# Pulse ยท September 25, 2026 > Google confirms a global September spam update, a video-platform AI-slop paper surfaces, VideoObject schema adds a creator property, and AI Overview links rise. Canonical: https://brandonlazovic.dev/pulse/2026-09-25/ Author: Brandon Lazovic Published: 2026-09-25 ## Google confirms a global September 2026 spam update, rolling out to all languages Status: Confirmed | Topics: organic-search-core Google's Search Status Dashboard confirms a September 2026 spam update released September 24 at 09:15 Pacific Daylight Time, applying globally and to all languages. The dashboard categorizes the incident as affecting ranking and states the rollout may take up to two weeks to complete. Google gives no further detail on which spam types or signals the update targets, or how large an effect to expect. In plain terms: A spam update is a Google algorithm change aimed specifically at pages built to game search rather than help readers, separate from Google's broader core updates. Why it matters: Any ranking movement over the next two weeks is plausibly this update rather than something else, but Google's own dashboard gives no signal detail to diagnose against. Our take: The two-week rollout window is the only concrete number here, which means the safest read for the next two weeks is wait, not diagnose. Attributing any single day's ranking swing to this update before the rollout completes is guessing. - [Google Search Status Dashboard: September 2026 spam update](https://status.search.google.com/incidents/XhUDXP7A67iHCD2kmbVu) ## Google publishes a research paper on SAFE, a multi-agent system for detecting coordinated AI-slop channel networks Status: Confirmed | Topics: llm-models-agents A Google Research paper describes SAFE (Scaled Abuse Forensics Examiner), a multi-agent system that investigates clusters of channels for coordinated AI-slop campaigns rather than scoring individual uploads. A root agent combines verdicts from three sub-agents examining channel relationships, behavioral patterns like synchronized upload timing, and content artifacts. The paper reports only that early deployment results cut investigation time versus human review, without naming a product, platform, or publish date; its PDF metadata dates the file to May 2026. In plain terms: AI slop is mass-produced, low-effort AI-generated content, and a coordinated channel cluster is a group of accounts working together, sharing infrastructure or posting patterns, that this system investigates as a group instead of judging each upload alone. Why it matters: The paper's own examples are channels and uploads, the vocabulary of a video platform, and it never mentions Search or rankings, so treating this as a new web-spam-fighting system for Search is exactly the misreading the last one drew. Our take: I argued in 'Google's new AI-slop defense is a video-platform paper, not a Search reveal' that this exact pattern would repeat: a video-platform integrity paper, no mention of Search or rankings, read by SEO coverage as a ranking story anyway. SAFE is the same shape again, right down to a headline already calling it a spam detector. - [Google Research: The Synthetic Gap - Automating Forensic Investigation of AI Slop with the Scaled Abuse Forensics Examiner (SAFE)](https://storage.googleapis.com/gweb-research2023-media/pubtools/1039292.pdf) ## Google adds a creator property and clarifies interactionStatistic types in its VideoObject documentation Status: Confirmed | Topics: structured-data-schema Google's structured data documentation for VideoObject now documents a creator property, usable alongside the existing author property, to name the person or organization that made a video. The page also clarifies that interactionStatistic supports four interaction types: WatchAction, LikeAction, CommentAction, and ShareAction, and continues to recommend interactionStatistic over the older interactionCount property for engagement metrics. In plain terms: VideoObject is the structured-data format that tells Google who made a video and how people engaged with it, using specific machine-readable fields rather than plain text Google has to guess at. Why it matters: A site marking up video with the older interactionCount property or omitting creator is now visibly behind Google's own documented schema, on a property type AI answer engines increasingly cite directly. Our take: A documentation clarification carries less urgency than a new required field, but the direction is consistent: Google keeps narrowing which interaction types it will actually read, so a custom or nonstandard interaction label is wasted markup either way. - [Google Search Central: Video (VideoObject) structured data](https://developers.google.com/search/docs/appearance/structured-data/video) ## Outbound links in Google AI Overviews rose to roughly 26% of answers by September 20, third-party tracking shows Status: Observed | Topics: ai-overviews-ai-mode, measurement-analytics Tracking data from Peec AI, cited by Search Engine Journal, shows the share of Google AI Overviews with an outbound link in the answer text rising from near zero between August 25 and September 10 to about 26.2% by September 20, with a sharp jump on September 11 whose cause is unclear. Some underlined links open an AI Mode conversation instead of an external page, and Google's own help documentation does not say whether those redirects count as an outside site's clicks or impressions. In plain terms: An outbound link in an AI Overview leaves Google's page for yours; a query refinement link just opens a follow-up AI conversation and never counts as a visit to your site. Why it matters: A rising link rate is a real signal for tracking how often AI answers still point back to the open web, but until Google labels which underlined phrases are real outbound links and which are AI Mode redirects, that rate cannot be turned into an expected-traffic number. Our take: I argued in 'Search Console counts your AI impressions and cannot connect one of them to a visit' that Google's AI reporting gives volume with no way to price it. This is the same gap from the other direction: even a rising link rate cannot tell you how many of those links were ever clickable in the first place. - [Search Engine Journal: Google AI Overviews Have More Links, But Not All Reach The Web](https://www.searchenginejournal.com/google-ai-overviews-have-more-links-but-not-all-reach-the-web/590762/) ## Bing tests replacing its Copilot Search tab label with AI Mode Status: Observed | Topics: ai-overviews-ai-mode Screenshots shared by Sachin Patel on X and covered by Search Engine Watch show a Bing test that renames the Copilot Search tab to AI Mode, the term Google uses for its own AI search experience. Microsoft has not confirmed the test, and no rollout scope, region, or timeline is documented; the change appears limited to the tab label rather than the underlying feature. In plain terms: Bing's AI-answers experience has gone by Copilot branding; this test just tries out calling that same tab AI Mode instead, matching what Google calls its own version. Why it matters: A label swap alone changes nothing about ranking or visibility, but Microsoft borrowing Google's own naming suggests Bing sees AI Mode as the term searchers now recognize, not Copilot. Our take: Naming tests like this are cheap and reversible, so the label itself is not the story; whether Bing later matches Google's AI Mode on scope and default visibility, not just its name, is the thing worth tracking. - [Search Engine Roundtable: Bing Tests Replacing Copilot With AI Mode In Search Tabs](https://www.seroundtable.com/bing-ai-mode-search-tabs-42165.html)