SEO Recap covering October 8, 2026: September 2026 spam update wraps and manual actions climb

Built by Stephanie Chung·5 min read·5 stories

Daily summary of what matters in SEO, GEO, AEO, and AI search generated with Claude Code (beware of hallucinations)

AI in Search / AI Overviews

A January 2026 DeepMind paper proves an autoregressive model with a fixed hidden dimension can rank an arbitrary number of documents, while a dual encoder's embedding dimension must grow linearly with the corpus, extending Google's 2021 "Rethinking Search" and 2022 Differentiable Search Index work.
The method encodes documents as short generated docIDs rather than URLs and uses a training loss called SToICaL that teaches the model to order the whole ranked list. On WordNet it pushed recall at positions 2-5 from the fifties into the mid-nineties.
Experiments stayed small, fine-tuning Mistral-7B with candidate IDs placed in the prompt (reranking, not open-corpus retrieval), on ~82,000 WordNet concepts and 310 shopping examples, so it is a foundation rather than a deployment.
Beam search caps results at the beam width, so visibility turns binary (in the generated set or absent) and there is no way to tell "ranked badly" from "never produced," shifting measurement from position to inclusion.
Removing the separate nearest-neighbor index is a structural cost reduction for generative results, weakening the economic reason to keep 10 blue links. Alphabet reported Google Search and other revenue of $63.27 billion, up 17%.

The program gives approved platforms a dedicated ingestion pipeline to push user-generated content and interaction signals to Google Search, independent of the Indexing API and traditional crawling, though it does not guarantee appearance in Search.
Eligibility requires platforms to primarily host UGC on dedicated pages with stable URLs, serve a high volume with a large audience, and keep content public and ungated with every post attributable to a creator with a public profile.
Technical requirements include OAuth 2.0 authentication and strictly validated JSON-LD payloads using schema.org markup such as SocialMediaPosting or DiscussionForumPosting with interactionStatistic sub-fields.
Content must be submitted within minutes and engagement counters updated within 72 hours. Applicants hear back within 6-8 weeks and Google will not accept all applications.

Google's September 2026 Spam Update Finished Rolling Out
Sources: Search Engine Journal · Search Engine Roundtable

Google's September 2026 spam update completed on October 8 at 1:00 am PT, roughly 13 days and 16 hours after launching September 24, far longer than this year's earlier spam updates that ran 19 hours (March), 2 days (June), and under 3 days (August).
Google called it a normal global spam update applying to all languages, but did not disclose what it targeted or why it took nearly the full 2-week window.
Diib's early read on 32,569 sites found the median share of unpaid Google search visits fell from 34% to 31%, with 10% of sites losing half or more of their share versus 5% a week earlier.
SEOmonitor warned that rank trackers showed sudden changes from September 26 that real searchers did not see, attributing it to incomplete results during automated scraping, so check Search Console impressions and clicks before concluding you were hit.
Barry Schwartz flagged the update felt more core-update-like in impact, with volatility clustering September 25-27, September 30, and October 4-6.

Spam updates evaluate technical configurations, automation footprints, and link profiles across multiple independent systems that refresh at different times, so mid-flight tracker swings rarely predict where a site lands.
Weekend search behavior shifts (more mobile, fewer business queries, larger local packs) can make charts look catastrophic independent of any update.
Freeze code, redirects, titles, and bulk content changes until the rollout is marked complete, then give the index an extra 10-14 days before measuring permanent loss.
Take dated Google Search Console snapshots noting the update start date, and only begin a structured review if traffic fails to recover after the settling period.

Marie Haynes said manual action help requests jumped from essentially none to several per week, mostly thin content, pure spam, and scaled content hitting long-standing legitimate businesses.
Daniel Foley Carter reported more manual actions appearing over recent months, tying them to scaled content abuse and cheap or spammy SEO shortcuts, with some sites losing 90%+ of traffic overnight.
Manual actions surface in Search Console after a quality rater review or when automated systems or user spam reports flag patterns, covering violations from thin content and unnatural links to cloaking and site reputation abuse.
Both ChatGPT and Gemini, asked to analyze the Webmaster Help forums, agreed complaints are up but not a dramatic spike.

Published on October 9, 2026