheets together instead of forcing a separate unstructured-data pipeline.
Our takeDatabricks did not say whether these capabilities are generally available or still in preview, and that distinction matters more than the feature list for anyone deciding whether to build on this now or wait for a stable release.
Observed
Google's Mueller reportedly says ranking recovery from algorithmic fixes can take months
Search Engine Journal reports that Google's John Mueller told a Reddit user that recovering rankings after an algorithmic issue is fixed is not a quick process. Mueller reportedly said broader algorithmic changes can take many months to show effect, and separately that link disavow changes take roughly three to six months to register, though no direct link to the original Reddit thread accompanies the report.
In plain termsFixing whatever caused a ranking drop does not mean the rankings bounce back right away, since Google's own systems can take months to notice and respond to the fix.
Why it mattersIf accurate, this sets a practical expectation for any post-fix reporting to a client or stakeholder, since a change that looks correct in week two can still be too early to call while the ranking signal itself lags by months.
Our takeThis is a secondhand Reddit quote with no link to the original thread, so treat the multi-month framing as a reasonable planning assumption rather than a documented Google policy until it surfaces in Google's own Search Central channels.
Observed
Site owners report GA4 standard reports showing zero traffic for September 1
Multiple site owners and several SEO trade outlets report that GA4 standard reports display zero or near-zero sessions and users for September 1, 2026, while GA4's real-time reports keep showing normal live activity. Complaints appeared on the official Google Analytics Community forum and across social platforms, and as of this writing Google has not posted a status update or root-cause explanation for the standard-reports gap.
In plain termsThe dashboard shows visits basically stopping for one day, but real-time visitor counts never actually dropped, which points to a reporting display bug rather than a real traffic loss.
Why it mattersAnyone reporting September traffic to a client or stakeholder this week should flag the September 1 gap as a known measurement artifact rather than a real conversion or organic-traffic drop.
Our takeUntil Google acknowledges a cause, the safest read is a reporting or processing delay rather than a genuine traffic collapse, since GA4's own real-time view never showed the drop that standard reports are displaying.